JP7343289B2 - Air conditioning load prediction method and air conditioning system - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act Published in the Proceedings of the 2018 Society of Air Conditioning and Sanitary Engineers Conference (pp. 221-224)

The present invention relates to a method for predicting air conditioning load, and an air conditioning system to which the same is applied.

FIG. 3 schematically shows an example of a general air conditioning system. Conditioned air A1 sent out from the air conditioner 1 is guided to a target space S, which is an indoor room such as an office, through an air supply duct 2. A plurality of air outlets 4 are installed in the ceiling 3 of the target space S, and the air supply duct 2 extending from the air conditioner 1 branches in the middle and is connected to each air outlet 4 on the downstream side. In this way, the conditioned air A1 flowing through the air supply duct 2 is supplied to the target space S from each outlet 4.

A variable air volume unit 5 is provided at a position in front of each outlet 4 in the air supply duct 2. The variable air volume unit 5 is a device that adjusts the volume of air passing through the interior by changing the opening degree of a damper provided therein, and the conditioned air A1 has its air volume adjusted by the variable air volume unit 5. is supplied to the target space S.

A return air port 6 is provided at an appropriate position within the target space S, and the return air port 6 and the air conditioner 1 are connected by a return air duct 7. Indoor air A2 in the target space S is taken in from the return air port 6 as return air A3, returns to the air conditioner 1 through the return air duct 7, and is sent out again as conditioned air A1 after its temperature and humidity are adjusted. It will be done. In this way, air is circulated between the air conditioner 1 and the target space S.

In such an air conditioning system, for example, an air conditioning method called a variable air volume method is used to cope with fluctuations in the air conditioning load within the target space S. A temperature sensor 8 that measures the temperature of the indoor air A2 is provided at a suitable location within the target space S (in the example shown here, the ceiling 3). The control unit 5a provided in each variable air volume unit 5 adjusts the damper opening degree according to the measured value acquired by each temperature sensor 8 and the set value of the air temperature in the target space S, and adjusts the damper opening degree to an appropriate amount. Conditioned air A1 is supplied from the outlet 4. That is, for example, during cooling, if the measured value of the temperature sensor 8 is significantly higher than the set value of the indoor temperature (temperature of the indoor air A2), the supply amount of the conditioned air A1 is increased so that the set value and the measured value are If the amount is close, control is performed to reduce the supply amount.

Prior art documents related to this type of air conditioning system include, for example, Patent Document 1 below.

In addition, in FIG. 3, for convenience of explanation, one air conditioner 1, one target space S, and two air outlets 4 are simply illustrated, but in an actual air conditioning system, multiple air conditioners 1 It may also be configured such that the conditioned air A1 is introduced into a plurality of target spaces S. The number of air outlets 4 installed per target space S can also be changed as appropriate depending on the size of the target space S and the like. In addition, in an actual air conditioning system, various devices and sensors are installed in addition to the devices shown here, but illustrations of configurations that are not directly related to the purpose of the present invention are omitted as appropriate. .

In addition to adjusting the blowout air volume in the variable air volume unit 5 described above, methods that can respond to fluctuations in the air conditioning load in the target space S include turning on and off each of the plurality of air conditioners 1 individually, Various methods can be adopted, such as adjusting the supply amount and supply temperature of A1.

Japanese Patent Application Publication No. 2015-190669

Incidentally, in the air conditioning system as described above, fluctuations in the air conditioning load in the target space S are grasped after the fact, and the operating conditions are adjusted in a feedback manner based on this. In other words, when the air conditioning load increases due to movement such as, for example, some new equipment operating in the target space S or a new person entering the target space S, the temperature of the indoor air A2 rises, and the temperature of the indoor air A2 increases. The temperature sensor 8 installed in the space S detects the temperature, and then countermeasures are taken, such as increasing the amount of air blown from the variable air volume unit 5. In other words, there is a time lag between the change in air conditioning load and the response to the change, so there is a lag between the load status and the operating status from when the air conditioning load fluctuates until the time when the changed air conditioning load is properly responded to. It will happen. Such a deviation in the operating condition with respect to the load condition may not only result in an air-conditioning condition that is not comfortable for people in the target space S, but also has the problem of being inefficient in terms of energy consumption.

In view of these circumstances, the present invention provides an air conditioning load prediction method and an air conditioning system that can quickly respond to changes in air conditioning load.

