JP7743876B2 - Growth prediction device, growth prediction method, and growth prediction program - Google Patents

Description

The disclosed technology relates to a growth prediction device, a growth prediction method, and a growth prediction program.

With the agricultural population declining, there is a demand for new farmers. To enable new farmers to perform farm work in the same way as experienced farmers, agricultural support systems (hereafter referred to as "agricultural support systems") are being launched. To ensure that appropriate farm work is performed at the appropriate time, agricultural support systems are required to have the ability to predict and suggest farm work times.

When it comes to paddy rice, Japan's staple food, it is said that fertilization, which is done to improve yields, should be done approximately 15 to 20 days before heading. Fertilization refers to the application of fertilizer. Furthermore, mowing at least one week before heading is said to be effective in preventing damage from spotted rice stink bugs, and many paddy rice farming practices (top dressing, pesticide spraying, etc.) are most effective when carried out around the heading date. Therefore, to support paddy rice farming practices, it is important to predict the heading date or the growth stage around the heading date (the young panicle differentiation stage, flag leaf formation stage, etc.) in advance.

SIMRIW (Simulation Model for Rice-Weather Relationships) is a well-known technology for rice growth models (see, for example, Non-Patent Document 1). SIMRIW is a highly reliable rice growth model that is widely used in Japan and has been adopted in the cultivation management support system provided by the National Agriculture and Food Research Organization (NARO).

SIMRIW can calculate the heading date and optimum harvest time by calculating the rice growth index (DVI) from weather data and variety information, as shown in Figure 9, for example.

Takeshi Horie, Hiroshi Nakagawa, "Study on Modeling and Prediction of Rice Development Processes," Journal of the Crop Science Society of Japan, 59(4): 687-695. (1990)

However, because SIMRIW only takes weather data and variety information as input, it does not take into account human errors that occur in actual farming work, resulting in discrepancies between actual growth conditions and calculated growth conditions.

In addition, the parameters used to calculate DVI differ for each variety, and in order to obtain accurate data, data must be collected for each variety, requiring a huge amount of data.

One parameter that can be measured and changes as rice grows is the Leaf Area Index (LAI). LAI represents the ratio of rice leaf area to a unit area of the ground. As shown in Figure 10, for example, LAI is an index of growth status that increases as rice grows, and peaks when the flag leaf appears, regardless of variety. By predicting the peak timing, it is possible to determine 10 days before heading. This is expected to enable agricultural work planning.

Like DVI, the maximum value of LAI also varies by variety, and also by agricultural conditions such as weather conditions and whether or not fertilizer is applied. For example, as shown in Figure 11, the trend in LAI differs depending on whether or not fertilizer is applied.

In other words, when predicting rice growth conditions using LAI or other methods, it is necessary to do so for each variety, and therefore a prediction model must be built for each variety. In this case, it is not possible to predict the growth of new varieties.

The disclosed technology has been developed in consideration of the above points, and aims to provide a growth prediction device, growth prediction method, and growth prediction program that can predict the growth status of rice without distinguishing between rice varieties.

A first aspect of the present disclosure is a growth prediction device comprising: a data input unit that receives input of time-series measurement data of index values that change with the growth of rice, cultivation information related to rice cultivation, and environmental information related to the environment in which the rice is cultivated; a maximum value estimation unit that estimates the maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information; a normalization unit that normalizes the measurement data of the index value from the measurement data of the index value and the maximum value of the index value to calculate normalized measurement data; and a prediction unit that predicts future trends in the index value from the normalized measurement data of the index value calculated by the normalization unit using a trained model that is generated by machine learning the normalized measurement data of the index values obtained in advance as training data, and that inputs the normalized measurement data of the index values and outputs predicted data that represents future trends in the index value.

A second aspect of the present disclosure is a growth prediction method that receives input of time-series measurement data of index values that change with the growth of rice, cultivation information related to rice cultivation, and environmental information related to the environment in which the rice is cultivated; estimates the maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information; normalizes the measurement data of the index value from the measurement data of the index value and the maximum value of the index value to calculate normalized measurement data; and predicts future trends in the index value from the calculated normalized measurement data of the index value using a trained model that is generated by machine learning using the normalized measurement data of the index values obtained in advance as training data, and that takes the normalized measurement data of the index values as input and outputs prediction data that represents future trends in the index values.

