Yield prediction model generation program and predicted yield output program

The yield prediction model generation program enhances crop yield prediction by identifying correlations between environmental data and harvest amounts over multiple stages, addressing the limitations of existing methods and achieving higher accuracy in yield forecasting.

JP7576849B2Active Publication Date: 2024-11-01NAT AGRI & FOOD RES ORG
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2022007190
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2021-01-28
Filing Date
2022-01-20
Publication Date
2024-11-01
Estimated Expiration
2042-01-20

AI Technical Summary

Technical Problem

Existing yield prediction methods for agricultural crops, such as those described in Patent Documents 1 and 2, either require the use of aerial images or have limitations in prediction accuracy, particularly in estimating crop yields based on weather conditions.

Method used

A yield prediction model generation program that utilizes environmental data to generate a yield prediction model by identifying correlations between environmental data and harvest amounts over multiple stages, using a combination of stages and types of environmental data with a correlation greater than a predetermined value, and generates a predictive model based on these correlations.

Benefits of technology

This approach allows for accurate prediction of crop yields by considering the periodic characteristics of harvest amounts and environmental data, resulting in improved prediction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007576849000007
    Figure 0007576849000007
  • Figure 0007576849000008
    Figure 0007576849000008
  • Figure 0007576849000009
    Figure 0007576849000009
Patent Text Reader

Abstract

To generate a yield prediction model for accurately predicting crop yield.SOLUTION: A prediction model generation unit sets a plurality of time periods obtained by dividing a predetermined time period traced back from a harvest day as a plurality of stages (S12). Further, the prediction model generation unit refers to past data DB 50 that stores yield data in the past and a plurality of kinds of environmental data when the yield data is obtained in association with each other to specify environment data (a combination of a stage and a kind) whose correlation with the yield data is equal to or more than a predetermined value among a plurality of kinds of environment data in each of a plurality of stages (S14, S16). Then, the prediction model generation unit acquires environment data corresponding to the combination of the specified stage and the kind and the yield data stored in association with the environment data from the past data DB to generate a prediction model for predicting the yield based on the acquired data (S20).SELECTED DRAWING: Figure 4
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a harvest yield prediction model generation program and a predicted harvest yield output program. [Background technology]

[0002] In order to plan the shipment of agricultural crops such as tomatoes and peppers and to maintain stable prices for these crops, it is necessary to accurately predict harvest yields.

[0003] Conventionally, techniques for estimating the yield of agricultural crops are known, such as those described in Patent Documents 1 and 2. The technique in Patent Document 1 estimates the yield using an aerial photograph of a farm field and time-series weather data at a specific growth stage of the crop. The technique in Patent Document 2 generates a prediction model by performing a decision tree analysis using the cultivation conditions as explanatory variables and the cultivation results as objective variables, based on learning data that stores data correlating the past cultivation conditions and cultivation results of a plant. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2015-49 A [Patent Document 2] JP 2019-170359 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, the above Patent Document 1 requires the use of aerial photographs. In addition, although the above Patent Document 2 describes that explanatory variables of the prediction model represent weather conditions, there is room for improvement in prediction accuracy.

[0006] The present invention aims to provide a yield prediction model generation program capable of generating a yield prediction model for accurately predicting crop yields, and a predicted yield output program capable of accurately predicting crop yields. [Means for solving the problem]

[0007] The yield prediction model generation program of the present invention is a yield prediction model generation program that causes a computer to generate a yield prediction model for predicting a crop yield from environmental data, and causes the computer to execute the following process: setting multiple settable periods within a specified period prior to the harvest date of the crop as multiple stages, referring to a memory unit that stores past yield data in correspondence with multiple types of environmental data at the time the yield data was obtained, identifying a combination of stage and type of environmental data among the multiple types of environmental data in each of the multiple stages that has a correlation with the yield data that is greater than or equal to a specified value, acquiring from the memory unit environmental data corresponding to the identified combination of stage and type, and harvest data stored in correspondence with the environmental data, and generating a harvest prediction model based on the acquired environmental data and harvest data.

[0008] In addition, the predicted yield output program of the present invention is a program that causes a computer to execute a process of inputting environmental data corresponding to the harvest prediction date for which the harvest is to be predicted into the harvest prediction model generated by the harvest prediction model generation program of the present invention, and outputting the output result from the harvest prediction model as the predicted harvest on the harvest prediction date. Effect of the Invention

[0009] The present invention provides a method for generating a crop yield prediction model by using a computer program that generates a crop yield prediction model for accurately predicting a crop yield. The present invention also provides a method for outputting a predicted crop yield by using a computer program that generates a crop yield prediction model for accurately predicting a crop yield. [Brief description of the drawings]