The present invention is an air conditioning load prediction method that predicts an air conditioning load using a load prediction model generated by machine learning based on a plurality of parameters related to the air conditioning load, comprising:
The load prediction model is an ARIMAX model that incorporates a plurality of parameters related to the air conditioning load as explanatory variables into the regression term, and uses the actual value of the air conditioning load at a point in the past from the target point in the ARIMA term,
Using a plurality of parameters selected from day of week, time of day, and outside temperature as explanatory variables of the regression term,
The method of predicting air conditioning load is characterized in that the calculation load is reduced by using an interaction term based on a combination of the plurality of parameters in the regression term .

In the air conditioning load prediction method of the present invention, at least one interaction term of day of the week and outside temperature, time of day and outside temperature, and day of the week, time of day, and outside temperature can be used as the regression term.

In the air conditioning load prediction method of the present invention, the load prediction model can use a time dummy as an explanatory variable. Further, in the air conditioning load prediction method of the present invention, the load prediction model can set the setting interval of the hour and minute dummy to 15 minutes or more and 30 minutes or less.

Further, the present invention relates to an air conditioning system characterized in that it is configured to adjust the operating status based on the predicted value of the air conditioning load by the above-mentioned air conditioning load prediction method.

The air conditioning system of the present invention includes an air conditioner that sends out conditioned air, an air duct that guides the conditioned air from the air conditioner to a target space, an outlet that blows out the conditioned air to the target space, and an air conditioner that blows out the air conditioner from the outlet. and a variable air volume unit that adjusts the air volume, and may be configured to adjust the air volume of the conditioned air supplied from the variable air volume unit to the target space based on the predicted value of the air conditioning load.

According to the air conditioning load prediction method and air conditioning system of the present invention, it is possible to achieve the excellent effect of being able to quickly respond to changes in air conditioning load.

1 is a schematic diagram showing an example of the configuration of an information system in an air conditioning system to which the present invention is applied. It is a flowchart explaining an example of the procedure of the prediction method of air conditioning load by implementation of the present invention. FIG. 1 is a schematic diagram showing an example of the configuration of an air conditioning system.

Embodiments of the present invention will be described below with reference to the accompanying drawings. Note that the configurations of the air conditioner 1, the target space S, etc. are the same as those of the conventional example shown in FIG. 3, so the description will be made below with reference to FIG. 3 as necessary.

As shown in FIG. 3, conditioned air A1 sent out from the air conditioner 1 is guided to a target space S, which is an indoor room such as an office, through an air duct 2, and is supplied from a plurality of air outlets 4 installed in the ceiling 3. be done. A variable air volume unit 5 is provided at a position in front of each outlet 4, and the conditioned air A1 is supplied to the target space S after the air volume is adjusted by the variable air volume unit 5. Indoor air A2 is taken in as return air A3 from the return air port 6, passes through the return air duct 7, and is returned to the air conditioner 1 again. A temperature sensor 8 is provided at an appropriate position (ceiling 3) in the target space S to measure the temperature of the indoor air A2.

The air conditioning system of this embodiment predicts the air conditioning load in the target space S, and based on the predicted value, the operating status (in the case of this embodiment, in particular, the air volume of the conditioned air A1 supplied from the variable air volume unit 5 to the target space S). ) can be adjusted in a feedforward manner.

Figure 1 shows the configuration of an information system that enables air conditioning load prediction and control based on the prediction. Various devices as shown in FIG. 3, ie, devices such as the air conditioner 1 and the variable air volume unit 5, are informationally connected to the network 10 (first network). First network 10 is connected to second network 16 via router 14 .

The first network 10 is a local network constructed by the air conditioner 1 and equipment around the target space S, and the second network 16 is a virtual network constructed, for example, on a cloud. In this way, various types of information can be exchanged between the devices connected to the first network 10 and the second network 16.

In the second network 16, a data storage section 17, a model generation section 18, a load prediction section 19, an air volume calculation section 20, and an input/output section 21 are virtually installed (note that one of these sections Of course, it is also possible to install part or all of it as physical hardware).

