JP7702735B2 - Method, device, and program for generating a prediction model for predicting agricultural crop production performance - Google Patents

Description

Article 30, paragraph 2 of the Patent Act applies. (1) Website address: http://www.naro.affrc.go.jp/publicity_report/press/laboratory/rcait/134410.html Date of publication: March 6, 2020, etc.

The present invention relates to a method, device, and program for generating a prediction model for predicting agricultural crop production performance.

At various agricultural production sites, attempts are being made to use meteorological data to predict the component content of crops during the growing and harvesting seasons in order to utilize the data for cultivation management, shipping plans, sales negotiations, etc.

Patent Document 1 describes a technology that predicts solar radiation and temperature information to bring the sugar content of a crop to a target sugar content, and controls the environment inside a greenhouse based on the predicted solar radiation and temperature information. In the technology described in Patent Document 1, the difference between a crop's specific standard sugar content and the target sugar content is used to predict sugar content by relating it to the difference between the average solar radiation and temperature values for the four weeks before harvest and the standard values.

Non-Patent Document 1 describes the construction of a model trained using time series data on weather fluctuations over the past few years and time series data on mandarin orange quality. The technology described in Non-Patent Document 1 uses the constructed model to predict the quality of mandarin oranges at the current year's harvest time based on time series data on weather and mandarin orange quality up until the current year's harvest time.

JP 2020-48551 A

Morimoto et al., Plant Environmental Engineering 17(2): 90-98, 2005

There are three problems with predicting the elemental content of agricultural crops. The first is the small number of data measurement points. It is rare to frequently measure the elemental content of agricultural crops during the growth process and accumulate the data; in most cases, elemental content is measured only during harvest or shipping. The second is that elemental content is influenced by factors other than weather. For example, in perennial crops, fluctuations over time can affect the elemental content, and in monoennial crops, trends over time in cultivation management that reflect consumer preferences can affect the elemental content. The third is that the type and timing of meteorological data variables that affect elemental content are often unknown.

Therefore, if a predictive model could be developed that reflects changes in the crop production environment over time using past measurements of component content and meteorological data, it would be of great benefit to farmers in the field.

The technology described in Patent Document 1 is difficult to apply to sugar content prediction when the crop-specific standard sugar content is unknown. In addition, when the harvest date is undecided, the technology described in Patent Document 1 is difficult to apply to sugar content prediction because the starting point of the solar radiation and temperature data used for the prediction is not determined.

In the technology described in Non-Patent Document 1, in order to create time-series data on mandarin orange quality, it is necessary to measure the component content multiple times during the growing process. In addition, although the model described in Non-Patent Document 1 can be used to predict the quality of the next period in the growing calendar, it is difficult to predict the quality at an earlier stage.

One aspect of the present invention has been made to solve the above-mentioned problems, and its purpose is to realize a technology for predicting the production performance of agricultural crops.

A method for generating a prediction model according to one embodiment of the present invention is a method for generating a prediction model for predicting the production performance of agricultural crops, and includes a step of generating a prediction model that represents changes in the production performance over time by referring to actual measured value data of the production performance for each past year, and predicts the production performance for the current year from at least a portion of the actual measured values of the production performance up to the previous year.

A prediction model generating device according to one embodiment of the present invention is a prediction model generating device for predicting the production performance of agricultural crops, and includes a generating unit that generates a prediction model that represents changes in the production performance over time by referring to actual measured value data of the production performance for each past year, and that predicts the production performance for the current year from at least a portion of the actual measured values of the production performance up to the previous year.

The predictive model generating device according to each aspect of the present invention may be realized by a computer. In this case, the control program for the generating device that causes the computer to operate as each part (software element) of the generating device to realize the generating device, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention.

According to one aspect of the present invention, it is possible to realize a technology for predicting the production performance of agricultural crops.

1 is a block diagram showing an example of a configuration of a main part of a prediction system according to an embodiment of the present invention. FIG. 1 is a diagram showing an example of actual measurement data of production results for each year up to the previous year used in a prediction device according to an embodiment of the present invention. FIG. 1 is a diagram illustrating changes in crop production performance over time and the effects of weather on production performance. 1 is a diagram illustrating the influence of meteorological data variables on production performance, which is used in a meteorological data prediction model generation device according to one embodiment of the present invention. FIG. FIG. 1 is a diagram illustrating the concept of predicting agricultural crop production performance using a prediction model, which is executed by a prediction system according to one embodiment of the present invention. 1 is a flowchart illustrating an example of a prediction process executed by a prediction device according to an aspect of the present invention. 1 is a flowchart illustrating an example of a predictive model generation process executed by a predictive model generation device according to an embodiment of the present invention. 1 is a flowchart illustrating an example of a prediction model generation process executed by a weather data prediction model generation device according to an aspect of the present invention. FIG. 13 is a diagram showing the verification results of sugar content prediction accuracy in an embodiment. FIG. 13 is a diagram showing the verification results of sugar content prediction accuracy in an embodiment. FIG. 13 is a diagram showing the verification results of the acidity prediction accuracy in the examples. FIG. 13 is a diagram showing the verification results of the acidity prediction accuracy in the examples. FIG. 13 is a diagram showing the verification results of sugar content prediction accuracy in an embodiment. FIG. 13 is a diagram showing the verification results of the acidity prediction accuracy in the examples. FIG. 1 is a diagram showing the effects of precipitation, average temperature, and sunshine hours on sugar content in an example. FIG. 1 illustrates the effect of soil moisture on production performance. FIG. 11 is a block diagram showing an example of a configuration of a main part of a prediction system according to another aspect of the present invention.

One aspect of the present invention is to realize a technology for predicting the production performance of agricultural crops. The production performance of agricultural crops predicted by one aspect of the present invention includes, but is not limited to, the yield of the agricultural crop; sugar content, acidity, protein content, molecular components measured by mass spectrometry, pigment content, and other component contents; morphology such as color, shape, size, and length; degree of lodging, growth index of leaves, flowers, stems, and fruits (stem length, degree of leaf opening, coloring of flowers and fruits), number of flowers, number of fruits, and number of leaves. In other words, according to one aspect of the present invention, at least one of the evaluation items for evaluating the produced agricultural crops can be predicted. Agricultural crops whose production performance can be predicted by one embodiment of the present invention are not particularly limited, and examples include perennial plants such as mandarin oranges, apples, peaches, grapes, loquats, plums, coffee, green tea, beets, angelica, licorice, peonies, and rhubarb; and annual plants such as rice, wheat, soybeans, peanuts, millet, tomatoes, spinach, eggplant, pumpkins, strawberries, perilla, cnidium, bupleurum, potatoes, sweet potatoes, and ornamental flowers.

Prediction System 100
A prediction system 100 used to predict the production performance of agricultural crops will be described with reference to Fig. 1. The prediction system 100 includes a prediction device 10, a prediction model generation device 20, and a weather data prediction model generation device 30. The prediction system 100 further includes an input device 40, a storage device 50, and an output device 60. The prediction system 100 may include the prediction device 10, the prediction model generation device 20, and the prediction model generation device 30 as independent devices, or may include them as an integrated device within a single device.