[0010] [Figure 1] 1 is a diagram showing a configuration of an agricultural system according to a first embodiment. [Diagram 2] FIG. 2 is a diagram illustrating a hardware configuration of a control device in FIG. [Diagram 3] FIG. 2 is a functional block diagram of a control device. [Figure 4] 13 is a flowchart showing a prediction model generation process performed by a prediction model generation unit. [Diagram 5] FIG. 5(a) is a diagram showing the period going back from the harvest date determined for each crop, and FIG. 5(b) is a diagram showing the number of divisions determined for each crop. [Figure 6] 6(a) to 6(c) are diagrams for explaining the correlation between the harvest yield and the environmental data. [Figure 7] 7(a) and 7(b) are diagrams for explaining the periodic characteristics of the harvest yield. [Figure 8] 13 is a flowchart showing a harvest yield prediction process performed by a harvest yield prediction unit. [Figure 9] Figure 9(a) is a graph showing the results of harvest yield predictions along with actual harvest yields and comparative examples, and Figure 9(b) is a graph showing the results of harvest yield predictions (using machine learning: Random Forest) along with the actual harvest yield. [Figure 10] 10(a) and 10(b) are graphs showing the results of crop yield predictions when using environmental measurements per unit of solar radiation. [Figure 11] FIG. 1 shows important stages for each work. [Figure 12] 12(a) and 12(b) are diagrams for explaining a method for calculating the total dry matter growth amount. [Figure 13] FIG. 11 is a diagram showing a plurality of stages set when predicting a shipping volume of spinach in the second embodiment. [Figure 14] FIG. 13 is a diagram showing the results of calculating the correlation coefficient between each environmental factor and growth amount at each stage for varieties A and B. [Figure 15] This is a table showing the water temperature, air temperature, and solar radiation for each day since the planting date (April 12, 2021). [Figure 16] 13 is a table showing the values ​​of Srst_6, Twst_8, and Tast_2 used on each calculation date when predicting the shipping volume of product A, and the predicted and actual values ​​of the shipping volume on the shipping date. [Figure 17] 13 is a table showing the values ​​of Srst_7, Twst_8, and Tast_5 used on each calculation date when predicting the shipping volume of variety B, and the predicted and actual values ​​of the shipping volume on the shipping date. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] First Embodiment A first embodiment will be described in detail below with reference to Fig. 1 to Fig. 9. Fig. 1 shows a schematic configuration of an agricultural system 100 according to the first embodiment. The agricultural system 100 of this embodiment is a system that predicts and outputs the harvest yield and shipping volume of crops in a facility (e.g., a greenhouse) where crops such as tomatoes, peppers, cucumbers, and spinach are cultivated.

[0012] 1, the agricultural system 100 includes a control device 10, an outdoor sensor 12, a greenhouse sensor 14 installed in a greenhouse 18, and an environmental control device (hereinafter referred to as a controlled device) 16 that adjusts the environment inside the greenhouse 18. The control device 10, the outdoor sensor 12, the greenhouse sensor 14, and the controlled device 16 are connected via a network such as the Internet, and information can be exchanged between the devices.

[0013] The control device 10 is an information processing device that can be used by an operator who cultivates crops (such as tomatoes and peppers) in a greenhouse 18, and predicts and outputs (displays, etc.) the crop yield for a target prediction date or target prediction week based on environmental data acquired by the outdoor sensor 12 and the greenhouse sensor 14 and information input by the operator. The control device 10 can also control the controlled device 16. The configuration and processing of the control device 10 will be described in detail later.

[0014] The outdoor sensor 12 includes a temperature sensor that detects the air temperature outside the greenhouse 18 and a solar radiation sensor that detects solar radiation, and inputs the detection results to the control device 10.

[0015] The greenhouse sensors 14 include a temperature sensor that detects the air temperature inside the greenhouse 18, a solar radiation sensor that detects the solar radiation inside the greenhouse 18, and a CO2 concentration sensor that detects the CO2 concentration inside the greenhouse 18, and input the detection results to the control device 10.

[0016] The controlled devices 16 include a heat pump, a ventilation window, a heater, a CO2 applicator, a shading / heat-retaining curtain, etc. The heat pump is a device that lowers the temperature inside the greenhouse 18, and the ventilation window is a window that takes in outside air into the greenhouse 18. The heater is a device that raises the temperature inside the greenhouse 18, and the CO2 applicator is a device that adjusts the CO2 concentration inside the greenhouse 18. The shading / heat-retaining curtain is a curtain that adjusts the solar radiation and temperature inside the greenhouse 18. The controlled devices 16 are capable of executing operations according to instructions from the control device 10, and the environment inside the greenhouse 18 is adjusted by the operation of the controlled devices 16.