The data storage unit 17 stores information acquired from devices connected to the first network 10, such as the air conditioner 1, variable air volume unit 5, and temperature sensor 8, as well as various other information as a database. In the case of this embodiment, the data acquired by the temperature sensor 8 is stored in the data storage unit 17 from the control unit 5a (see FIG. 3) of the variable air volume unit 5 via the first and second networks 16. . The model generation unit 18 performs a load prediction that predicts the air conditioning load in the target space S at a certain point in time (for example, at the present time or at a point in the immediate future of the present time) based on the learning data stored in the data storage unit 17. Generate model M. The load prediction unit 19 stores the load prediction model M generated by the model generation unit 18, and predicts the air conditioning load in the target space S at a certain point in time based on various parameters. The air volume calculation unit 20 calculates the set air volume in the variable air volume unit 5 based on the air conditioning load predicted by the load prediction unit 19. The control unit 5a of the variable air volume unit 5 adjusts the damper opening degree based on the air volume setting value calculated by the air volume calculation unit 20. The input/output unit 21 writes information acquired from each device connected to the first network 10 to the data storage unit 17, takes out necessary data from the data storage unit 17 and inputs it to the load prediction unit 19, Alternatively, information is input/output between each section, such as inputting the air volume setting value calculated by the air volume calculation section 20 to the control section 5a of the variable air volume unit 5.

The model generation unit 18 has a function of generating a load prediction model M by machine learning based on learning data. As the load prediction model M, various types of models such as linear regression, ridge regression, gradient boosting, and random forest can be adopted.

A method for predicting air conditioning load using such a system will be described with reference to the flowchart in FIG. 2.

First, learning data for generating the load prediction model M is created and input to the data storage section 17 (step S1). The learning data is a data set that records various parameters related to the air conditioning load at various times and actual values of the air conditioning load at those times when the air conditioning system shown in FIG. 3 is actually operated. The actual value of the air conditioning load is the "temperature difference between the conditioned air A1 blown out from the variable air volume unit 5 and the return air A3 taken in from the return air port 6" and the "air volume blown out of the conditioned air A1 in the variable air volume unit 5". It can be calculated as the product multiplied by Further, some of the parameters related to the air conditioning load can be acquired from the air conditioner 1, the temperature sensor 8, and the control unit 5a of the variable air volume unit 5.

In predicting the air conditioning load, various parameters related to the air conditioning load can be used as explanatory variables. Examples of parameters related to air conditioning load are listed below.
- Measured value of indoor temperature (air temperature in target space S) - Set value of indoor temperature - Temperature difference between indoor and outdoor - Outdoor temperature - Amount of solar radiation - Date (at least one of month or day)
・Time (at least one of hours or minutes)
・Day of the week (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, and Saturday, or holiday or weekday)
・Actual value of air conditioning load ・Number of operating air conditioners 1 ・Supply air temperature of air conditioner 1 (temperature of conditioned air A1)
・Weather/expected temperature

Indoor temperature is a parameter directly related to the value of air conditioning load. The outside air temperature and the amount of solar radiation affect the flow of heat into and out of the target space S, and affect the air conditioning load. The date and time are based on changes in outside air temperature depending on the season or time, or changes in the amount of heat generated in the target space S (for example, it is assumed that there is equipment in the target space S that operates only in a specific season or time). It corresponds to the air conditioning load, so it is related to the air conditioning load. The day of the week corresponds to the schedule of people's activities and the operation of equipment in the target space S, so it is related to the air conditioning load (for example, in the case of an office, there are fewer people on holidays and equipment is not operated as much, so it is a holiday) It is assumed that the air conditioning load will be lower on days of the week). In addition, the actual value of the air conditioning load in the last few days or one year before the target time, the number of operating air conditioners 1, the supply air temperature, etc. are parameters related to the value of the air conditioning load at the target time, It can be used for prediction. The weather and expected temperature at the desired time also affect the air conditioning load.

When it is desired to predict the air conditioning load at a desired point in time, the time point or period at which parameters used as explanatory variables for the prediction are acquired can be selected as appropriate. For example, if you want to predict the air conditioning load 15 minutes from now, you can use the indoor temperature, outdoor temperature, and solar radiation at multiple points in the past week as explanatory variables, or you can use Each parameter at the same date and time may be used as an explanatory variable, or each parameter at the current time may be used as an explanatory variable. Of the above parameters, indoor temperature, outdoor temperature, solar radiation, actual value of air conditioning load, number of operating air conditioners 1, supply air temperature, and expected temperature are qualitative variables, and date, time, day of the week, and weather are dummy variables. Each is set as a variable.

When predicting the air conditioning load, some or all of the parameters listed above can be used as explanatory variables. Further, in place of or in addition to the above parameters, some other parameter related to the air conditioning load may be used as appropriate.