The input device 40 accepts input operations by a user to the prediction system 100. As an example, the input device 40 accepts input of data used to predict the production performance of agricultural crops in the prediction device 10. The input device 40 also accepts input of data used to generate a prediction model in the prediction model generation device 20. Furthermore, the input device 40 accepts input of data used to generate a prediction model of weather data in the prediction model generation device 30. The input device 40 may be, for example, a keyboard, a mouse, a touch sensor, etc.

The storage device 50 stores programs and data used in the prediction system 100. As an example, the storage device 50 stores various data input via the input device 40. As an example, the storage device 50 also stores a prediction model, input information, and output information used to predict the production performance of agricultural crops in the prediction device 10. As an example, the storage device 50 also stores learning data used to generate a prediction model and the generated prediction model in the prediction model generation device 20. As an example, the storage device 50 also stores learning data used to generate a prediction model and the generated prediction model in the prediction model generation device 30. The storage device 50 may have a database on the cloud or a server that stores various data.

The output device 60 outputs the results predicted by the prediction device 10. The output device 60 may also output cultivation management guidance information based on the results predicted by the prediction device 10. The cultivation management guidance information includes, for example, the timing of mulching, information on drainage channel maintenance, the number and timing of fruit thinning, information on fertilization, information on root pruning, etc.

The form of output by the output device 60 is not particularly limited. The output device 60 may be, for example, a display device that displays the information as an image, a printing device that prints the information, or an alarm device that outputs the information as sound. The output device 60 may also be a display of a mobile device such as a smartphone that displays the results predicted by the prediction device 10 and cultivation management guidance information based on the results.

(Prediction device 10)
The prediction device 10 is a prediction device that predicts the production performance of agricultural crops. The prediction device 10 predicts the production performance of the current year from at least a part of the actual measured values of the production performance up to the previous year, using a prediction model that represents a change in the production performance over time and that is generated by referring to the actual measured value data of the production performance for each past year.

The prediction device 10 includes a control unit 11. The control unit 11 controls each part of the prediction device 10, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions contained in the program. This constitutes each part of the control unit 11. As each part, the control unit 11 includes an input data acquisition unit 12 and a prediction unit 13.

The input data acquisition unit 12 acquires input data, which is input information for a prediction model that predicts production performance. Based on an input signal representing an instruction to start prediction from the input device 40, the input data acquisition unit 12 reads out actual measurement value data including at least a portion of the actual measurement values of production performance up to the previous year from the storage device 50. The input data acquisition unit 12 may also acquire actual measurement value data input via the input device 40. The input data acquisition unit 12 outputs the acquired input data to the prediction unit 13.

Here, the actual measured values of production performance up to the previous year are intended to mean the actual measured values of production performance measured prior to the current year, which is the year for which production performance is to be predicted. The actual measured values of production performance need only be measured at least once each year during the growing season, harvest season, or at the time of shipping. The actual measured value data need only include the actual measured values of production performance for at least the previous year, but may also include the actual measured values of production performance for multiple years up to the previous year.

The prediction unit 13 predicts the current year's production performance from at least a portion of the actual measured values of the production performance up to the previous year using a prediction model 24 that represents changes in production performance over time and was generated by referring to actual measured value data of production performance for each past year. The prediction unit 13 reads out the prediction model 24 from the storage device 50, inputs the actual measured value data sent from the input data acquisition unit 12 to the prediction model 24, and acquires the production performance for the current year output from the prediction model 24. The prediction unit 13 outputs the acquired data representing the production performance for the current year to the output device 60 as the prediction result.

The prediction model 24 used by the prediction unit 13 to predict the production performance may be a prediction model 24 generated by a prediction model generation device 20 described below. As an example, the prediction model 24 is generated by performing machine learning using actual measurement value data as learning data, and is a prediction model in which at least a portion of the actual measurement values of the production performance up to the previous year are input information and the production performance of the current year is output information. In other words, by using the prediction model 24, it is possible to predict the production performance by taking into account the annual fluctuations in the production performance of agricultural crops (fluctuations in production performance) as changes over time.

As an example, it is known that perennial plants such as mandarin oranges tend to have a decline in production the year following a good production year due to a decline in tree vigor, such as a decrease in stored nutrients in the tree. Furthermore, perennial plants alternate between good years (years with good production results) and bad years (years with poor production results). As such, the production results of agricultural crops fluctuate on an annual basis, so by using a prediction model 24 that predicts the production results of the current year from the actual measured values of production results up to the previous year, it is possible to predict production results that take into account annual fluctuations.

As another example, in the case of monoannual plants, the cultivation management method may vary from year to year, and this variation in the cultivation management method leads to fluctuations in production performance from year to year. Therefore, even in the case of monoannual plants, by using a prediction model 24 that predicts the production performance of the current year from the actual measured values of the production performance up to the previous year, it is possible to predict the production performance taking into account the year-to-year fluctuations caused by the variation in the cultivation management method.

Here, the actual measurement data used to predict production performance will be described with reference to FIG. 2. FIG. 2 is a diagram showing an example of actual measurement data of production performance for each year up to the previous year used by the prediction device 10. In the graph on the left side of FIG. 2, the X axis shows the calendar, the Y axis shows sugar content, and the Z axis shows the year, and the graph shows the annual sugar content change for each year. By extracting the data on the annual sugar content change for each year on a specific day from such data on the annual sugar content change for each year, a graph showing the annual sugar content change for each year on a specific day can be obtained, as shown in the graph on the right side of FIG. 2. Note that the graph on the left side of FIG. 2 was obtained from the results of measuring sugar content multiple times in each year. FIG. 2 is shown to explain the concept of the present invention, and the present invention is not limited to the embodiment shown in FIG. 2.

As an example, Figure 2 shows a graph (left graph) showing the annual sugar content change for each year for the past 15 years, and a graph (right graph) showing the sugar content change for the past 15 years as of October 30th. In this way, by using data showing the annual sugar content change on a specific day, the prediction device 10 can predict production results taking into account annual fluctuations.

When measuring the annual change in production performance, measurements are repeated several times a year on a reference plant individual (such as a single designated reference tree). In other words, for sampling, leaves and fruits are thinned out from the reference plant individual several times a year. The loss of leaves and fruits due to sampling may change the translocation of photosynthetic products within the plant individual, or the damage caused by sampling may cause injury stress, which may change the condition of the plant individual. Therefore, production performance is measured in a state different from normal growth, and the data may not represent the actual growth process. If the production performance of agricultural crops is predicted based on the data obtained in this way, there is a risk of the prediction accuracy decreasing. The prediction device 10 predicts production performance taking into account the change in production performance from year to year on a specific day, rather than the change in production performance over the course of a year, making it possible to make accurate predictions.