A third aspect of the present disclosure is a growth prediction program that causes a computer to receive input of time-series measurement data of index values that change with the growth of rice, cultivation information related to rice cultivation, and environmental information related to the environment in which the rice is cultivated; estimate the maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information; calculate normalized measurement data by normalizing the measurement data of the index value from the measurement data of the index value and the maximum value of the index value; and predict future trends in the index value from the calculated normalized measurement data of the index value using a trained model that is generated by machine learning using the normalized measurement data of the index values obtained in advance as training data, and that takes the normalized measurement data of the index values as input and outputs prediction data that represents future trends in the index values.

The disclosed technology has the effect of making it possible to predict the growth status of rice without distinguishing between rice varieties.

1 is a diagram illustrating an example of a configuration of a farm work support system according to an embodiment. FIG. 1 is a block diagram showing an example of a hardware configuration of a growth prediction device according to an embodiment. 1 is a block diagram showing an example of a functional configuration of a growth prediction device according to an embodiment. FIG. 10 is a diagram illustrating an example of a plurality of variables included in environmental information according to the embodiment. 10 is a graph showing an example of a relationship between actual measurement data and predicted data of LAI according to the embodiment. 10 is a graph showing an example of LAI when the LAI is predicted using all environmental information. 10 is a graph showing an example of LAI when the LAI is predicted using environmental information selected according to the contribution rate to the LAI. 10 is a flowchart showing an example of the flow of a learning process by the growth prediction program according to the embodiment. 10 is a flowchart showing an example of the flow of a prediction process by a growth prediction program according to an embodiment. FIG. 1 is a diagram illustrating the prior art. FIG. 1 shows the relationship between LAI and heading date. FIG. 1 is a diagram showing the difference in LAI transition depending on whether fertilization is performed or not.

An example of an embodiment of the disclosed technology will be described below with reference to the drawings. Note that in each drawing, identical or equivalent components and parts are given the same reference numerals. Also, the dimensional proportions in the drawings have been exaggerated for the sake of explanation and may differ from the actual proportions.

The growth prediction device of this embodiment provides specific improvements over conventional methods for predicting growth conditions for each type of rice, and represents an advancement in the technical field of predicting rice growth conditions.

The growth prediction device of this embodiment reflects the current growth status of rice plants and predicts growth status using LAI, an example of an index value that can be measured actually, and determines the timing of agricultural work from the predicted growth stage, providing support for agricultural work. In other words, by learning time-series measurement data of LAI and predicting future trends in LAI, it determines the timing approximately 10 days before heading date (when flag leaves appear) and predicts rice growth.

When performing growth prediction using LAI, the growth prediction device of this embodiment uses the maximum LAI value to normalize the LAI measurement data obtained by actual measurement to calculate normalized measurement data. This normalized measurement data is used for machine learning to generate a prediction model, which is then used to predict future trends in LAI. This makes it possible to train and perform predictions using the normalized LAI measurement data as training data, regardless of rice variety. This makes it possible to predict growth even for new varieties.

Figure 1 is a diagram showing an example of the configuration of the agricultural work support system 100 related to this embodiment.

The agricultural work support system 100 shown in FIG. 1 comprises a growth prediction device 10, an LAI measurement sensor 20, a database 30, an agricultural work instruction unit 40, and a user terminal 50. The growth prediction device 10 is connected to each of the LAI measurement sensor 20, the database 30, and the agricultural work instruction unit 40. The agricultural work instruction unit 40 is connected to the user terminal 50. The database 30 may be stored in an external storage device, or may be stored in the growth prediction device 10. Furthermore, the agricultural work instruction unit 40 may be realized as a function of an information processing device different from the growth prediction device 10, or may be realized as a function of the growth prediction device 10.

The LAI measurement sensor 20 is a sensor that measures LAI, an example of an index value that changes as rice grows, from images obtained by actually photographing rice plants. The LAI measurement sensor 20 is assumed to be capable of measuring weather (temperature, etc.), soil solution concentration, etc., and measuring LAI. LAI may be estimated using other methods, not just methods of estimating it from images. It is conceivable that the types of sensors installed in each field will vary from field to field. For example, in field A, water temperature, air temperature, and humidity are acquired by sensors, while in field B, water temperature, EC (electrical conductivity), air temperature, and humidity are acquired by sensors. In other words, the items that can be acquired from sensors may differ in each field, and items that cannot be acquired can be acquired from the database 30.