In the learning data, the actual values of the air conditioning load at the time are recorded together with each of the parameters at a plurality of time points. The learning data may be obtained by, for example, operating the air conditioning system by feedback control based on the temperature sensor 8 for a period sufficient to construct the load prediction model M (for example, from several days to one year), and collecting the above-mentioned parameters during that period. It can be created with .

The model generation unit 18 performs machine learning using the learning data and generates a load prediction model M that predicts the air conditioning load at a target time based on various parameters (step S2). The generated load prediction model M is a model that predicts the air conditioning load at a desired point in time in the target air conditioning system from parameters related to operating conditions at a certain point or period.

As described above, various types of models can be employed as the load prediction model M generated in step S2. However, when the purpose is to predict the air conditioning load, it is preferable to use the ARIMAX model as the load prediction model M in terms of accuracy and convenience.

The ARIMAX model will be explained. In order to predict the target value (here, air conditioning load) at a certain point in time, time-series data at a point before the target point in time can be used. For example, in order to calculate the predicted value of the air conditioning load 15 minutes later, it is possible to use the actual values of the air conditioning load acquired in chronological order from one day before to the current time.

A model that predicts the air conditioning load at a target time using values of the air conditioning load at each time past the target time is called an AR (autoregressive) model, and is expressed, for example, by a formula such as Equation 1 below. y _t is the predicted value at the target point, and y _t-1 , y _t-2 , ......, y _t-p are the predicted values at each point in the past (from 1 point before to before point p) The actual value in is substituted.
[Formula 1]
y _t =φ ₁ y _t-1 +φ ₂ y _t-2 +......+φ _p y _t-p +ε _t

Additionally, time series data usually contains errors. Therefore, considering that data at each point in the past includes errors, a model that smoothes and handles the movement along the time series is called an MA (moving average) model. The MA model is expressed, for example, by a mathematical formula such as Equation 2 below.
[Formula 2]
y _t =θ ₁ ε _t-1 +θ ₂ ε _t-2 +...+θ _q ε _t-q +ε _t

A model that combines the AR model and MA model is called an ARMA (autoregressive moving average) model, and is expressed by a combination of Equations 1 and 2, as in Equation 3 below, for example.
[Formula 3]
y _t =φ ₁ y _t-1 +φ ₂ y _t-2 +......+φ _p y _t-p +θ ₁ ε _t-1 +θ ₂ ε _t-2 +...+θ _q ε _t-q +ε _t

Furthermore, if the target value fluctuates over time, more accurate predictions can be made by considering not only the value itself at each point in time, but also the movement of that value over time. Therefore, a model in which the above-mentioned ARMA model is applied to time-series data obtained by taking the difference between each point in time is called an ARIMA (autoregressive integrated moving average) model. The ARIMA model is expressed, for example, by a mathematical formula such as Equation 4 below.
[Formula 4]
Δ ^d y _t =φ ₁ Δ ^d y _t-1 +φ ₂ Δ ^d y _t-2 +......+φ _p Δ ^d y _t-p +θ ₁ ε _t-1 +θ ₂ ε _t-2 +. .....＋θ _q ε _t−q +ε _t

The ARIMA model is designed to reflect the movement of values at each time point from 1 time point before to p time points ago, and calculate the value at the target time point by taking into account errors. Here, the value at the target point in time may be influenced by another parameter other than the value in a certain period in the past or the movement of that value. For example, in the case of air conditioning load, the value at each point in time and its fluctuation pattern vary greatly depending on the day of the week and time of day. That is, for example, when considering air conditioning in an office or the like, the load is large during the daytime on weekdays, and the load is small at night and on holidays. For this reason, for example, when trying to predict the air conditioning load at 18:00 based on data from 9:00 to 17:30 on weekdays, or predicting the air conditioning load on Saturday based on data from Monday to Friday, the air conditioning load or Predictions based only on these fluctuations may not be able to achieve sufficient accuracy.

Therefore, the ARIMAX model is a model that combines the above-mentioned ARIMA model with a regression model using another parameter as an explanatory variable, and is expressed, for example, by a formula such as Equation 5 below. βx on the right side is a regression term representing the influence of another parameter (regression model).
[Formula 5]
Δ ^d y _t = βx+φ ₁ Δ ^d y _t-1 +φ ₂ Δ ^d y _t-2 +......+φ _p Δ ^d y _t-p +θ ₁ ε _t-1 +θ ₂ ε _t-2 +. .....＋θ _q ε _t−q +ε _t

Parameter settings when adopting the ARIMAX model as the load prediction model M will be explained.

