Method and device for creating a prediction model for predicting tea leaf growth, method and device for predicting tea leaf growth, and prediction system

Abstract

To provide models suitable for predicting tea leaf growth.SOLUTION: A method for generating a prediction model for predicting tea leaf growth comprises a generation step of performing machine learning using the integrated value obtained by accumulating the past weather data of the field where tea plants are grown for a predetermined period and the degree of past tea leaf growth as learning data, and thereby generating a plurality of prediction models, in which the input information is the integrated value of weather data for a predetermined period from an arbitrary day of a target year and the output information is the information representing the degree of tea leaf growth, according to the cumulative number of days of the weather data included in the learning data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a method and apparatus for generating a prediction model for predicting the growth of tea leaves, a prediction method and apparatus for predicting the growth of tea leaves, and a prediction system.

Background Art

[0002] Tea is a crop that picks new shoots. If the picking is delayed, the quality deteriorates, while if it is too early, the yield decreases. Therefore, in order to manage the quality and ensure a stable yield, it is necessary to pick at the appropriate time. Conventionally, it has been common to predict the appropriate picking time relying on visual judgment and empirical knowledge, but the prediction is likely to vary, and in the case where the tea garden is on a mountainside, visual observation involves a great deal of labor. As techniques for reducing such prediction variations and labor, the techniques described in Non-Patent Documents 1 to 2 and Patent Documents 1 to 2 are known.

[0003] Non-Patent Document 1 describes a technique for predicting the appropriate picking time using an expression representing the relationship between the accumulated temperature, the number of unfolded leaves, and the length of new shoots. Non-Patent Document 2 describes a method for predicting the tea picking period from the meteorological data of AMeDAS.

[0004] Patent Document 1 describes a technique for predicting the appropriate picking time of tea leaves based on the optical data of the image information of tea leaves. Patent Document 2 describes a technique for formulating a plan for the appropriate timing and management mode of each management element such as autumn pruning, fertilization, and pest control in a tea garden based on at least one of the tea tree environmental information and the economic environmental information.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Non-Patent Documents

[0006] [Non-Patent Document 1] Journal of Tea Science, 97:39-47, "Relationship between the growth of the first flush shoots and the accumulated temperature in the main tea cultivar 'Yabukita'", 2004 [Non-Patent Document 2] Research and Development Results Information of Fujinokuni Research Institute for Fiscal Year 2016, "Tea Garden Management Support Software for Predicting the Picking Period of Tea Using Mesh Agriculture Meteorology" [Summary of the Invention] [Problems to be Solved by the Invention]

[0007] The growth of tea leaves is affected by various weather conditions and cultivation conditions. Therefore, in order to predict the appropriate picking period of tea leaves, it is necessary to consider various weather conditions and cultivation conditions. When constructing a prediction model for predicting the appropriate picking period of tea leaves, the prediction models obtained may differ depending on the types and numbers of explanatory variables selected from various weather conditions and cultivation conditions. Since the prediction accuracy of the appropriate picking period of tea leaves may differ depending on the prediction model used, it is necessary to select the optimal prediction model for accurate prediction. Non-Patent Documents 1-2 and Patent Documents 1-2 do not describe anything about selecting and updating the optimal prediction model including variable changes according to the number of days to be predicted from among a plurality of prediction models for predicting the appropriate picking period of tea leaves.

[0008] One aspect of the present invention has been made to solve the above-described problems, and an object thereof is to realize a technique for providing a model suitable for predicting the growth of tea leaves. [Means for Solving the Problems]

[0009] The generation method according to one aspect of the present invention is a method for generating a prediction model for predicting the growth of tea leaves. By performing machine learning using, as learning data, the integrated value of past weather data of a field where tea trees grow accumulated over a predetermined period and the degree of past tea leaf growth, a prediction model in which the integrated value of weather data for a predetermined period from an arbitrary day of the target year is input information and information representing the degree of tea leaf growth is output information is generated in plural according to the number of days of integration of the weather data included in the learning data.

[0010] The generation method according to one aspect of the present invention is a method for generating a prediction model for predicting the growth of tea leaves. By performing machine learning using, as learning data, the difference between the integrated values of past weather data of a field where tea trees grow accumulated over a predetermined period for any years and the difference in the degree of tea leaf growth for the any years, a prediction model in which the difference between the integrated values of the past arbitrary year and the target year of the field where tea trees grow is input information and information representing the difference in the degree of tea leaf growth between the past arbitrary year and the target year is output information is generated.

[0011] The prediction method for predicting the growth of tea leaves according to one aspect of the present invention includes a step of predicting the growth of tea leaves using the prediction model generated by the generation method according to one aspect of the present invention.

[0012] The generation apparatus according to one aspect of the present invention is a generation apparatus for a prediction model for predicting the growth of tea leaves, and includes a generation unit that generates, in plural according to the number of days of integration of weather data, a prediction model in which the integrated value of weather data for a predetermined period from an arbitrary day of the target year is input information and information representing the degree of tea leaf growth is output information by performing machine learning using, as learning data, the integrated value of past weather data of a field where tea trees grow accumulated over a predetermined period and the degree of past tea leaf growth.