Database 30 stores cultivation information related to rice cultivation and environmental information related to the environment in which the rice is cultivated. Database 30 associates dates with the LAI measurement data and environmental information obtained from LAI measurement sensor 20, and stores the LAI measurement data and environmental information as time-series data. Furthermore, the time-series LAI measurement data (hereinafter simply referred to as "LAI measurement data") is stored in association with the cultivation information. The cultivation information includes multiple variables related to rice cultivation. The multiple variables related to rice cultivation include, for example, the transplanting date, variety, timing of top dressing for the relevant variety, and whether or not fertilization was performed (fertilization or no fertilization). Furthermore, the environmental information includes multiple variables related to the environment in which the rice is cultivated. The environmental information includes, for example, variables related to the weather during the rice cultivation period and variables related to the soil in which the rice is cultivated. Weather-related variables include temperature, solar radiation, humidity, etc., while soil-related variables include pH (hydrogen ion exponent), soil solution concentration (ion concentration), etc.

The growth prediction device 10 acquires LAI measurement data, cultivation information, and environmental information from the database 30. The growth prediction device 10 may acquire this information periodically or in response to a user request. The LAI measurement data may also be acquired directly from the LAI measurement sensor 20. The growth prediction device 10 determines the day on which LAI will peak and the maximum LAI value from the acquired LAI measurement data, cultivation information, and environmental information, and outputs these to the agricultural work instruction unit 40. The LAI peak day and maximum LAI value are used as data representing future growth conditions.

When the agricultural work instruction unit 40 obtains the date on which the LAI will peak and the maximum LAI value from the growth prediction device 10, it determines the need for top dressing and the timing of agricultural work based on the obtained date on which the LAI will peak and the maximum LAI value, and outputs the determination result to the user terminal 50.

The user terminal 50 is a terminal device used by the farmer (user), and may be, for example, a personal computer (PC), smartphone, etc. The user terminal 50 displays the judgment results from the farm work instruction unit 40 and instructs the user on the need for top dressing and the timing of farm work.

Next, referring to Figure 2, we will explain the hardware configuration of the growth prediction device 10 in this embodiment.

Figure 2 is a block diagram showing an example of the hardware configuration of the growth prediction device 10 in this embodiment.

As shown in Figure 2, the growth prediction device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, storage 14, an input unit 15, a display unit 16, and a communication interface (I/F) 17. Each component is connected to each other via a bus 18 so that they can communicate with each other.

The CPU 11 is a central processing unit that executes various programs and controls each component. That is, the CPU 11 reads programs from the ROM 12 or storage 14 and executes the programs using the RAM 13 as a working area. The CPU 11 controls the above components and performs various arithmetic processing in accordance with the programs stored in the ROM 12 or storage 14. In this embodiment, the ROM 12 or storage 14 stores a growth prediction program for executing growth prediction processing.

ROM 12 stores various programs and data. RAM 13 temporarily stores programs or data as a working area. Storage 14 is composed of an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various programs, including the operating system, and various data.

The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used to make various inputs to the device.

The display unit 16 is, for example, a liquid crystal display, and displays various information. The display unit 16 may also function as the input unit 15 by adopting a touch panel system.

The communication interface 17 is an interface that allows the device to communicate with other external devices. For this communication, a wired communication standard such as Ethernet (registered trademark) or FDDI (Fiber Distributed Data Interface) or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark) is used.

The growth prediction device 10 of this embodiment is implemented using a general-purpose computer device such as a server computer or a personal computer (PC).

Next, referring to Figure 3, the functional configuration of the growth prediction device 10 will be explained.

Figure 3 is a block diagram showing an example of the functional configuration of the growth prediction device 10 in this embodiment.

As shown in Figure 3, the growth prediction device 10 has, as its functional components, a data input unit 101, a maximum value estimation unit 102, a normalization unit 103, a variable selection unit 104, a learning unit 105, a prediction unit 106, and a data output unit 107. Each functional component is realized when the CPU 11 reads out a growth prediction program stored in the ROM 12 or storage 14, expands it into the RAM 13, and executes it.

Storage 14 stores a first trained model 141 and a second trained model 142 generated by the learning unit 105. The first trained model 141 is an example of a trained model, and is a predictive model generated by machine learning using previously obtained normalized measurement data of LAI as training data. The first trained model 141 takes the normalized measurement data of LAI as input and outputs prediction data representing the future trend of LAI. The second trained model 142 is an example of another trained model, and is a predictive model generated by machine learning using previously obtained normalized measurement data of LAI and a predetermined number of variables selected from environmental information in descending order of contribution rate to the trend of LAI as training data. The second trained model 142 takes the normalized measurement data of LAI and a predetermined number of selected variables as input and outputs prediction data representing the future trend of LAI. In addition, for the machine learning of each of the first trained model 141 and the second trained model 142, for example, LSTM (Long Shot-Term Memory) is used.