The ARIMA term (after the second item on the right side of Equation 5 above) uses the actual value of the air conditioning load at each point in the past period from the target point in time (for example, each point in time from several days to one month ago). It will be done. More specifically, the actual values from one time point before the target time point to time point p are substituted into y _t-1 , y _t-2 , . . . , y _t-p .

Various parameters related to the air conditioning load can be incorporated into the regression term (the part corresponding to βx on the right side of Equation 5 above), but in particular, multiple parameters selected from the following three parameters can be incorporated as explanatory variables. It is suitable to use.
・Day of the week (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, and Saturday, or holiday or weekday)
・Time (hour or hour/minute)
·Outside air temperature

These parameters are easy to obtain among parameters related to air conditioning load. Dummy variables can be set for the day of the week and time, and if outside temperature is used as an explanatory variable, past actual measurements, current actual measurements obtained by an outdoor temperature sensor (not shown), or current forecasts from weather forecasts can be used. It is sufficient to use values etc. Therefore, for example, it is possible to predict the air conditioning load without purchasing detailed weather forecast data, which is convenient and can reduce costs.

Furthermore, when using the above parameters as explanatory variables in a regression term, it is recommended to create a regression term using the interaction terms of "day of the week and outside temperature," "time of day and outside temperature," and "day of the week, time of day, and outside temperature." . If these three interaction terms are used as regression terms, we can obtain "the relationship between outside air temperature and air conditioning load on holidays," "the relationship between outside air temperature and air conditioning load at a certain time," and "the relationship between outside air temperature and air conditioning load at a certain time on weekdays." This pattern can be reflected in the predicted value. In addition, since the number of terms is small, the calculation load is reduced, and overfitting is suppressed, making it possible to make highly accurate predictions.

When using hours and minutes as dummy variables, for example, a term (time dummy) representing m hours (0 o'clock, 1 o'clock, 2 o'clock, ......, 23 o'clock) and n minutes (0 minutes, 1 o'clock), (minutes, 2 minutes, ......, 59 minutes, or 0 minutes, 5 minutes, 10 minutes, 15 minutes, ..., 55 minutes, etc.) (minute dummy) as variables. There are two possible methods: setting m hours and n minutes (hour and minute dummy) as a variable. As a result of extensive research, the inventors of the present application have found that when the latter method is adopted, it is possible to create a load prediction model M that can predict the air conditioning load with particularly high accuracy.

Here, when employing hour and minute dummies, it goes without saying that the finer the minutes, the more variables there are for the hour and minute dummies. That is, 144 time and minute dummy terms are required for 10 minute increments, 288 for 5 minute increments, and 1440 for 1 minute increments. If the number of variables is large, the time required to predict the air conditioning load using the load prediction model M becomes long. On the other hand, it is expected that the finer the minute increments, the more accurately the air conditioning load can be predicted. Furthermore, since the interval at which the air conditioning load can be predicted becomes shorter, it becomes possible to operate the system in a manner that more precisely responds to the air conditioning load that changes over time.

Therefore, the inventors of the present application studied the relationship between the interval at which hour and minute dummies are set and the time required for predictive calculation (processing time). The results are shown in Table 1 below.

When the set interval was 10 minutes or less, the processing time exceeded the set interval. As mentioned above, if the setting interval is short, improved prediction accuracy and detailed operation control can be expected, but if you want to use the predicted air conditioning load for air conditioning system operation, if the processing time is longer than the hour/minute dummy setting interval, the acquisition By the time a predicted result is obtained from the parameters, the target time has already passed. Therefore, when employing hour and minute dummies, it is preferable that the setting interval of the hour and minute dummies is 15 minutes or more. On the other hand, from the viewpoint of prediction accuracy and followability of the operating status to the air conditioning load, it is preferable that the setting interval of the hour and minute dummy is not too long. From the above, it is preferable that the setting interval of the hour and minute dummy is between 15 minutes and 30 minutes (although processing time depends on the performance of the computer, so it cannot be generalized). It is considered that the air conditioning load can be accurately predicted at sufficient time intervals.