The prediction unit 13 may predict the current year's production performance using a prediction model 24 that represents the change in production performance over time and the influence of weather, and that is generated by referring to measurement data including actual measurement data of production performance for each past year and observed value data of weather data for those years. The prediction unit 13 reads out the prediction model 24 from the storage device 50, inputs the actual measurement data and the forecast value of the weather data for the current year sent from the input data acquisition unit 12 to the prediction model 24, and obtains the production performance for the current year output from the prediction model 24. Forecast values of the weather data can be obtained, for example, from a database published by an organization that provides weather information, such as the Japan Meteorological Agency. The prediction device 10 may store the acquired forecast values of the weather data in the storage device 50, and read them out from the storage device 50 when predicting the production performance in the prediction unit 13.

The prediction model 24 is generated by performing machine learning using measurement data including actual measurement data of past yearly production performance and observed weather data for those years as learning data. Such a prediction model 24 is a prediction model in which at least a portion of the actual measurement values of production performance up to the previous year and forecast values of weather data for the current year are input information, and the current year's production performance is output information. By using such a prediction model 24, production performance can be predicted taking into account the fluctuations in annual production performance of agricultural crops as well as weather data that affects production performance. By further taking into account weather data closely related to agriculture, more accurate predictions can be made than when predicting production performance by considering only the fluctuations in annual production performance.

The effects of aging and weather on production performance will be explained with reference to Figure 3. Figure 3 is a diagram explaining aging changes in agricultural crop production performance and the effects of weather on production performance. The graph in Figure 3 shows the year on the X axis and sugar content on the Y axis, and the graph shows the yearly change in sugar content on a specific day. It is believed that this yearly change in sugar content is influenced not only by aging changes in production performance but also by weather. Therefore, as shown in the graph on the right side of Figure 3, the yearly change in sugar content (solid line data) is decomposed into the change in sugar content due to aging (dashed line data) and the change in sugar content due to weather effects (arrange of arrows) represented by the difference between the dashed line data and the solid line data.

The sugar content, which indicates the effect of changes over time and is represented by the dashed line in Figure 3, is a predicted sugar content that takes into account the above-mentioned fluctuations in annual production performance, and is referred to as the standard sugar content (standard production performance). Because the standard sugar content is predicted taking into account only annual fluctuations, it may differ from the actual sugar content, which is represented by the solid line in Figure 3. The difference between this actual sugar content and the standard sugar content predicted taking into account annual fluctuations is considered to be the effect that weather has on sugar content.

Because forecast model 24 is a forecast model that forecasts the current year's production performance from measurement data that includes actual measurement data of production performance for each past year and observed weather data for those years, it is possible to forecast production performance that takes into account not only year-to-year fluctuations but also the influence of weather. Therefore, in the graph on the right side of Figure 3, a correction is made to the reference production performance represented by the dashed line within the range of the arrow representing the influence of weather, making it possible to forecast production performance with high accuracy that is closer to the actual production performance represented by the solid line.

The prediction model 24 is a prediction model generated for each prediction target location, and can be a model that predicts the production performance of the prediction target location for the current year based on at least a portion of the actual measured values of production performance up to the previous year at the prediction target location and the forecast values of weather data for the current year at the prediction target location.

The effects of changes over time and weather on production performance may differ depending on the region in which the crop is grown. Therefore, by predicting production performance using a prediction model 24 created for each prediction target point, which is the region in which the crop whose production performance is to be predicted is grown, it is possible to predict production performance more accurately.

The prediction device 10 can predict the current year's production performance based on at least a portion of the actual measured values of production performance up to the previous year, so that even in the early stages of the cultivation period, production performance at the harvest time or shipping date of that year can be predicted. This makes it possible to predict the production performance of that year early in the cultivation period, and to improve the production performance of that year by changing cultivation management such as fertilization, irrigation, and fruit thinning based on the prediction results. In addition, the optimal work and harvest times can also be determined based on the predicted production performance.

(Prediction model generating device 20)
The prediction model generating device 20 is a device for generating a prediction model for predicting the production performance of agricultural crops. The prediction model generating device 20 generates a prediction model that represents changes in production performance over time by referring to actual measured value data of production performance for each past year, and predicts the production performance for the current year from at least a part of the actual measured values of production performance up to the previous year.

The predictive model generating device 20 includes a control unit 21. The control unit 21 controls each part of the predictive model generating device 20, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions included in the program. This constitutes each part of the control unit 21. As each part, the control unit 21 includes a learning data acquisition unit 22 and a predictive model generating unit (generating unit) 23.

The learning data acquisition unit 22 acquires learning data for training the prediction model 24. The learning data acquisition unit 22 reads out the learning data from the storage device 50 based on an input signal representing an instruction to start learning from the input device 40. The learning data acquisition unit 22 may also acquire learning data input via the input device 40. The learning data acquisition unit 22 outputs the acquired learning data to the prediction model generation unit 23.

The prediction model generation unit 23 generates a prediction model 24 that represents changes in production performance over time by referring to actual measured production performance data for each past year, and predicts the current year's production performance from at least a portion of the actual measured production performance data for the previous year.

The prediction model generation unit 23 generates a prediction model 24 by performing machine learning using the learning data acquired from the learning data acquisition unit 22. Here, the learning data is actual measurement value data. The prediction model generation unit 23 generates the prediction model 24 using, for example, a known machine learning method such as a neural network, a decision tree, a random forest, or a support vector machine. The prediction model 24 generated by the prediction model generation device 20 is a prediction model that uses at least a portion of the actual measured values of production performance up to the previous year as input information and the production performance of the current year as output information.

In addition, as an example, the prediction model generation unit 23 may generate a statistical model that represents the correlation between the production performance for each year up to the year before a certain year and the production performance for that year. In other words, the prediction model 24 may be a statistical model such as a linear regression model, sparse modeling, a generalized linear model, a state space model, a hierarchical Bayes model, a time series model, or clustering.

The prediction model 24 is a model that represents changes in production performance over time by referring to actual measured production performance data for each past year, and can therefore be used to predict production performance taking into account fluctuations in the annual production performance of agricultural crops as changes over time. The prediction model 24 is a prediction model used to predict production performance in the prediction device 10 described above. The prediction model generation unit 23 may store the generated prediction model 24 in the storage device 50.

The prediction model generating unit 23 may also generate a prediction model 24 that represents changes in production performance over time and the effects of weather, by referring to measurement data including actual measurement data of production performance for each past year and observed value data of weather data for those years. Such a prediction model 24 is a prediction model that predicts production performance for the current year from at least some of the actual measurement values of production performance up to the previous year and forecast values of weather data for the current year. As an example, the observed value data of the weather data can be obtained from a database published by an organization that provides weather information, such as the Japan Meteorological Agency. The prediction model generating device 20 may store the obtained observed values of the weather data in the storage device 50, and read them from the storage device 50 when the prediction model generating unit 23 generates the prediction model 24.

The prediction model generation unit 23 performs machine learning using the measurement data as learning data to generate a prediction model in which at least a portion of the actual values of production performance up to the previous year and forecast values of weather data for the current year are input information, and the production performance for the current year is output information. In addition, as an example, the prediction model generation unit 23 may generate, as the prediction model 24, a statistical model that represents the correlation between the production performance for each year up to the previous year and the observed values of the weather data for that year, and the production performance for that year.