The data input unit 101 periodically or upon request from the user accepts input of LAI measurement data, cultivation information (transplanting date, variety, whether or not fertilization was performed, etc.), and environmental information (weather, soil solution concentration, etc.) from the database 30 (or LAI measurement sensor 20). Because LAI changes relatively slowly, the time-series LAI measurement data may be processed, for example, by averaging daily or thinning out every few days. Hereinafter, the LAI measurement data will be referred to as LAI measurement data LAI(t). The data input unit 101 converts the LAI measurement data LAI(t) into a format that can be input to each unit. The data input unit 101 outputs the LAI measurement data LAI(t), cultivation information, and environmental information to the maximum value estimation unit 102, and outputs the environmental information to the variable selection unit 104.

The maximum value estimation unit 102 estimates the maximum value of LAI from the LAI measurement data LAI(t), cultivation information, and environmental information acquired from the data input unit 101. Hereinafter, the maximum value of LAI will be referred to as the maximum LAI value LAI _MAX . Since there are trends such as fertilization > no fertilization and high temperature > low temperature, the maximum LAI value LAI _MAX can be calculated using the following formula (1):

LAI _MAX =α×β×LAI _ave ...(1)

where α is a coefficient determined by the presence or absence of fertilization, β is a coefficient determined by the transplanting date and temperature, and LAI _ave is the maximum average LAI value (approximately 5). The transplanting date may be expressed in three stages: early, middle, and late, depending on the planting time. LAI _ave is calculated from measurement data of LAI obtained from the past to the present.

The maximum value estimation unit 102 outputs the LAI measurement data LAI(t) and the maximum LAI value LAI _MAX estimated using the above equation (1) to the normalization unit 103. Note that the maximum value estimation unit 102 may have the correspondence relationship of the above equation (1) in the form of a table.

The normalization unit 103 calculates normalized measurement data by normalizing the LAI measurement data LAI(t) from the LAI measurement data LAI(t) and the maximum LAI value LAI _MAX acquired from the maximum value estimation unit 102. Hereinafter, this normalized measurement data will be referred to as normalized measurement data LAI _nom (t). The normalized measurement data LAI _nom (t) can be calculated using the following equation (2), that is, by dividing the LAI measurement value LAI(t) by the maximum LAI value LAI _MAX .

LAI _nom (t)=LAI(t)/LAI _MAX ...(2)

The normalization unit 103 outputs the LAI normalized measurement data LAI _nom (t) calculated using the above formula (2) to the variable selection unit 104 .

The variable selection unit 104 acquires environmental information from the data input unit 101 and acquires LAI normalized measurement data LAI _nom (t) from the normalization unit 103. The variable selection unit 104 selects a predetermined number of variables from all variables included in the environmental information in descending order of contribution rate to the transition of LAI. The variable selection unit 104 analyzes the relationship between multiple candidate explanatory variables (weather, soil solution concentration, etc.) and the objective variable LAI, selects variables with a high contribution rate to describing LAI, and outputs the selected variables and the LAI normalized measurement data LAI _nom (t) to the learning unit 105 or the prediction unit 106. Note that if the only variable to be used is LAI, the variable selection function is not used, and only the LAI normalized measurement data LAI _nom (t) is output to the learning unit 105 or the prediction unit 106.

For example, environmental information highly correlated with growth may differ for each field. Therefore, by appropriately selecting variables with high correlation (i.e., high contribution rates) from among the multiple variables contained in the environmental information, it is possible to expect improved prediction accuracy for LAI trends. Here, the variable selection process performed by the variable selection unit 104 will be specifically described with reference to Figures 4, 5, 6A, and 6B.

Figure 4 shows an example of multiple variables included in the environmental information in this embodiment.

As shown in Figure 4, environmental information includes multiple variables related to weather, such as maximum temperature, minimum temperature (lowest temperature), average temperature, emissivity, solar radiation, humidity, precipitation, water temperature, and soil temperature. Environmental information also includes multiple variables related to soil, such as EC representing electrical conductivity, pH representing hydrogen ion exponent, ORP representing oxidation-reduction potential, DO representing dissolved oxygen, NH4+ representing ammonium ions, NO3- representing nitrate ions, Na+ representing sodium ions, K+ representing potassium ions, and Ca2+ representing calcium ions. NDVI stands for normalized difference vegetation index (also known as normalized difference vegetation index).