[0017] Here, the configuration and processing of the control device 10 will be described in detail. FIG. 2 shows a schematic diagram of the hardware configuration of the control device 10. As shown in FIG. 2, the control device 10 includes a CPU 90, a ROM 92, a RAM 94, a storage unit (HDD in this case) 96, a network interface 97, a display unit 93 as a display device, an input unit 95, and a drive for a portable storage medium 99. The display unit 93 includes a liquid crystal display and the like, and the input unit 95 includes a keyboard, a mouse, a touch panel, and the like. These components of the control device 10 are connected to a bus 98. In the control device 10, the CPU 90 executes a program (including a program for generating a harvest prediction model and a program for outputting a predicted harvest) stored in the ROM 92 or the HDD 96, or a program read by the drive for a portable storage medium 99 from the portable storage medium 91, thereby realizing the functions of the components shown in FIG. 3. The functions of the components shown in FIG. 3 may be realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0018] Fig. 3 shows a functional block diagram of the control device 10. In the control device 10, the CPU 90 executes a program to realize functions as an input receiving unit 30, an environmental information acquiring unit 32, an information managing unit 34, a predictive model generating unit 36, a harvest yield predicting unit 38, and an output unit 40, as shown in Fig. 3.

[0019] The input receiving unit 30 acquires actual harvest data and harvest date information input by the worker via the input unit 95, and transmits them to the information management unit 34. The input receiving unit 30 also acquires operation information (e.g., input information for an instruction to start prediction) input by the worker via the input unit 95, and transmits them to the harvest yield prediction unit 38.

[0020] The environmental information acquisition unit 32 acquires the detection results of the outdoor sensor 12 and the greenhouse sensor 14, and transmits them to the information management unit 34. The environmental information acquisition unit 32 also acquires past data (normal values, etc.) and forecast data (weather forecast mesh, etc.) from an external server or the like as future environmental data, and transmits them to the harvest yield prediction unit 38. Note that a value obtained by adding a predetermined temperature (α° C.) to the normal value may be used as the forecast data. The value of α may be appropriately set by the user, or the environmental information acquisition unit 32 may automatically set α based on the long-term forecast of the Japan Meteorological Agency, etc.

[0021] The information management unit 34 associates the information on the actual harvest date and harvest volume data input from the input reception unit 30 with environmental data corresponding to the harvest date (environmental data for a predetermined period prior to the harvest date), and manages them in the past data DB 50. The environmental data may be, for example, the average temperature, the average amount of solar radiation, and the average CO2 concentration for each day.

[0022] The prediction model generating unit 36 ​​generates a prediction model for predicting the harvest yield, using the data stored in the past data DB 50. The method of generating the prediction model will be described in detail later.

[0023] The harvest yield prediction unit 38 predicts the crop yield by inputting the environmental data corresponding to the prediction target date and the prediction target week into the prediction model generated by the prediction model generation unit 36. The method for predicting the harvest yield will be described in detail later.

[0024] The output unit 40 displays the results of the prediction by the yield prediction unit 38 (the predicted value of the yield for the prediction target day / week) on the display unit 93. In addition, the output unit 40 outputs control information to the controlled device 16.

[0025] (Regarding the processing of the control device 10) Next, the processing of the control device 10 will be described in detail with reference to the flowcharts of Figures 4 and 8 and other drawings as appropriate. The processing of Figure 4 is a prediction model generation process by the prediction model generation unit 36, and the processing of Figure 8 is a harvest yield prediction process for a prediction target day / week by the harvest yield prediction unit 38.

[0026] (Prediction model generation process) As a premise for the processing in FIG. 4, it is assumed that the information management unit 34 associates the actual harvest date information and harvest data input to the input receiving unit 30 with environmental data corresponding to the harvest data (environmental data for a predetermined period prior to the harvest date) and stores them in the past data DB 50. When the target of the prediction model to be generated is a single greenhouse, the environmental data to be associated with the harvest data is environmental data within the greenhouse. On the other hand, when the target of the prediction model to be generated is a multiple number of facilities in a single region (i.e., when the target is a "production area"), the environmental data to be associated with the harvest data is outdoor environmental data.

[0027] In the processing of Figure 4, first, in step S10, the predictive model generation unit 36 ​​identifies one crop and refers to the historical data DB 50 to obtain past yield data for the identified crop and environmental data for a specified period going back from the harvest date corresponding to the past yield data.

[0028] Here, the predetermined period is determined for each crop, and for example, as shown in FIG. 5(a), for cucumbers, it is a period of 300°C accumulated temperature going back from the harvest date, for spinach, it is a period of 600°C accumulated temperature going back from the harvest date, for tomatoes, it is a period of 1200°C accumulated temperature going back from the harvest date, and for peppers, it is a period of 1400°C accumulated temperature going back from the harvest date. Note that the predetermined period does not have to be determined based on the accumulated temperature. For example, the predetermined period may be a period from the harvest date to a date a predetermined number of days back. In addition to the crop, the predetermined period may be determined for each region or season. The environmental data is, for example, the average temperature of each day, the average solar radiation of each day, and the average CO2 concentration of each day.

[0029] Next, in step S12 of FIG. 4, the prediction model generating unit 36 ​​divides the predetermined period for which the environmental data was acquired by a division number n corresponding to the identified crop, and sets the divided n periods as n stages. The division number n is determined based on past knowledge and the like, and as shown in FIG. 5(b), for example, n=3 for cucumber, n=3 for spinach, n=6 for tomato, and n=7 for pepper. Therefore, in the case of tomatoes, the period from the harvest date of tomatoes to the time when the accumulated temperature is 1200°C is divided into ST1 to ST6. The prediction model generating unit 36 ​​may divide the period based on the accumulated temperature or the number of days. The predetermined period may also be divided unequally based on the knowledge of the operator or the like.