In addition, regarding the case where the day of the week is used as an explanatory variable, the case where the dummy variable is set in two ways, holiday or weekday, has been described above, but it is also possible to set it in seven ways: Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, and Saturday. However, according to research conducted by the inventors of the present application, it has been found that prediction accuracy is higher when the day of the week dummy variable is set to either holiday or weekday. If there were seven dummy variables for the days of the week, there was a possibility that overfitting would occur as a result of too many cases. By setting the dummy variable for the day of the week in two ways: holiday or weekday, it is possible to further improve prediction accuracy while reducing the amount of calculation.

In addition, if it is desired to make a prediction that corresponds more precisely to the schedule of an office or the like, "whether it is the day of the week when air conditioning starts or not" or "whether it is air conditioning start time or not" may be added as a dummy variable. For example, on Monday, which is a weekday after a holiday, it can be assumed that the load will be especially heavy at the time when air conditioning starts due to the influence of heat storage in the building structure. Further, even on a day in the middle of consecutive weekdays, the air conditioning load may increase immediately after the start of air conditioning (for example, before or after the start time).

Once the generation of the load prediction model M is completed, the operation of actually predicting the air conditioning load can be started.

The input/output unit 21 acquires values of various parameters used as explanatory variables in the load prediction model M via the networks 10 and 16, and sequentially records them in the data storage unit 17 (step S3). The load prediction unit 19 reads the numerical value of the parameter at a certain time or period from the data storage unit 17, and predicts the air conditioning load at the target time using the load prediction model M (step S4).

The air volume calculation unit 20 calculates the set value of the air volume in each variable air volume unit 5 based on the predicted value of the air conditioning load output from the load prediction unit 19 (step S5). The variable air volume unit 5 adjusts the damper opening degree based on the set value of the air volume, and supplies an appropriate volume of conditioned air A1 to the target space S. Note that the set air volume can be calculated based on the following formula (the predicted value of the air conditioning load calculated in step S4 is substituted for QSH in formula 6 below).
[Formula 6]
During cooling...V={3600QSH/Cp・ρ}×(tr-ts)
During heating...V={3600QSH/Cp・ρ}×(ts-tr)
however,
V: Air volume [m ³ /h]
QSH: Indoor sensible heat load [W]
Cp: Specific heat of air at constant pressure (approximately 1000 [J/(kg・K)])
ρ: Density of air (1.2 [kg/m ³ ])
tr: Set value of indoor temperature [℃]
ts: Set value of supply air temperature [℃]

The predicted value of the air conditioning load in step S4 is recorded in the data storage section 17 (step S6). Furthermore, the actual value of the air conditioning load at the predicted time is also recorded in the data storage section 17 (step S7).

The predicted value recorded in step S6 and the actual value recorded in step S7 do not necessarily match. Therefore, the model generation unit 18 modifies the load prediction model M based on the data recorded in steps S6 and S7 as necessary (step S8).

When steps S6 to S8 are completed, the process returns to step S3 to obtain new parameters, and predict the air conditioning load at the next point in time (step S4) and calculate the air volume (step S5).

According to such an air conditioning load prediction method and air conditioning system, it is possible to predict the air conditioning load at a target time and to reflect the predicted value in the operation of the air conditioning system in a feedforward manner. Therefore, the operating conditions can quickly follow the fluctuations in the load in the target space S, and air conditioning can be suitably performed. For example, if the air conditioning load in the target space S increases at a certain time on a certain day, the air volume in the variable air volume unit 5 may be increased after the air conditioning load actually increases and the air temperature in the target space S rises. Therefore, the air volume can be increased almost at the same time as when the air conditioning load is predicted to increase, or in advance of the predicted increase in the air conditioning load. Therefore, for the people in the target space S, a comfortable air-conditioned state can be maintained even if there are fluctuations in the air-conditioning load, and energy can be saved during air-conditioning by operating the air-conditioning appropriately with respect to the air-conditioning load. Can be used efficiently.

Note that the system configuration of the air conditioning system and the procedure for predicting the air conditioning load described here are merely examples. As long as the load prediction model M can be generated from the above-mentioned various parameters related to the operating status of the air conditioning system, and the air conditioning load can be predicted from the operating status at any time or period using this model, the system configuration and prediction procedure can be varied. Can be changed.

In addition, here, a case has been described in which the air conditioning load is handled through the air volume blown by the variable air volume unit 5, but the air conditioning load may be handled by adjusting the operating status of other devices. For example, operations such as turning on/off a plurality of air conditioners 1 and changing the temperature and supply amount of the conditioned air A1 from the air conditioner 1 are also possible according to the prediction results.