As mentioned above, when predicting LAI, it is expected that prediction accuracy will be improved by appropriately selecting the environmental information (weather, soil solution concentration, etc.) to be used. Specifically, a regression analysis is performed using the LAI during the cultivation period as the dependent variable and the soil solution concentration (concentration of each ion) and meteorological data (temperature, solar radiation, humidity, etc.) as explanatory variables. From the coefficients obtained in the regression analysis (see Figure 4, for example), variables that have a high correlation with LAI, i.e., a high contribution rate, are selected. Specifically, the larger the coefficient of a variable, the higher its contribution rate to the progression of LAI. For this reason, a predetermined number of variables (for example, 1 to 5) are selected in descending order of coefficient.

Figure 5 is a graph showing an example of the relationship between actual LAI data A and predicted LAI data P in this embodiment. The vertical axis represents LAI, and the horizontal axis represents the number of days after transplantation.

In Figure 5, actual measurement data A shows the actual measured value of LAI, and predicted data P shows the predicted value of LAI predicted using environmental information selected according to the contribution rate to LAI.

Figure 6A is a graph showing an example of LAI when predicted using all environmental information. On the other hand, Figure 6B is a graph showing an example of LAI when predicted using environmental information selected according to the contribution rate to LAI. The vertical axis shows LAI. Note that the solid line shows the actual measured LAI data, and the dotted line shows the predicted LAI data.

The example in Figure 6B shows that it is easier to identify the position of the LAI peak compared to the example in Figure 6A. Hereinafter, the environmental information selected based on its contribution rate to the LAI will be referred to as "selected environmental information."

3 , the learning unit 105 performs a learning process for the prediction model. When the learning unit 105 acquires only the LAI normalized measurement data LAI _nom (t) from the variable selection unit 104, the learning unit 105 performs machine learning using the LAI normalized measurement data LAI _nom (t) as learning data to generate a first trained model 141. The first trained model 141 is a model that receives the LAI normalized measurement data LAI _nom (t) as input and outputs prediction data that represents a future change in the LAI.

On the other hand, when the learning unit 105 acquires the LAI normalized measurement data LAI _nom (t) and the selected environment information from the variable selection unit 104, it performs machine learning using the LAI normalized measurement data LAI _nom (t) and the selected environment information as learning data to generate a second trained model 142. The second trained model 142 is a model that receives the LAI normalized measurement data LAI _nom (t) and the selected environment information as input and outputs prediction data that represents a future trend of the LAI. Note that, as described above, for example, an LSTM or the like is used for the machine learning of each of the first trained model 141 and the second trained model 142.

The prediction unit 106 performs prediction processing using the prediction model learned by the learning unit 105. When the prediction unit 106 acquires only the LAI normalized measurement data LAI _nom (t) from the variable selection unit 104, the prediction unit 106 predicts a future change in the LAI from the LAI normalized measurement data LAI _nom (t) using the first trained model 141. Naturally, the LAI normalized measurement data LAI _nom (t) acquired by the prediction unit 106 is data that is temporally later than the LAI normalized measurement data LAI _nom (t) acquired by the learning unit 105. In other words, the LAI normalized measurement data LAI _nom (t) used for prediction by the prediction unit 106 is data that was acquired after the first trained model 141 was generated.

On the other hand, when the prediction unit 106 acquires the LAI normalized measurement data LAI _nom (t) and the selected environment information from the variable selection unit 104, it predicts a future change in the LAI from the LAI normalized measurement data LAI _nom (t) and the selected environment information using the second trained model 142. As described above, the LAI normalized measurement data LAI _nom (t) and the selected environment information acquired by the prediction unit 106 are data that are temporally later than the LAI normalized measurement data LAI _nom (t) and the selected environment information acquired by the learning unit 105. In other words, the LAI normalized measurement data LAI _nom (t) and the selected environment information used for prediction by the prediction unit 106 are data that were acquired after the second trained model 142 was generated.

The prediction unit 106 may also predict future trends in the LAI using either the first trained model 141 or the second trained model 142, whichever has the higher LAI prediction accuracy. The prediction accuracy of the model may be determined using a method that determines the difference between the actual measured value and the predicted value, such as RMSE (Root Mean Squared Error), or other methods.