[0030] Next, in step S14, the prediction model generation unit 36 ​​calculates the correlation between each environmental data of each stage and the yield data. For example, in the case of tomatoes, the average temperature of each of the stages ST1 to ST6, the amount of solar radiation of each of the stages ST1 to ST6, and the CO2 concentration of each of the stages ST1 to ST6 can be acquired, so the correlation between each of these environmental data and the yield data (for example, the coefficient of determination R 2 For example, as shown in Fig. 6(a) to Fig. 6(c), if the varieties (A, B) are different even for the same crop, the correlation (coefficient of determination R 2 ) is also different. Therefore, the prediction model generating unit 36 ​​calculates the correlation (coefficient of determination R 2 ) is required.

[0031] Next, in step S16, the prediction model generation unit 36 ​​selects environmental data (a combination of stage and type) to be used in the prediction model based on the correlation obtained. There are various possible selection methods, but for example, the coefficient of determination R 2 Select environmental data (stage and type combinations) with a coefficient of determination R of 0.5 or higher. 2 is less than 0.5, the coefficient of determination R 2 A method such as selecting a predetermined number of environmental data (combinations of stage and type) in descending order of value can be adopted.

[0032] Next, in step S18, the prediction model generating unit 36 ​​calculates the periodic characteristics of the harvest yield. The harvest yield often increases or decreases periodically due to holidays at the collection center and crop characteristics. It is believed that alternate yearly fruit bearing in fruit trees and flashing in paprika are caused by the physiological characteristics (sink-source) of crops. For example, there may be a period in which a peak appears again 65 days after the peak harvest. In addition, if harvesting is done to coincide with an event, the shipping volume the following week may decrease significantly, and seasonal changes also play a major role. Therefore, in this step S18, such periodic characteristics of the harvest yield are calculated.

[0033] Specifically, the prediction model generating unit 36 ​​calculates the periodic characteristic H v_tr The calculation formula for the approximation curve is generated using past harvest data as shown in the following equation (1). H v_tr =Intercept+ΣBi×t i …(1)

[0034] Here, i=1, 2, ... in the above formula (1), and t means each actual measurement time point. The calculation coefficients B1 to B8 and the intercept of the calculation formula of the approximation curve are assumed to be values ​​as shown in FIG. 7(b). As shown in FIG. 7(a), the prediction model generation unit 36 ​​arranges the actual measurement values ​​of past harvest yields in chronological order (days, weeks) from the start of harvesting to the end of harvesting. Then, as shown in FIG. 7(b), the prediction model generation unit 36 ​​uses the least squares method to obtain an approximation curve that is an approximation curve of a nonlinear function for the harvest yield. In this approximation, the prediction model generation unit 36 ​​determines a coefficient that minimizes the sum of squares of the residuals. For example, a specific example of the above formula (1) is the following formula (1)'. H v_tr =0.005×t 3 -0.1539×t 2 +8.527×t+616.95 …(1)'

[0035] Next, in step S20, the prediction model generation unit 36 ​​generates a prediction model using the selected environmental data and periodic characteristics as shown in the following equation (2). Y a =f(Sr st_x ,Tm st_y ,…)·g(H v_z ,H v_w ,…) …(2)

[0036] Y in the above formula (2) a is the yield of crop a, f is a function derived from the analysis results, Sr st_x is the average solar radiation during stage STx, Tm st_y is the average temperature during stage STy, g is a function derived from the analysis results, H v_z is the yield at harvest week z, H v_w means the yield during harvest week w.

[0037] For example, let Y be the yield of crop a in week (w+2). a_w+2 In step S16, the environmental data selected for the item a is the amount of solar radiation Sr st_6 (R 2 =0.555), mean temperature Tm st_4 (R 2 =0.468), mean temperature Tm of stage ST1 st_1 (R 2 =0.470), solar radiation Sr st_4 (R 2 = 0.320). In addition, the periodic characteristic H v_tr Coefficient of determination R 2 is 0.210. Also, the yield in week w, H v_w and the yield in week (w+2) (coefficient of determination R 2 ) was 0.186.

[0038] In this case, for example, the prediction model generation unit 36 ​​generates a multiple regression equation such as the following equation (2)′. Y a_w+2 =276.03902×Tm st_4 -229.35906×Tm st_1 -152.73690×Tm st_2 +0.19759×H v_tr +0.04157×Hv_w +7999.60182 …(2)'

[0039] The above formula is an example, and the functions f and g can be changed in various ways. The prediction model generating unit 36 ​​may generate a machine learning model in which Y is the objective variable and Sr, Tm, and H are explanatory variables, instead of the above formula (multiple regression formula) of the prediction model.

[0040] The prediction model generation unit 36 ​​stores the generated prediction model in the prediction model storage unit 52 in association with the work.