The prediction model 24 can be used to predict production performance, taking into account the fluctuations in crop production performance from year to year as well as meteorological data that affects production performance. By taking into account meteorological data closely related to agriculture, it can be used to make more accurate predictions than when predicting production performance by considering only the fluctuations in crop production performance from year to year.

It is preferable that the prediction model generation unit 23 generates a prediction model 24 for each prediction target location. The prediction model generation unit 23 generates a prediction model 24 for each prediction target location by referring to measurement data including actual measured values of production performance for each past year at the prediction target location and observed values of weather data for those years at the prediction target location.

The effects of changes over time and weather on production performance may differ depending on the region in which the crop is grown. Therefore, by generating a prediction model 24 for each prediction target point, which is the region in which the crop whose production performance is to be predicted is grown, it can be used to predict production performance with greater accuracy.

The prediction model generation unit 23 may regenerate the prediction model 24 based on the prediction result in order to improve prediction accuracy. If the prediction model 24 is a learning model generated by machine learning, the prediction model 24 may be re-learned. As an example, the prediction model generation unit 23 regenerates the prediction model 24 so as to minimize the difference between the actual production performance value for a certain past year and the predicted value of the production performance for that year predicted using the prediction model 24.

As an example, the prediction model generation unit 23 calculates for each year the difference between the actual measured values of production performance for each year up to the previous year and the predicted values of production performance for those years predicted using the prediction model 24, and regenerates the prediction model 24 so that the difference is minimized for all years. The prediction model generation unit 23 may also regenerate the prediction model 24 so that the sum of squares of the difference between the actual measured values of production performance for a certain past year and the predicted values of production performance for that year predicted using the prediction model 24 is minimized.

Furthermore, when regenerating the prediction model 24, the prediction model generation unit 23 may regenerate the prediction model using regression analysis in which the difference between the actual measured value of production performance in a past year and the predicted value of production performance for that year predicted by the prediction model 24 is regressed based on the observed value of the weather data for that year.

As explained with respect to the prediction device 10, referring to FIG. 3, the change in production performance from year to year can be decomposed into the change in production performance due to the influence of aging and the change in production performance due to the influence of weather. The prediction model generation unit 23 decomposes the change in production performance from year to year into the change in production performance due to the influence of aging and the change in production performance due to the influence of weather, taking into account the balance between the change in production performance due to the influence of aging and the change in production performance due to the influence of weather.

The prediction model generation unit 23 uses regularization, as an example, to calculate standard production performance, which is a predicted value of production performance taking into account changes over time. First, the strength of regularization is varied to calculate standard production performance for each regularization coefficient. Then, the difference between the measured production performance and the standard production performance for each regularization coefficient is found, and this difference is calculated as the influence of weather on the production performance. The influence of weather is then regressed using the observed values of the weather data for that year. The prediction model generation unit 23 repeats this process to determine the regularization coefficient, standard production performance, and weather data regression coefficient that minimize the difference between the predicted value and the actual measured value of production performance, and regenerates the prediction model 24.

An example of a regression analysis of the effect of weather on production performance using weather data, which is performed by the prediction model generation unit 23 when regenerating the prediction model 24, may be a regression analysis performed by the prediction model generation unit 33 in the prediction model generation device 30 described below.

In this way, by regenerating the prediction model 24, it is possible to generate a prediction model 24 that can predict production performance with greater accuracy.

The prediction model generation device 20 can generate a prediction model that predicts the current year's production performance based on at least a portion of the actual measured values of production performance up to the previous year, so it can also be used to predict production performance for the harvest season or shipping date of that year at the beginning of the cultivation period. This makes it possible to predict the production performance of that year at the beginning of the cultivation period and improve the production performance of that year by changing cultivation management such as fertilization, irrigation, and fruit thinning based on the prediction results. It can also be used to determine the optimal work and harvest times based on the predicted production performance.

(Prediction model generating device 30)
The weather data prediction model generation device 30 is a device for generating a prediction model that predicts the production performance of agricultural crops. The prediction model generation device 30 generates a prediction model 34 that is a prediction model that represents the production performance and the influence of weather by referring to observed value data of the weather data before the target prediction date, and predicts the production performance from the forecast value of the weather data on the target prediction date. The prediction model generation device 30 generates the prediction model 34 by selecting a combination of variables of the weather data so as to minimize the difference between the actual measured value of the production performance in a certain past year and the predicted value of the production performance for that year predicted using the prediction model 34.

The predictive model generating device 30 includes a control unit 31. The control unit 31 controls each part of the predictive model generating device 30, and is realized by a processor and a memory, for example. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions contained in the program. This constitutes each part of the control unit 31. As each part, the control unit 31 includes a learning data acquisition unit 32 and a predictive model generating unit 33.

The learning data acquisition unit 32 acquires learning data for training the prediction model 34. The learning data acquisition unit 32 reads out the learning data from the storage device 50 based on an input signal representing an instruction to start learning from the input device 40. The learning data acquisition unit 32 may also acquire learning data input via the input device 40. The learning data acquisition unit 32 outputs the acquired learning data to the prediction model generation unit 33.

The prediction model generation unit 33 generates a prediction model 34, which is a prediction model that represents production performance and the effects of weather by referring to observed values of weather data prior to the prediction target date, and predicts production performance from the forecast values of the weather data for the prediction target date.

The prediction model generation unit 33 generates a prediction model 34 that represents production performance and the effects of weather, by referring to the learning data including observed values of daily weather data and actual values of production performance on those days. The prediction model generation unit 33 generates the prediction model 34 by performing machine learning using the learning data acquired from the learning data acquisition unit 22. The prediction model generation unit 33 generates the prediction model 34 by using, for example, known machine learning methods such as neural networks, decision trees, random forests, and support vector machines.

In addition, as an example, the prediction model generation unit 33 may generate a statistical model that represents the correlation between daily weather data and the production performance of those days. In other words, the prediction model 34 may be a statistical model such as a linear regression model, sparse modeling, a generalized linear model, a state space model, a hierarchical Bayesian model, a time series model, or clustering.

The prediction model generation device 30 generates a prediction model 34 by selecting a combination of weather data variables so as to minimize the difference between the actual production performance of a certain past year and the predicted production performance of that year predicted using the prediction model 34. Weather data that is expected to affect the production performance of agricultural crops includes many variables such as precipitation, average temperature, maximum temperature, minimum temperature, total solar radiation, and hours of sunshine. Some weather data variables are interdependent, and as an example, there are combinations of variables that are highly interdependent, such as lower temperatures on days when it rains (the relationship between precipitation and temperature). Therefore, by selecting a combination of weather data variables that is optimal for predicting production performance, a prediction model that can accurately predict production performance can be generated.