That is, the prediction unit 106 predicts the transition of the LAI for the next unit time. For the prediction, for example, a method such as the above-mentioned LSTM is used. When predicting the transition of the LAI, the following variations are assumed to be available for the selection of variables to be used.
(1) Prediction is performed using only the LAI.
(2) Prediction is made using the LAI and selected environmental information.
(3) Predictions are made in both (1) and (2) above, and the one with the better prediction accuracy is selected.

The prediction unit 106 outputs predicted data predicting the future trend of LAI to the data output unit 107.

The data output unit 107 identifies the day on which the LAI will peak from the prediction data acquired from the prediction unit 106, and outputs the day on which the LAI will peak and the maximum value of the LAI to the agricultural work instruction unit 40. The growth prediction device 10 may output this information periodically or in response to a request from the user.

Returning to Figure 1, the agricultural work instruction unit 40 obtains the LAI peak date and maximum LAI value from the growth prediction device 10, determines the need for top dressing and the timing of agricultural work from the obtained LAI peak date and maximum LAI value, and outputs the determination result to the user terminal 50. Specifically, for example, the recommended timing of agricultural work for the variety being cultivated is obtained from the cultivation information stored in the database 30. The following items are estimated from the LAI peak date, maximum LAI value, and the current date, and instructed to the user:

- Whether or not additional fertilization is necessary (determined by the maximum LAI value)
- Top dressing time (day when LAI peaks)
- Pesticide application date (approximately 3 days after the peak of LAI)
・Optimal harvest time (approximately one and a half months after the peak of LAI)

The need for top dressing is determined by whether the maximum LAI value is equal to or greater than a threshold value. The time for top dressing (top dressing timing) is determined, for example, as a period of X days before or after the day when LAI peaks. The optimal date for X may differ depending on the variety. For example, for variety A, it is 15 days before the heading date (≒ 5 days before the LAI peak), and for variety B, it is 12 days before the heading date (≒ 2 days before the LAI peak). Therefore, when the top dressing time for each variety is based on the day when LAI peaks, it is desirable to set it to Y days before the heading date minus 10 days (Y-10). In this case, information on the time for top dressing for each variety can be stored in database 30 as cultivation information.

Next, referring to Figures 7 and 8, the operation of the growth prediction device 10 of this embodiment will be explained.

Figure 7 is a flowchart showing an example of the flow of the learning process using the growth prediction program according to this embodiment. The learning process using the growth prediction program is realized by the CPU 11 of the growth prediction device 10 writing the growth prediction program stored in the ROM 12 or storage 14 to the RAM 13 and executing it.

In step S101 of Figure 7, the CPU 11 periodically or in response to a user request, acquires LAI measurement data LAI(t), cultivation information (transplanting date, variety, whether or not fertilization was performed, etc.), and environmental information (weather, soil solution concentration, etc.) from the database 30.

In step S102, the CPU 11 estimates the maximum LAI value LAI _MAX from the LAI measurement data LAI(t), cultivation information, and environmental information acquired in step S101, using, for example, the above-mentioned formula (1).

In step S103, the CPU 11 normalizes the LAI measurement data LAI(t) from the LAI measurement data LAI(t) acquired in step S101 and the maximum LAI value LAI _MAX estimated in step S102, using, for example, the above-mentioned formula (2), to calculate normalized measurement data LAI _nom (t).

In step S104, the CPU 11 selects as selected environmental information, as an example, a predetermined number of variables from all variables included in the environmental information acquired in step S101, in order of their highest contribution rate to the change in LAI, as shown in Figure 4 above.

In step S105, the CPU 11 generates a second trained model 142 by machine learning the LAI-normalized measurement data LAI _nom (t) normalized in step S103 and the selected environment information selected in step S104 as training data. The second trained model 142 is a model that receives the LAI-normalized measurement data LAI _nom (t) and the selected environment information as input and outputs prediction data representing a future change in the LAI. Note that when the CPU 11 acquires only the LAI-normalized measurement data LAI _nom (t), it generates a first trained model 141 by machine learning the LAI-normalized measurement data LAI _nom (t) as training data. The first trained model 141 is a model that receives the LAI-normalized measurement data LAI _nom (t) as input and outputs prediction data representing a future change in the LAI.

In step S106, the CPU 11 stores the first trained model 141 and the second trained model 142 generated in step S105 in the storage 14, and terminates the learning process using this growth prediction program.

Figure 8 is a flowchart showing an example of the flow of prediction processing using the growth prediction program according to this embodiment. The prediction processing using the growth prediction program is realized by the CPU 11 of the growth prediction device 10 writing the growth prediction program stored in the ROM 12 or storage 14 to the RAM 13 and executing it.