[0041] With the above, all the processing in FIG. 4 is completed. If there are other crops for which a prediction model should be generated, the prediction model generation unit 36 ​​may repeatedly execute the processing in FIG. 4. The prediction model generation unit 36 ​​may also execute the processing in FIG. 4 at predetermined time intervals or whenever a predetermined number or more of new harvest volume data or environmental data are input, thereby updating the prediction model as needed. This allows the prediction model to be updated with the latest data, thereby improving the prediction accuracy of the prediction model.

[0042] (Harvest yield prediction processing) Next, the process by the harvest yield prediction unit 38 will be described with reference to the flowchart in Fig. 8. Note that, as a prerequisite for starting the flowchart in Fig. 8, it is assumed that an operator has input, via the input unit 95, the crop for which the harvest yield is to be predicted.

[0043] When the process of FIG. 8 is started, first, in step S50, the harvest yield prediction unit 38 acquires the input crop name, and acquires from the prediction model storage unit 52 the prediction model corresponding to the acquired crop name.

[0044] Next, in step S52, the harvest prediction unit 38 acquires environmental data and harvest data to be input to the prediction model. Specifically, the harvest prediction unit 38 identifies the environmental data (combination of stage and type) to be input to the prediction model (for example, the above formula (2)'), and acquires the identified environmental data and harvest data from the past data DB 50. Furthermore, when future environmental data needs to be input to the prediction model, the harvest prediction unit 38 acquires the future environmental data from the environmental information acquisition unit 32.

[0045] Next, in step S54, the harvest yield prediction unit 38 inputs the acquired environmental data and harvest yield data into a prediction model (for example, the above formula (2)'). As a result, the harvest yield prediction unit 38 obtains an output value from the prediction model as a predicted harvest yield.

[0046] Next, in step S56, the harvest yield prediction unit 38 displays the obtained predicted harvest yield on the display unit 93 via the output unit 40. This allows the worker to check the predicted harvest yield.

[0047] In step S56, the harvest prediction unit 38 may output a control instruction to the controlled device 16 based on the predicted harvest via the output unit 40, together with or instead of displaying the predicted harvest. This makes it possible to adjust the environment in the greenhouse so that the predicted harvest becomes an appropriate value. In this way, when the output unit 40 outputs a control instruction to the controlled device 16, the harvest may be predicted again based on the environmental data after controlling the controlled device 16. In this case, the output unit 40 may display the predicted value of the harvest after control on the display unit 93.

[0048] With the above, the entire process in FIG. 8 is completed.

[0049] Fig. 9(a) shows an example of the result of predicting the yield of tomatoes using the above formula (2)'. Also, Fig. 9(a) shows, as a comparative example, the result of predicting the yield of tomatoes based on the following formula (3) without considering the correlation between the yield of each environmental data. Y a_w+2 =13.6×Sr st_i +13.6×Tm st_i …(3)

[0050] In addition, Sr st_i is the average solar radiation during stages ST1, ST2, …ST6, Tm st_i is the average temperature during stages ST1, ST2, …ST6. For example, to predict the harvest in week 6 (35-41 days from the start date), the Tm st_4 , Tm st_1 , Tm st_2 , H v_tr , H v_w The data used are the values ​​in Table 1.

[0051] [Table 1]

[0052] These Tm st_4 , Tm st_1 , Tm st_2 , H v_tr , H v_w By substituting this into formula (2)', the predicted daily yield is calculated, and by aggregating these, the predicted weekly yield of 48,117.80 kg is calculated. Note that the actual yield at this time was 46,072 kg, so it can be seen that the use of formula (2)' allows for accurate prediction of the yield.

[0053] In addition, when predicting the yield in the 10th week (63-69 days from the start date), the daily Tm st_4 , Tm st_1 , Tm st_2 , H v_tr , H v_w Use the values ​​in Table 2 as data.

[0054] [Table 2]

[0055] These Tm st_4 , Tmst_1 , Tm st_2 , H v_tr , H v_w By substituting this into formula (2)', the predicted daily yield is calculated, and by aggregating these, the predicted weekly yield of 52,771.12 kg is calculated. Note that the actual yield at this time was 54,381 kg, so it can be seen that the use of formula (2)' allows for accurate prediction of the yield.

[0056] As shown in FIG. 9(a), by using the method according to this embodiment, the yield can be predicted with higher accuracy than in the comparative example.

[0057] Figure 9(b) shows the results of harvest yield prediction using a prediction model obtained by machine learning (Random Forest) along with actual measured values. As shown in Figure 9(b), even if the prediction model is generated by machine learning (Random Forest), the harvest yield can be predicted with high accuracy.