The effect of combinations of weather data variables on production performance will be explained with reference to Figure 4. Figure 4 is a diagram explaining the effect of weather data variables on production performance. In Figure 4, the graph on the top left shows the effect of precipitation on sugar content, the graph in the top center shows the effect of average temperature on sugar content, and the graph on the top right shows the effect of sunshine hours on sugar content. In these graphs, the X-axis shows the date, the Y-axis shows the increase or decrease in sugar content, and the Z-axis shows the measurement location. In other words, these graphs show the effect of weather data variables on production performance for each measurement location.

In Figure 4, the bottom graph is an overlay of the top three graphs, showing only the graphs for specific measurement locations. In other words, this graph shows the overall effect of each weather data variable on sugar content. In the bottom graph, the solid line shows the effect of precipitation, the dotted line shows the effect of average temperature, and the circle plots show the effect of sunshine hours.

As shown in the top graph of Figure 4, the degree to which production performance is affected by variables varies from place to place and from day to day. As shown in the bottom graph of Figure 4, by overlaying the graphs of each variable, it becomes clear that the types and degrees of affected variables vary even for the same place and the same day.

In this way, by selecting the optimal combination of variables for predicting production performance based on the degree of influence of weather data variables on production performance, a prediction model that can predict production performance with greater accuracy can be generated. In addition, by selecting the optimal combination of variables for each day or location, a prediction model that can predict production performance with greater accuracy can be generated.

The prediction model generation unit 33 selects a combination of weather data variables by regression analysis based on the integrated value or average value for each variable of the weather data before the target prediction date. The prediction model generation unit 33 calculates the integrated value for each variable over a specified period or the average value within each specified period, and performs regression analysis based on the calculated integrated value or average value to select a combination of variables that minimizes the difference between the actual production performance value and the predicted value. As an example, the prediction model generation unit 33 selects the optimal combination of variables by performing regression analysis using machine learning that can analyze highly interdependent variables.

The prediction model generation unit 33 selects a combination of weather data variables by regression analysis based on the start and end dates for calculating the integrated value or average value for each variable, and the number of calculation days. The impact of weather on production performance can change not only on an annual basis, but also on short time periods such as the flowering period and fruit setting period. Therefore, by performing regression analysis based on the start and end dates for the integration of each variable, or the start and end dates for the calculation of the average value of each variable, it is possible to select the optimal combination of variables for each period. Similarly, by performing regression analysis based on the number of days for integration or the number of days for averaging (number of calculation days) for each variable, it is possible to select the optimal combination of variables for a shorter time period.

The prediction model generation unit 33 selects a combination of the weather data variables by regression analysis based on the integrated value or average value of the calculated number of days for each variable of the weather data for each day from the start date to the end date. In this way, by calculating the integrated value or average value of each variable for the calculated number of days for each day within the calculation period from the start date to the end date, the window for the calculated number of days is shifted by one day, and a vector representing the influence of each variable on the production performance for each day within the calculation period can be created. Then, a regression analysis is performed using the created daily vector, and a combination of variables that minimizes the difference between the actual production performance and the predicted production performance is selected, thereby selecting the combination of variables that is optimal for predicting the production performance. In addition, by performing regression analysis using the daily vector, a combination of variables can be selected taking into consideration the influence on the production performance of the continuity of the variables from day to day, such as the influence of a heat wave that continues for days.

The prediction model generating unit 33 may generate a prediction model 34 that represents the change over time in production performance and the influence of weather, generated by referring to measurement data including the actual measurement value data of the production performance for each past year and the observation value data of the weather data for those years, and predicts the production performance from at least a part of the actual measurement value of the production performance up to the previous year and the forecast value of the weather data for the prediction target day. In addition, such a prediction model 34 may be generated by performing machine learning using the measurement data as learning data, thereby generating a prediction model 34 in which at least a part of the actual measurement value of the production performance up to the previous year and the forecast value of the weather data for the prediction target day are input information and the production performance for the current year is output information. In other words, the prediction model generating unit 33 may generate a prediction model 34 that represents the change over time in production performance and the influence of weather by regressing the influence of weather when the change in production performance for each year is decomposed into the influence of the change over time and the influence of weather in the prediction model generating unit 23 of the prediction model generating device 20 described above.

The prediction model generating unit 33 preferably generates a prediction model 34 for each prediction target location. The prediction model generating unit 33 generates a prediction model 24 for each prediction target location by referring to measurement data including actual values of production performance for each past year at the prediction target location and observed values of meteorological data for those years at the prediction target location. By generating a prediction model 34 for each prediction target location, which is an area where the agricultural crops whose production performance is to be predicted grow, it can be used to predict production performance with higher accuracy.

The prediction model generation device 30 can predict the current year's production performance from the forecasted weather data for the prediction date, so it can also be used to predict production performance for the harvest season or shipping date of that year early in the cultivation period. This makes it possible to predict the production performance of that year early in the cultivation period and change cultivation management such as fertilization, irrigation, and fruit thinning based on the prediction results, thereby improving the production performance of that year. It can also be used to determine the optimal work and harvest times based on the predicted production performance.

(Prediction of production performance in prediction system 100)
The concept of the flow of the process of predicting the production performance using the prediction model in the prediction system 100 will be described with reference to Fig. 5. Fig. 5 is a diagram for explaining the concept of predicting the production performance of agricultural crops using the prediction model, which is executed by the prediction system according to one embodiment of the present invention. Note that the variables of the weather data shown in Fig. 5 are merely examples, and the variables of the weather data are not limited thereto.

As shown in Figure 5, a prediction model that has been trained using measurement data including actual production performance data for each past year and observed weather data for those years is used. The actual (past) production performance values up to the previous year and forecast weather data for the current year are input into the prediction model. The prediction model divides the predicted value of the production performance into the base production performance for the current year and the influence of the weather for the current year. The prediction model regresses the influence of the weather for the current year based on a combination of weather data variables and selects the optimal combination of variables. The prediction model regresses the influence of the weather for the current year based on the selected combination of variables. The prediction model outputs a predicted value of the production performance for the current year from the base production performance for the current year and the influence of the weather for the current year.

(Prediction process flow)
The flow of the prediction process (prediction method) by the prediction device 10 will be described with reference to FIG. 6. FIG. 6 is a flowchart showing an example of the prediction process executed by the prediction device according to one aspect of the present invention. As shown in FIG. 6, first, the input data acquisition unit 12 acquires the actual measured values of (past) production results up to the previous year (step S1). Next, the input data acquisition unit 12 acquires the forecast value of the weather data for the current year (step S2). Then, the prediction unit 13 inputs the acquired actual measured values of the production results up to the previous year and the forecast value of the weather data for the current year into the prediction model 24, and acquires the output forecast value of the production results for the current year (step S3, prediction process). The prediction unit outputs the acquired forecast value to the output device 60 as a prediction result (step S4), and ends the process.