In step S111 of Figure 8, the CPU 11 periodically or in response to a user request, obtains LAI measurement data LAI(t), cultivation information (transplanting date, variety, whether or not fertilization was performed, etc.), and environmental information (weather, soil solution concentration, etc.) from the database 30.

In step S112, the CPU 11 estimates the maximum LAI value LAI _MAX from the LAI measurement data LAI(t), cultivation information, and environmental information acquired in step S111, using, for example, the above-mentioned formula (1).

In step S113, the CPU 11 normalizes the LAI measurement data LAI(t) from the LAI measurement data LAI(t) acquired in step S111 and the maximum LAI value LAI _MAX estimated in step S112, using, for example, the above-mentioned formula (2), to calculate normalized measurement data LAI _nom (t).

In step S114, the CPU 11 selects as selected environmental information, as an example, a predetermined number of variables from all variables included in the environmental information acquired in step S111, in order of their highest contribution rate to the change in LAI, as shown in Figure 4 above.

In step S115, the CPU 11 inputs the LAI normalized measurement data LAI _nom (t) normalized in step S113 and the selected environment information selected in step S114 into the second trained model 142, and causes the second trained model 142 to output predicted data representing a future change in the LAI. Note that, when the CPU 11 acquires only the LAI normalized measurement data LAI _nom (t), the CPU 11 inputs the LAI normalized measurement data LAI _nom (t) normalized in step S113 into the first trained model 141, and causes the first trained model 141 to output predicted data representing a future change in the LAI. Furthermore, the CPU 11 may predict a future change in the LAI using either the first trained model 141 or the second trained model 142, whichever model has a higher LAI prediction accuracy.

In step S116, the CPU 11 identifies the day on which the LAI will peak from the predicted data obtained in step S115, outputs the day on which the LAI will peak and the maximum value of the LAI to the agricultural work instruction unit 40 as future growth conditions, and terminates the prediction processing by this growth prediction program.

In this way, according to this embodiment, by predicting LAI, it is possible to predict the growth status (the timing of flag leaf emergence and the approximate heading date).

In addition, by using the measured LAI, it is possible to understand the current growth situation, which makes it possible to respond to growth delays caused by human factors.

In addition, by normalizing the LAI measurement data, it is possible to use the LAI measurement data for learning without distinguishing between varieties. This eliminates the need to build a model for each variety, making it possible to predict the growth of new varieties as well.

In the above embodiment, the growth prediction process executed by CPU 11 after reading the growth prediction program may be executed by various processors other than CPU 11. Examples of processors in this case include PLDs (Programmable Logic Devices) such as FPGAs (Field-Programmable Gate Arrays), whose circuit configuration can be changed after manufacture, and dedicated electrical circuits, such as ASICs (Application Specific Integrated Circuits), which are processors with circuit configurations specifically designed to execute specific processes. Furthermore, the growth prediction process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor devices.

In addition, while the above embodiment describes a configuration in which the growth prediction program is pre-stored (also referred to as "installed") in ROM 12 or storage 14, this is not limited to this. The growth prediction program may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The growth prediction program may also be downloaded from an external device via a network.

All publications, patent applications, and technical standards mentioned in this specification are incorporated by reference into this specification to the same extent as if each individual publication, patent application, and technical standard was specifically and individually indicated to be incorporated by reference.

The following notes are further disclosed regarding the above embodiments.

(Additional note 1)
Memory and
at least one processor coupled to said memory;
Including,
The processor:
The system accepts input of time-series measurement data of index values that change with the growth of rice, cultivation information on rice cultivation, and environmental information on the environment in which rice is cultivated,
estimating a maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information;
Normalizing the measurement data of the index values from the measurement data of the index values and the maximum value of the index values to calculate normalized measurement data;
A growth prediction device configured to predict future trends in index values from the calculated normalized measurement data of the index values using a trained model that is generated by machine learning using normalized measurement data of the index values obtained in advance as learning data, and that takes the normalized measurement data of the index values as input and outputs prediction data that represents future trends in the index values.