[0058] As described above in detail, according to the first embodiment, the prediction model generating unit 36 ​​sets a plurality of periods obtained by dividing a predetermined period preceding the harvest date of the crop as a plurality of stages (S12). The prediction model generating unit 36 ​​also refers to the past data DB 50 that stores past yield data in association with a plurality of types of environmental data at the time the yield data was obtained, and identifies environmental data (a combination of a stage and a type) that has a correlation with the yield data that is greater than or equal to a predetermined value among a plurality of types of environmental data in each of the plurality of stages (S14, S16). The prediction model generating unit 36 ​​then acquires from the past data DB 50 the environmental data corresponding to the identified combination of stage and type, and the yield data stored in association with the environmental data, and generates a prediction model for predicting the yield based on the acquired data (S20). As a result, in this embodiment, the yield is predicted using environmental data with a correlation greater than or equal to a predetermined value, and therefore a prediction model that can predict the yield with higher accuracy can be generated than a prediction model that can be generated using environmental data including environmental data with a correlation less than a predetermined value. Furthermore, by predicting the harvest yield using the prediction model generated as described above, the harvest yield can be predicted with high accuracy.

[0059] Furthermore, in this embodiment, a prediction model is generated (S18, S20) based on the tendency of change in the harvest data (the cyclical characteristics of the harvest) stored in the past data DB 50. This makes it possible to generate a prediction model that is capable of predicting the harvest, taking into account the cyclical characteristics of the harvest.

[0060] In addition, in this embodiment, rules for the period going back from the harvest date and the number of divisions are set for each crop (Figures 5(a) and 5(b)), and multiple stages are set based on these periods going back and the number of divisions, so that appropriate stage settings can be made taking into account the characteristics of the crop.

[0061] In the first embodiment, the environmental data may be a value obtained by dividing the environmental measurement value (average temperature or CO2 concentration) by the amount of solar radiation. Since solar radiation has a strong effect on the yield, the use of the environmental measurement value per solar radiation makes it easier to reflect the influence of factors other than solar radiation in the prediction model. FIG. 10(a) shows the result of predicting the yield using the environmental measurement value per solar radiation (using machine learning: Random Forest, as in FIG. 9) together with the actual measurement value. By using the environmental measurement value per solar radiation, the yield can be predicted with high accuracy, as shown in FIG. 10(a). In addition, when using a linear regression model or a gradient boosting model in predicting the yield using the environmental measurement value per solar radiation, the yield can be predicted with high accuracy, as shown in FIG. 10(b).

[0062] In the above first embodiment, the prediction model generating unit 36 ​​specifies the environmental data (combination of stage and type) used to generate the prediction model based on the correlation with the harvest yield data, but the present invention is not limited to this. For example, when information on important stages that have a large physiological environmental impact is known in advance, the prediction model generating unit 36 ​​may specify the environmental data (combination of stage and type) to be used to generate the prediction model by taking that information into consideration. For example, as shown in FIG. 11, when an important stage is determined for each crop, the environmental data of that stage may be used unconditionally (irrespective of the magnitude of correlation) to generate the prediction model. Furthermore, the environmental data of the important stage may be preferentially used to generate the prediction model. In this case, for example, the correlation (coefficient of determination R 2 11, important stages are indicated by specific stage names, but in reality, they are indicated by the abbreviation "STx" of the stage shown in FIG. 5(b).

[0063] In the above first embodiment, a case has been described in which the periodic characteristic is used in the prediction model, but the present invention is not limited to this, and the prediction model may not use the periodic characteristic.

[0064] Second Embodiment Next, a second embodiment will be described. In the above first embodiment, the prediction model generating unit 36 ​​sets a plurality of periods obtained by dividing a predetermined period going back from the harvest date of the crop as a plurality of stages. In contrast, in the present second embodiment, the prediction model generating unit 36 ​​sets a plurality of periods based on a predetermined date (a calculation date described later) that can be set within a predetermined period going back from the harvest date (shipping date) of the crop as a plurality of stages. Note that the system configuration, device configuration, etc. of the present second embodiment are the same as those of the first embodiment, and therefore description thereof will be omitted.

[0065] In the second embodiment, the shipping volume forecast of spinach will be described as an example. In the shipping volume forecast of spinach, the correlation between the dry matter weight gain (photosynthetic products) of spinach and weather data is used to calculate the daily dry matter growth volume as shown in FIG. 12(a). Then, the total dry matter growth volume is calculated by accumulating the daily dry matter growth volume as shown in FIG. 12(b). The shipping volume is calculated based on the following formula (4) using the fresh weight dry matter ratio of spinach (ratio coefficient of dry matter weight and fresh weight). Fresh weight (shipping amount) = (total dry weight) ÷ (fresh weight dry weight rate) … (4)

[0066] Fig. 13 is a diagram showing a number of stages that are set when predicting the shipping amount of spinach in the second embodiment. In the second embodiment, a number of stages as shown in Fig. 13 are set in a period going back from the target date (calculation date) for calculating the dry matter growth amount (10 days including the calculation date). Specifically, stage ST0 is the calculation date, stage ST1 is one day before the calculation date, stage ST2 is two days before the calculation date, ..., stage ST9 is nine days before the calculation date.