(Flow of prediction model generation process)
A flow of a process of generating a prediction model 24 (a method of generating a prediction model) by the prediction model generating device 20 will be described with reference to FIG. 7. FIG. 7 is a flowchart showing an example of a prediction model generating process executed by the prediction model generating device according to one aspect of the present invention. As shown in FIG. 7, first, the learning data acquiring unit 22 acquires actual values of production performance for each past year (step S11). Next, the learning data acquiring unit 22 acquires observed values of weather data for those years (step S12). Then, the learning data acquiring unit 22 associates the actual values of production performance for each past year with the observed values of weather data for those years, and generates learning data (measurement data) (step S13). The prediction model generating unit 23 uses the generated learning data to generate a prediction model 24 in which the actual values of (past) production performance up to the previous year and the forecast values of weather data for the current year are input information, and the production performance for the current year is output information (step S14, generating process). The prediction model generating unit 23 stores the generated prediction model 24 in the storage device 50.

(Flow of weather data forecast model generation process)
The flow of the generation process (generation method of a prediction model) of the prediction model 34 by the prediction model generation device 30 will be described with reference to FIG. 8. FIG. 8 is a flowchart showing an example of the generation process of a prediction model executed by the weather data prediction model generation device according to one aspect of the present invention. As shown in FIG. 8, first, the prediction model generation unit 33 sets the start date and end date for calculating the integrated value or average value of each variable of the weather data, and the number of days to be integrated or averaged (calculation number of days) (step S21). Next, the prediction model generation unit 33 calculates the integrated value or average value of the calculation number for each variable for each day from the start date to the end date (step S22). Then, the prediction model generation unit 33 regresses the prediction model 34 using the calculated value to obtain the error between the predicted value and the actual value (step S23). The prediction model generation unit 33 obtains the combination of variables that minimizes the error, the start date and end date of the integrated value or average value, and the calculation number of days (step S24). The prediction model generating unit 33 generates a prediction model 34 based on the obtained combination of variables, the start date and end date of the integrated value or average value, and the calculated number of days (step S25, generating step). The prediction model generating unit 33 stores the generated prediction model 34 in the storage device 50 (step S26).

[Software implementation example]
The functions of the prediction device 10, the prediction model generating device 20, and the prediction model generating device 30 (hereinafter referred to as the "devices") can be realized by programs for causing a computer to function as these devices, and for causing a computer to function as each control block of these devices (in particular, each part included in the control unit 11, the control unit 21, and the control unit 31).

In this case, these devices are equipped with a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the above program. The functions described in the above embodiment are realized by executing the above program using this control device and storage device.

The above program may be recorded on one or more computer-readable recording media, not on a temporary basis. These devices may or may not have these recording media. In the latter case, the above program may be supplied to these devices via any wired or wireless transmission medium.

In addition, some or all of the functions of each of the above control blocks can be realized by a logic circuit. For example, an integrated circuit in which a logic circuit that functions as each of the above control blocks is formed is also included in the scope of the present invention. In addition, it is also possible to realize the functions of each of the above control blocks by, for example, a quantum computer.

The processes described in the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI may run on the control device, or on another device (such as an edge computer or a cloud server).

[Example]
For mandarin oranges, sugar content and acidity were predicted using a prediction model generated using past sugar content and acidity data at multiple target prediction locations, and the accuracy of the prediction was verified. The accuracy of sugar content and acidity predictions was verified by comparing actual measured values in fiscal year 2020 with values predicted using the prediction model. Weather observation values for fiscal years 2018 and 2019, and weather forecast values as of March 22, fiscal year 2020 were used.

The root mean square error (RMSE) between predicted and measured values was calculated for all prediction locations and all prediction dates over the two-year period. When meteorological observation values were used in sugar content prediction, the RMSE was 0.6561 (Figure 9). When meteorological forecast values were used in sugar content prediction, the RMSE was 1.5267 (Figure 10). Next, when meteorological observation values were used in acidity prediction, the RMSE was 0.2517 (Figure 11). When meteorological forecast values were used in acidity prediction, the RMSE was 0.2636 (Figure 12).

In addition, sugar content and acidity for fiscal year 2020 were predicted based on weather forecast values as of March 22, 2020, and accuracy was verified using measured values up to November 9, 2020. Weather forecast values were used to verify accuracy up to November 9, 2020. The RMSE for sugar content prediction was 0.7355 (Figure 13). The RMSE for acidity prediction was 0.1615 (Figure 14).

In this way, it was shown that by using a predictive model, it is possible to make predictions with sufficient accuracy as of March to grasp the trends in production performance for the following year.

Figure 15 shows the influence of three meteorological variables by period in the prediction of high sugar content sorting in FY2020 at a certain forecast location. In Figure 15, the solid line indicates the influence of precipitation, the dotted line indicates the influence of average temperature, and the circle plot indicates the influence of sunshine hours. Note that the weather forecast values as of March 22, 2020 are used, so the influence of the actual weather in FY2020 differs. When the weather is normal from late April, which is the normal value, the influence of precipitation on sugar content is the largest among precipitation, average temperature, and sunshine hours. As shown in Figure 15, the precipitation up to the end of June 2020 was found to have the effect of revising the sugar content at the time of sorting up by up to 0.01 per day. On the other hand, the precipitation in July and August 2020 was found to have the effect of revising the sugar content at the time of sorting down by 0.01 to up to 0.025 per day. Although mulch sheet covering was used during this period, it was suggested that the influence of precipitation in July and August was not zero.

Other Embodiments
(Aspect of using average production results)
The prediction device 10 may predict the current year's production performance from at least a part of the actual measured values of the production performance up to the previous year using a prediction model that represents the average value of the production performance, which is generated by referring to the actual measured value data of the production performance for each past year. As an example, when the accumulated years of the production performance for each past year are short, such as about 2 to 5 years, a prediction model that represents the average value of the production performance may be used instead of a prediction model that represents the change in the production performance over time. Here, the average value of the production performance may be a value obtained by averaging the production performance for each year for a predetermined number of years.
Furthermore, the prediction model 24 generated by the prediction model generating device 20 and the prediction model 34 generated by the prediction model generating device 30 may be prediction models that represent the average value of production performance and the influence of weather, generated by referring to measurement data including actual measurement value data of production performance for each past year and observed value data of weather data for those years. Such prediction models 24 and 34 can predict production performance by taking into account the annual fluctuation of agricultural crops as an average value.

In other words, in each of the above-mentioned embodiments, the technical scope of the present invention also includes embodiments in which all of the descriptions regarding changes in production performance over time are replaced with average production performance values.

(Aspect in which measurement data of soil components is used as measurement data)
The prediction device 10 may predict the current year's production performance using a prediction model that represents the secular change or average value of the production performance and the influence of weather, which is generated by referring to the actual measurement value data of the production performance for each past year and the measurement data including at least one of the observation value data of the weather data for those years and the measurement value data of the soil components for those years. Here, the measurement value data of the soil components may be, for example, soil component values (soil moisture value, soil pH value, etc.) measured by a soil sensor such as a soil moisture meter installed in a field where agricultural crops are grown.