As described above, in the air conditioning load prediction method of the present embodiment, the air conditioning load is predicted using the load prediction model M generated by machine learning based on a plurality of parameters related to the air conditioning load. There is. In this way, the air conditioning load can be predicted with high accuracy.

In the air conditioning load prediction method of this embodiment, the load prediction model M incorporates multiple parameters related to the air conditioning load into the regression term as explanatory variables, and calculates the actual value of the air conditioning load at a point in the past from the target point using ARIMA. It can be the ARIMAX model used in the section. In this way, the air conditioning load can be predicted with even greater accuracy.

In the air conditioning load prediction method of this embodiment, a plurality of parameters selected from the day of the week, time of day, and outside air temperature can be used as explanatory variables of the regression term. In this way, the air conditioning load can be predicted with even greater accuracy.

In the air conditioning load prediction method of this embodiment, interaction terms between day of the week and outside temperature, time of day and outside temperature, and day of the week, time of day, and outside temperature can be used as regression terms. Prediction accuracy can be further improved while reducing the calculation load.

In the air conditioning load prediction method of this embodiment, the load prediction model M can use hour and minute dummies as explanatory variables. In this way, the air conditioning load can be predicted with even greater accuracy.

Further, the air conditioning system of the present embodiment is configured to adjust the operating status based on the predicted value of the air conditioning load by the above-mentioned air conditioning load prediction method. In this way, it is possible to predict the air conditioning load at a desired point in time and reflect it in the operation of the air conditioning system in a feedforward manner, so that the operating status can quickly follow changes in the load.

The air conditioning system of this embodiment includes an air conditioner 1 that sends out conditioned air A1, an air duct 2 that guides the conditioned air A1 from the air conditioner 1 to the target space S, and an outlet 4 that blows out the conditioned air A1 to the target space S. , and a variable air volume unit 5 that adjusts the air volume of the conditioned air A1 blown out from the outlet 4, and the air volume of the conditioned air A1 supplied from the variable air volume unit 5 to the target space S based on the predicted value of the air conditioning load. It is configured to adjust. In this way, by sending the conditioned air A1 in an amount corresponding to the predicted air conditioning load into the target space S, air conditioning can be suitably performed.

Therefore, according to this embodiment, it is possible to quickly respond to changes in air conditioning load.

Note that the air conditioning load prediction method and air conditioning system of the present invention are not limited to the above-described embodiments, and it goes without saying that various changes may be made without departing from the gist of the present invention.

1 Air conditioner 2 Air supply duct 4 Outlet 5 Variable air volume unit A1 Conditioned air M Load prediction model S Target space

Claims (6)

An air conditioning load prediction method that predicts an air conditioning load using a load prediction model generated by machine learning based on a plurality of parameters related to the air conditioning load, the method comprising:
The load prediction model is an ARIMAX model that incorporates a plurality of parameters related to the air conditioning load as explanatory variables into the regression term, and uses the actual value of the air conditioning load at a point in the past from the target point in the ARIMA term,
Using a plurality of parameters selected from day of week, time of day, and outside temperature as explanatory variables of the regression term,
An air conditioning load prediction method, characterized in that the calculation load is reduced by using an interaction term based on a combination of the plurality of parameters in the regression term . 2. The air conditioning load prediction method according to claim 1 , wherein the regression term uses at least one of interaction terms of day of the week and outside temperature, time of day and outside temperature, and day of the week, time of day, and outside temperature. 3. The air conditioning load prediction method according to claim 1 , wherein the load prediction model uses a time dummy as an explanatory variable. 4. The air conditioning load prediction method according to claim 3, wherein the load prediction model sets the setting interval of the hour and minute dummy to 15 minutes or more and 30 minutes or less. An air conditioning system configured to adjust operating conditions based on a predicted value of air conditioning load by the air conditioning load prediction method according to any one of claims 1 to 4 . An air conditioner that sends out conditioned air,
an air supply duct that guides conditioned air from the air conditioner to the target space;
An outlet that blows out conditioned air into the target space;
and a variable air volume unit that adjusts the air volume of the conditioned air blown out from the air outlet,
The air conditioning system according to claim 5, wherein the air conditioning system is configured to adjust the volume of conditioned air supplied from the variable air volume unit to the target space based on the predicted value of the air conditioning load.

Images