[0067] In the second embodiment, the prediction model generation unit 36 ​​calculates the correlation coefficient between the daily growth amount on the calculation date and the environmental factors (average water temperature, average temperature, average solar radiation) of each stage. Figure 14 shows the results of calculating the correlation coefficient between each environmental factor and the growth amount of each stage for varieties A and B.

[0068] The prediction model generation unit 36 ​​determines the environmental factors for which the absolute value of the correlation coefficient is 0.5 or more and which have the largest absolute value of the correlation coefficient among the environmental factors to be used in generating the prediction model. Note that if there is no stage for each environmental factor with an absolute value of the correlation coefficient of 0.5 or more, that environmental factor is not used in the prediction model.

[0069] (In the case of variety A) For example, in the case of variety A, as shown in Fig. 14, for the environmental factor: average water temperature, the absolute value of the correlation coefficient for stage ST8 is 0.7532, which is the largest, so the prediction model generation unit 36 ​​determines to use the average water temperature for stage ST8 in the prediction model. Also, for the environmental factor: average temperature, the absolute value of the correlation coefficient for stage ST2 is 0.7638, which is the largest, so the prediction model generation unit 36 ​​determines to use the average temperature for stage ST2 in the prediction model. Furthermore, for the environmental factor: average solar radiation, the absolute value of the correlation coefficient for stage ST6 is 0.7424, which is the largest, so the prediction model generation unit 36 ​​determines to use the average solar radiation for stage ST6 in the prediction model.

[0070] Then, the prediction model generating unit 36 ​​creates a prediction model as shown in the following formula (5) using the past shipment amount data.

[0071]

number

[0072] Y is the shipping amount, f is the calculation function for the daily dry weight, and Sr st_6 is the average solar radiation in stage ST6, Tw st_8 is the average water temperature of stage ST8, Ta st_2is the average temperature in stage ST2. Also, g(t) is the fresh weight dry matter ratio of variety A (ratio coefficient of dry matter weight to fresh weight = dry matter weight / fresh weight).

[0073] Specifically, the prediction model for variety A can be expressed as the following equation (5)' based on the selected correlation coefficient.

[0074]

number

[0075] Figure 15 is a table showing the water temperature, temperature, and solar radiation for each day after the planting date (April 12, 2021). The shipping date of this variety A is May 9, 2021, which is the 28th day after the planting date. st_6 -0.2629Tw st_8 -0.3.567Ta st_2 +7.1802) on the 28th day is the average solar radiation (Sr st_6 ), the average water temperature of stage ST8 (Tw st_8 ), the average temperature of stage ST2 (Ta st_2 ) is used to calculate the average solar radiation (Sr st_6 ) is the average solar radiation for six days from May 3 to May 8. Also, the average water temperature (Tw st_8 ) is the average water temperature for eight days from May 1 to May 8. The average temperature of stage ST2 (Ta st_2 ) is the average temperature over the two days from May 7th to May 8th.

[0076] Similarly, the calculation dates are the 27th, 26th, ..., and the (-0.1397Sr st_6 -0.2629Tw st_8 -0.3.567Ta st_2 +7.1802) is also calculated. st_6 -0.2629Tw st_8 -0.3.567Ta st_2+7.1802) into the above formula (5)' to calculate the shipment volume on the 28th day. st_6 , Tw st_8 , Ta st_2 By using the values ​​in Figure 16, the predicted shipping amount on the shipping date (28th day) was calculated to be 71.27g. In contrast, the actual measured shipping amount of spinach was 68.97g, so it was found that the shipping amount can be predicted with high accuracy by using this method.

[0077] (For variety B) 14, for variety B, for the environmental factor: average water temperature, the absolute value of the correlation coefficient for stage ST8 is 0.7465, which is the largest, so the prediction model generation unit 36 ​​determines to use the average water temperature for stage ST8 in the prediction model. Also, for the environmental factor: average temperature, the absolute value of the correlation coefficient for stage ST5 is 0.6318, which is the largest, so the prediction model generation unit 36 ​​determines to use the average temperature for stage ST5 in the prediction model. Furthermore, for the environmental factor: average solar radiation, the absolute value of the correlation coefficient for stage ST7 is 0.7397, which is the largest, so the prediction model generation unit 36 ​​determines to use the average solar radiation for stage ST7 in the prediction model.

[0078] Then, the prediction model generating unit 36 ​​creates a prediction model as shown in the following formula (6) using the past shipment amount data.

[0079]

number

[0080] Y is the shipping amount, f is the calculation function for the daily dry weight, and Sr st_7 is the average solar radiation in stage ST7, Tw st_8 is the average water temperature of stage ST8, Ta st_5 is the average temperature at stage ST5. w(t) is the fresh weight dry matter percentage of variety B.

[0081] Specifically, the prediction model for variety B can be expressed as the following equation (6)' based on the selected correlation coefficient.