The prediction model 24 generated by the prediction model generating device 20 may be a prediction model that represents the change over time or average value of the production performance and the influence of at least one of the weather and the soil components, generated by referring to the actual measurement value data of the production performance for each past year and the measurement data including at least one of the observation value data of the weather data and the measurement data of the soil components for those years. Such a prediction model 24 is generated by performing machine learning using the measurement data as learning data, and at least a part of the actual measurement value of the production performance up to the previous year and at least one of the forecast value of the weather data for the current year and the measurement value of the soil components for the current year are input information, and the production performance for the current year is the output value. According to the prediction model 24, the production performance can be predicted taking into account the soil components.

The prediction model 24 is a prediction model generated for each prediction target point, and may predict the production performance for the current year at the prediction target point from at least a portion of the actual measured values of the production performance at the prediction target point up to the previous year and at least one of the forecast values of meteorological data for the current year at the prediction target point and the measured values of soil components for the current year.
The prediction model 24 may be regenerated using regression analysis in which the difference between the actual measured production performance for a past year and the predicted production performance for that year predicted by the prediction model 24 is regressed based on at least one of the observed meteorological data for that year and the measured values of soil components for that year.

The prediction model 34 generated by the prediction model generating device 30 is a prediction model that represents the production performance and the influence of at least one of the weather and soil components, generated by referring to at least one of the observed value data of the weather data and the measured value data of the soil components before the prediction target date, and can be a prediction model that predicts the production performance from at least one of the forecast value of the weather data and the measured value of the soil components on the prediction target date.

Such a prediction model 34 is generated by selecting a combination of weather data variables and a combination of soil component variables so as to minimize the difference between the actual production performance of a certain past year and the predicted production performance of that year predicted using the prediction model 34. These combinations of variables may be selected by regression analysis based on the integrated values or average values of each variable of the weather data and soil components prior to the prediction target date. Furthermore, these combinations of variables may be selected by regression analysis based on the start and end dates for calculating the integrated values or average values, and the number of days for calculation. Furthermore, these combinations of variables may be selected by regression analysis based on the integrated values or average values of the number of days for calculation for each day from the start date to the end date for each variable of the weather data and soil components.

The prediction model 34 is a prediction model that represents the change over time or average value of the production performance and the influence of at least one of the weather and the soil components, generated by referring to the actual measurement data of the production performance for each past year and the measurement data including at least one of the observation value data of the weather data for those years and the measurement value data of the soil components for those years, and predicts the production performance from at least a part of the actual measurement values of the production performance up to the previous year and at least one of the forecast value of the weather data and the measurement value of the soil components for the prediction target day. Such a prediction model 34 can be generated by performing machine learning using the measurement data as learning data. Such a prediction model 34 is a prediction model in which at least a part of the actual measurement values of the production performance up to the previous year and at least one of the forecast value of the weather data and the measurement value of the soil components for the prediction target day are input information, and the production performance of the current year is output information.

The effect of soil components on production performance will be explained with reference to Figure 16. Figure 16 is a diagram explaining the effect of soil moisture value on production performance. The X-axis in the graph shown in Figure 16 shows the date, and the Y-axis shows the effect value of soil moisture value converted into sugar content. The graph shown in Figure 16 shows the effect of soil moisture value on sugar content at harvest.

As shown in Figure 16, the degree to which soil moisture value affects sugar content at harvest varies from day to day. As an example, Figure 16 shows that soil moisture value from July to September has a large impact on sugar content at harvest. This finding is consistent with existing knowledge in the fruit tree field that drought stress in July and August increases sugar content, and the effect saturates thereafter.

In this way, by predicting production performance taking into account soil components, it is possible to predict production performance with greater accuracy. Furthermore, by predicting production generators taking into account both weather and soil components, it is expected that prediction accuracy will be further improved.

In other words, in each embodiment described in this specification, the technical scope of the present invention also includes embodiments in which all descriptions of meteorological data are replaced with (i) measurement data of soil components, or (ii) meteorological data and measurement data of soil components.

(Selection of variables based on information criteria)
The selection of the combination of meteorological data variables and the combination of soil component variables in the prediction model 34 generated by the prediction model generation device 30 may be performed based on an information criterion. In other words, in the method for generating a prediction model, an embodiment in which a combination of meteorological data variables and a combination of soil component variables are selected based on an information criterion to generate the prediction model 34 is also included in the technical scope of the present invention. Here, as the information criterion, a conventionally known information criterion can be used, and examples thereof include the Kullback-Leibler information criterion, the Akaike information criterion, and the Bayesian information criterion.

As an example, if the estimation algorithm of each prediction model generated for a combination of meteorological data variables and soil component variables does not include a model selection criterion, the combination of meteorological data variables and soil component variables may be selected based on an information criterion. This allows the variable selection process to be performed uniformly without being affected by the operator's knowledge or level of expertise.

(Predicting Growth Stage)
The prediction device 10 may predict the growth stage of the crop by comparing the predicted production performance with a reference value that indicates the performance of each growth stage of the crop. Here, the growth stage of the crop refers to the flowering period, harvest period, transplanting period, etc. of the crop. The prediction of the growth stage in the prediction device 10 may be performed by the prediction unit 13. As an example, when a predicted value of the production performance such as sugar content or acidity obtained by the prediction model 24 exceeds a preset reference value (threshold value), the prediction unit 13 of the prediction device 10 determines that the harvest standard is met and predicts that the time when the production performance was predicted is the harvest time.

The prediction device 10 may also calculate the amount of influence of at least one of the weather and soil components in the prediction model, and predict the growth stage of the crops by referring to the calculated amount of influence.

Agricultural crops include those that are grown indoors as seedlings and then transplanted outdoors, and those that are harvested and stored in the shade under a roof. For such crops, if the prediction model generation unit 23 and the prediction model generation unit 33 calculate the amount of influence (influence portion) of weather or soil components on production performance, it is possible to detect the date on which the amount of influence switches discretely. For example, in the case of crops that are stored in the shade under a roof after harvest, the date on which the amount of influence of sunlight hours suddenly becomes almost zero corresponds to the date on which the crops are transferred to a storage environment under a roof after harvest. Therefore, the prediction device 10 can predict the growth stage of the crops by referring to such influence amounts.

(Prediction device for predicting growth stages)
The technical scope of the present invention also includes a growth stage prediction device that predicts the growth stage transition of the current year from at least a part of the actual growth stage transition up to the previous year and at least one of the forecast values of the weather data for the current year and the measured values of the soil components for the current year, using a prediction model that represents the growth stage transition and the influence of at least one of weather and soil components, which is generated by referring to data representing the actual growth stage transition of agricultural crops for each past year and measurement data including at least one of the observed value data of the weather data for those years and the measured value data of the soil components for those years. Here, the growth stage transition is intended to be, for example, information on the date of change of each growth stage of the agricultural crop, such as the flowering date, harvest date, transplanting date, etc.