(Additional note 2)
A non-transitory storage medium storing a program executable by a computer to perform growth prediction processing,
The growth prediction process includes:
The system accepts input of time-series measurement data of index values that change with the growth of rice, cultivation information on rice cultivation, and environmental information on the environment in which rice is cultivated,
estimating a maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information;
Normalizing the measurement data of the index values from the measurement data of the index values and the maximum value of the index values to calculate normalized measurement data;
A non-transitory storage medium that predicts future trends in an index value from the calculated normalized measurement data of the index value using a trained model that is generated by machine learning using normalized measurement data of the index value obtained in advance as training data, and that uses the normalized measurement data of the index value as input and outputs prediction data that represents future trends in the index value.

10 Growth prediction device 11 CPU
12 ROM
13 RAM
14 Storage 15 Input unit 16 Display unit 17 Communication I/F
18 Bus 20 LAI measurement sensor 30 Database 40 Farm work instruction unit 50 User terminal 100 Farm work support system 101 Data input unit 102 Maximum value estimation unit 103 Normalization unit 104 Variable selection unit 105 Learning unit 106 Prediction unit 107 Data output unit 141 First trained model 142 Second trained model

Claims (8)

a data input unit that receives input of time-series measurement data of index values that change with the growth of rice, cultivation information related to rice cultivation, and environmental information related to the environment in which rice is cultivated;
a maximum value estimation unit that estimates a maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information;
a normalization unit that normalizes the measurement data of the index values from the measurement data of the index values and the maximum value of the index values to calculate normalized measurement data;
a prediction unit that predicts future trends in the index values from the normalized measurement data of the index values calculated by the normalization unit, using a trained model that is generated by machine learning using normalized measurement data of the index values obtained in advance as training data, and that receives the normalized measurement data of the index values as input and outputs prediction data that represents future trends in the index values;
A growth prediction device equipped with the above. The cultivation information includes a transplanting date indicating the date on which the rice was transplanted and whether or not fertilization was performed indicating whether or not fertilizer was applied,
the environmental information includes temperature during the rice cultivation period;
The growth prediction device according to claim 1 , wherein the maximum value estimation unit estimates the maximum value of the index value from a coefficient determined by the presence or absence of fertilization, a coefficient determined by the transplanting date and the temperature, and an average maximum value of the index value. the environmental information includes a plurality of variables related to an environment in which rice is grown;
a variable selection unit that selects a predetermined number of variables from all variables included in the environmental information in descending order of contribution rate to the transition of the index value;
3. The growth prediction device according to claim 1, wherein the prediction unit predicts a future change in the index value from the normalized measurement data of the index value calculated by the normalization unit and the predetermined number of variables selected by the variable selection unit using another trained model instead of the trained model, the another trained model being generated by machine learning using, as learning data, normalized measurement data of the index value obtained in advance and a predetermined number of variables selected from the environmental information in descending order of contribution rate to the change in the index value, the another trained model receiving the normalized measurement data of the index value and the selected predetermined number of variables as input and outputting prediction data representing a future change in the index value. The growth prediction device according to claim 3 , wherein the prediction unit predicts a future change in the index value using one of the trained model and the other trained model, which has a higher prediction accuracy of the index value. The growth prediction device according to claim 3 or 4, wherein the environmental information includes variables related to weather during a rice cultivation period and variables related to soil in which the rice is cultivated. The growth prediction device according to any one of claims 1 to 5, wherein the index value is a leaf area index that represents the ratio of the area of rice leaves to a unit area of the ground surface. The system accepts input of time-series measurement data of index values that change with the growth of rice, cultivation information on rice cultivation, and environmental information on the environment in which rice is cultivated,
estimating a maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information;
Normalizing the measurement data of the index values from the measurement data of the index values and the maximum value of the index values to calculate normalized measurement data;
a trained model that is generated by machine learning normalized measurement data of the index value obtained in advance as learning data, and that receives the normalized measurement data of the index value as input and outputs prediction data that represents a future change in the index value, and predicts a future change in the index value from the calculated normalized measurement data of the index value;
A computer-implemented growth prediction method. The system accepts input of time-series measurement data of index values that change with the growth of rice, cultivation information on rice cultivation, and environmental information on the environment in which rice is cultivated,
estimating a maximum value of the index value from the measurement data of the index value, the cultivation information, and the environmental information;
Normalizing the measurement data of the index values from the measurement data of the index values and the maximum value of the index values to calculate normalized measurement data;
a trained model that is generated by machine learning normalized measurement data of the index value obtained in advance as learning data, and that receives the normalized measurement data of the index value as input and outputs prediction data that represents a future change in the index value, and predicts a future change in the index value from the calculated normalized measurement data of the index value;
A growth prediction program to be run on a computer.