[0082]

number

[0083] Like variety A, the planting date of variety B is April 12, 2021, and the shipping date is May 9, 2021 (the 28th day after the planting date). st_7 +0.1051Tw st_8 -0.3.516Ta st_5 +2.2777) on the 28th day is the average solar radiation (Sr st_7 ), the average water temperature of stage ST8 (Tw st_8 ), the average temperature of stage ST5 (Ta st_5 ) is used to calculate the average solar radiation (Sr st_7 ) is the average solar radiation for 7 days from May 2 to May 8. Also, the average water temperature (Tw st_8 ) is the average water temperature for eight days from May 1 to May 8. The average temperature of stage ST5 (Ta st_5 ) is the average temperature over the five days from May 4th to May 8th.

[0084] Similarly, the calculation dates are the 27th, 26th, ..., and the (-0.0537Sr st_7 +0.1051Tw st_8 -0.3.516Ta st_5 +2.2777) is also calculated. st_7 +0.1051Tw st_8 -0.3.516Ta st_5 +2.2777) into the above formula (6)' to calculate the shipment volume on the 28th day. st_7 , Tw st_8 , Ta st_5By using the values ​​in Figure 17, the predicted shipping amount on the shipping date (28th day) was calculated to be 58.23g. In contrast, the actual measured shipping amount of spinach was 57.17g, so it was found that the shipping amount can be predicted with high accuracy by using this method.

[0085] The above processing functions can be realized by a computer. In that case, a program is provided that describes the processing contents of the functions that the processing device should have. The above processing functions are realized on the computer by executing the program on the computer. The program that describes the processing contents can be recorded on a computer-readable storage medium (excluding carrier waves).

[0086] When a program is distributed, it is sold in the form of a portable storage medium on which the program is recorded, such as a DVD (Digital Versatile Disc), a CD-ROM (Compact Disc Read Only Memory), etc. The program can also be stored in a storage device of a server computer and transferred from the server computer to other computers via a network.

[0087] A computer that executes a program stores, for example, a program recorded on a portable storage medium or a program transferred from a server computer in its own storage device. The computer then reads the program from its own storage device and executes processing in accordance with the program. The computer can also read the program directly from a portable storage medium and execute processing in accordance with that program. The computer can also execute processing in accordance with the received program each time a program is transferred from the server computer.

[0088] The above-described embodiment is a preferred example of the present invention, but the present invention is not limited to this embodiment and can be modified in various ways without departing from the scope of the present invention. [Explanation of symbols]

[0089] 10 Control device 12 Outdoor Sensors 14 Greenhouse sensor 16 Controlled Equipment 30 Input reception section 32 Environmental Information Acquisition Department 34 Information Management Department 36 Prediction model generation unit 38 Crop Yield Forecasting Department 40 Output section 100 Agriculture System

Claims

1. A crop yield prediction model generation program for causing a computer to generate a crop yield prediction model for predicting a crop yield from environmental data, comprising: A plurality of periods that can be set within a predetermined period preceding the harvest date of the crop are set as a plurality of stages; referring to a storage unit that stores past harvest data in association with a plurality of types of environmental data at the time the harvest data was obtained, identifying a combination of a stage and a type of environmental data that has a correlation with the harvest data that is equal to or higher than a predetermined value among the plurality of types of environmental data at each of the plurality of stages; acquiring, from the storage unit, environmental data corresponding to the identified combination of the stage and type, and harvest yield data stored in association with the environmental data, and generating a harvest yield prediction model based on the acquired environmental data and harvest yield data. A harvest yield prediction model generating program that causes the computer to execute a process.

2. 2 . The harvest yield prediction model generating program according to claim 1 , wherein in the generating process, the harvest yield prediction model is generated based on a trend of change in the harvest yield data stored in the storage unit.

3. 3. The harvest yield prediction model generating program according to claim 1, wherein the environmental data is a value obtained by dividing an environmental measurement value other than the amount of solar radiation by the amount of solar radiation.

4. In the setting process, the predetermined period and the plurality of stages are set based on rules defined for each type of crop, In the process of identifying, a combination of a stage and a type of the environmental data is identified for each type of the crop; 4. The harvest yield prediction model generating program according to claim 1, wherein in the generating process, the harvest yield prediction model is generated for each type of crop.

5. A harvest yield prediction model generation program as described in any one of claims 1 to 4, characterized in that in the setting process, multiple periods obtained by dividing the specified period are set as the multiple stages, or multiple periods based on a specified date that can be set within the specified period are set as the multiple stages.

6. Inputting environmental data corresponding to a harvest prediction target date for predicting a harvest into the harvest prediction model generated by the harvest prediction model generation program according to any one of claims 1 to 5, An output result from the harvest yield prediction model is output as a predicted harvest yield on the harvest yield prediction target date. A predicted yield output program that causes the computer to execute a process.

Citation Information

Patent Citations

  • Harvest-predicting system and harvest-predicting apparatus

    JP2015000049A

  • Farming management system, server, farming management method, server control method and server program

    JP2017046613A

  • Culture support device, culture support method, computer program and recording medium

    JP2019030253A

  • Plant cultivation result prediction system

    JP2019170359A

  • Crop production assisting program, crop production assisting method, and crop production assisting device

    JP2021056572A