That is, the prediction model used by such a growth stage prediction device is a growth stage prediction model that represents the change over time or average value of the growth stage transition of agricultural crops and the influence of at least one of weather and soil components. The growth stage prediction device inputs at least a portion of the actual growth stage transition up to the previous year and at least one of the forecast values of weather data for the current year and the measured values of soil components for the current year into the growth stage prediction model to obtain the growth stage transition that is output. As an example, by using the growth stage prediction model, the growth stage prediction device can obtain prediction results such as the date of flowering, the number of days from a specified date to flowering, and the number of days from sowing date to flowering.

In addition, in the embodiments described in this specification, embodiments in which all of the descriptions regarding prediction of production performance are replaced with prediction of the progression of growth stages are also included in the technical scope of the present invention.

(Aspects of calculating potential value of production performance)
Another embodiment of the prediction device will be described with reference to Fig. 17. For ease of explanation, the same reference numerals are given to components having the same functions as those described in the prediction system 100 and the prediction device 10 described above, and the description thereof will not be repeated. Fig. 17 is a block diagram showing the main configuration of a prediction system 200 according to another embodiment of the present invention. The prediction system 200 differs from the prediction system 100 shown in Fig. 1 in that the control unit 211 of the prediction device 210 includes an input value generating unit 212, an output value acquiring unit 213, a setting unit 214, and a calculating unit 215.

The input value generation unit 212 generates simulation input values of at least one of the observed values of meteorological data and the measured values of soil components based on at least one of the observed values of past meteorological data and the measured values of soil components (input value generation process). The input value generation unit 212 replaces at least one of the observed values of meteorological data and the measured values of soil components acquired by the input data acquisition unit 12 on a daily basis to generate simulation input values. The data replacement is performed by replacing it with data for the same day in a different past year at the same location. In other words, the simulation input value generated by the input value generation unit 212 can be at least one of the observed values of meteorological data and the measured values of soil components, which are data for multiple years patched together on a daily basis. The input value generation unit 212 outputs the generated simulation input value to the output value acquisition unit 213.

The output value acquisition unit 213 acquires a simulation output value of the production performance predicted from the simulation input value using the prediction model 24 or the prediction model 34 (a process of acquiring an output value). The output value acquisition unit 213 inputs the simulation input value to the prediction model 24 or the prediction model 34, and acquires the predicted value of the production performance to be output as a simulation output value. The output value acquisition unit 213 outputs the acquired simulation output value to the setting unit 214.

The setting unit 214 sets the simulation input value when the obtained simulation output value is the highest as the potential value (setting process). The setting unit 214 compares the simulation output values obtained for various simulation input values, and sets the simulation input value when the simulation output value is the highest as the potential value. In other words, the setting unit 214 compares production performance predicted using multiple pieces of weather data or soil component data with different patchwork patterns, and sets the potential value. The setting unit 214 outputs the set potential value to the calculation unit 215.

The calculation unit 215 calculates the simulation input value when the potential value is obtained as the optimal environmental value (calculation process). That is, the calculation unit 215 calculates the weather data or soil component data when the best production performance is obtained, and calculates at least one of the weather conditions and soil component conditions for obtaining the production performance, as the optimal environmental value.

In the prediction device 210, simulation input values in which at least one of the observed values of meteorological data and the measured values of soil components is replaced with data for the same day in a different past year at the same location are used as virtual environmental conditions, and the production performance predicted based on the virtual environmental conditions is used as the potential value. Such simulation input values are generated by, for example, executing a process for all dates in the growing season, such as replacing "maximum temperature, hours of sunshine, and amount of precipitation on March 3, 2010" with "maximum temperature, hours of sunshine, and amount of precipitation on March 3, 2011." In this way, by comparing production performance for each virtual environmental condition, the most ideal environmental conditions can be predicted.

The input value generation unit 212 may determine which weather conditions or soil component conditions to replace at which time based on information regarding the influence of weather or soil components on production performance obtained from the prediction model generation unit 23 and the prediction model generation unit 33. The input value generation unit 212 may also generate simulation input values using an algorithm that determines the data to be replaced based on information regarding the influence of weather or soil components on production performance.

The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. The technical scope of the present invention also includes embodiments obtained by appropriately combining the technical means disclosed in different embodiments.

10, 210 Prediction device 13 Prediction unit 20 Prediction model generation device 23 Prediction model generation unit (generation unit)
30 Prediction model generating device 33 Prediction model generating unit 100, 200 Prediction system

Claims (7)

A method for generating a prediction model for predicting crop production performance, comprising:
the generating method is executed by a computer under the control of software;
A computer executes a step of generating a prediction model that represents a change in the production performance over time by referring to actual measurement value data of the production performance on a specific day of each past year, the prediction model predicting the production performance of the current year from at least a part of the actual measurement values of the production performance up to the previous year;
In the generating step, machine learning is performed using actual measured value data of the production performance on a specific day of each year as learning data to generate a predictive model in which at least a portion of the actual measured values of the production performance up to the previous year are input information and the production performance for the current year are output information. The generating method according to claim 1, wherein in the generating step, machine learning is performed using as learning data actual measured value data of the production performance on a specific day of each year and measurement data including at least one of observed value data of meteorological data for those years and measured value data of soil components for those years, to generate a predictive model that represents changes in the production performance over time and the influence of at least one of weather and soil components, wherein at least a portion of the actual measured values of the production performance up to the previous year and at least one of forecast values of meteorological data for the current year and measured values of soil components for the current year are input information, and the production performance for the current year is output information. The generation method of claim 2, wherein in the generating step, a prediction model is generated for each prediction target location by performing machine learning using as learning data actual measured value data of the production performance on a specific day of each year at the prediction target location and measurement data including at least one of observed value data of meteorological data for those years at the prediction target location and measured values of soil components for those years. The method of claim 2 or 3, further comprising the step of re-learning the prediction model so that the difference between the actual value of the production performance in a past year and the predicted value of the production performance in that year predicted using the prediction model is minimized. The method of claim 4, wherein in the re-learning step, the prediction model is re-learned using a regression analysis that regresses the difference between the actual measured value of the production performance for a certain past year and the predicted value of the production performance for that year predicted by the prediction model based on at least one of the observed values of meteorological data for that year and the measured values of soil components for that year. A prediction model generating device for predicting crop production performance, comprising:
a generation unit that generates a prediction model that represents a change in the production performance over time by referring to actual measurement value data of the production performance on a specific day of each past year, and that predicts the production performance of the current year from at least a part of the actual measurement values of the production performance up to the previous year;
The generation unit performs machine learning using actual measured value data of the production performance on a specific day of each year as learning data to generate a predictive model in which at least a portion of the actual measured values of the production performance up to the previous year are input information and the production performance for the current year are output information. A program for generating a prediction model for predicting crop production performance,
a program for causing a computer to execute a step of generating a prediction model that represents a change in the production performance over time by referring to actual measurement value data of the production performance on a specific day of each past year, and predicts the production performance of the current year from at least a part of the actual measurement values of the production performance up to the previous year,
In the generating step, a generation program performs machine learning using actual measured value data of the production performance on a specific day of each year as learning data to generate a predictive model in which at least a portion of the actual measured values of the production performance up to the previous year are input information and the production performance for the current year are output information.

Images