[0013] The generation device according to one aspect of the present invention is a generation device for a prediction model that predicts the growth of tea leaves. By performing machine learning using, as learning data, the difference between any annual integrated values obtained by integrating past meteorological data of a tea plantation for a predetermined period and the difference between any annual degrees of tea leaf growth, the difference between the integrated values of any past year and the target year of the tea plantation where the tea trees grow is input information, and a generation unit that generates a prediction model in which information representing the difference in the degree of tea leaf growth between any past year and the target year is output information is provided.

[0014] The prediction device according to one aspect of the present invention includes a prediction unit that predicts the growth of tea leaves using the prediction model generated by the generation device according to one aspect of the present invention.

[0015] The prediction system for predicting the growth of tea leaves according to one aspect of the present invention includes the generation device according to one aspect of the present invention and the prediction device according to one aspect of the present invention.

Advantages of the Invention

[0016] According to one aspect of the present invention, it is possible to realize a technique for providing a model suitable for predicting the growth of tea leaves.

Brief Description of the Drawings

[0017]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Mode for Carrying Out the Invention

[0018] Hereinafter, an aspect of the present invention realizes a prediction system for predicting the growth of tea leaves. The prediction system according to an aspect of the present invention can predict the leaf-opening stage of tea leaves, and can respectively predict the 1-leaf-opening stage, 1.5-leaf-opening stage, 2-leaf-opening stage, 3-leaf-opening stage, 4-leaf-opening stage, 5-leaf-opening stage, 6-leaf-opening stage, etc. Note that the 1-leaf-opening stage to 6-leaf-opening stage represent the number of newly opened leaves and the corresponding time periods, and represent the time periods from when a new leaf starts to open to when 6 new leaves have opened, for each number of opened leaves. The appropriate picking period of tea leaves is appropriately set according to the balance between quality and yield, but generally, the period from the 3-leaf-opening stage to the 5-leaf-opening stage can be the appropriate picking period. Therefore, the prediction system according to an aspect of the present invention preferably predicts the period from the 3-leaf-opening stage to the 5-leaf-opening stage of tea leaves, and more preferably predicts the 4-leaf-opening stage of tea leaves. The prediction system according to an aspect of the present invention can also be referred to as a prediction system for predicting the appropriate picking period of tea leaves.

[0019] In addition, the prediction system according to an aspect of the present invention can support the harvesting operation of tea leaves and the cultivation management of tea trees by predicting the growth of tea leaves. Regarding the support for the cultivation management of tea trees, the information on the growth of tea leaves predicted by the prediction system according to an aspect of the present invention can be used for determining the pruning period of tea trees, the start time of covering cultivation of tea trees (the time when tea trees are covered with light-shielding materials, etc.).

[0020] 〔Prediction System 100〕 Based on FIG. 1, a prediction system 100 for predicting the growth of tea leaves will be described. FIG. 1 is a block diagram showing an example of the main configuration of the prediction system 100 according to one aspect of the present invention. The prediction system 100 includes a model generation device (generation device) 10 and a prediction device 20. Further, the prediction system 100 further includes an input device 30, a storage device 40, and an output device 50. The prediction system 100 may include the model generation device 10 and the prediction device 20 as independent devices, or may be integrally provided in one device.

[0021] Here, the prediction of the growth of tea leaves by the prediction system 100 will be described with reference to FIG. 2. FIG. 2 is a diagram for explaining the concept of predicting the growth of tea leaves using a prediction model, which is executed by the prediction system 100 according to one aspect of the present invention. As shown in FIG. 2, as an example, the prediction model is generated by performing machine learning using, as learning data, data associating past weather data in the field where the tea tree to be predicted grows and the picking period of that year.

[0022] In the prediction model generated in this way, the weight values are learned so that the output layer outputs the appropriate picking period of the tea leaves in the target year by inputting the weather data of the target year as input information into the input layer. The generated prediction model performs an operation based on the learned weight values on the input weather data of the target year to be predicted, and is a learning model for causing a computer to function so as to output the appropriate picking period of the tea leaves in that year from the output layer. Thus, according to the prediction system 100, it is possible to predict the appropriate picking period of the year only by inputting the weather data of the target year to be predicted.

[0023] The input device 30 receives an input operation by the user to the prediction system 100. As an example, the input device 30 receives the input of data used for generating a prediction model in the model generation device 10. Further, the input device 30 receives the input of data used for predicting the growth of tea leaves in the prediction device 20.

[0024] The storage device 40 stores the programs and data used in the prediction system 100. As an example, the storage device 40 stores various data input via the input device 30. Also, as an example, the storage device 40 stores the learning data used for generating the prediction model and the generated prediction model in the model generation device 10. Further, as an example, the storage device 40 stores the prediction model, input information, and output information used for predicting the growth of tea leaves in the prediction device 20. The storage device 40 may have a database for storing various data on the cloud or a server.

[0025] The output device 50 outputs the result predicted by the prediction device 20. Also, the output device 50 may output harvesting work support information and cultivation management support information based on the result predicted by the prediction device 20. The harvesting work support information includes, for example, information indicating the appropriate harvesting time. The cultivation management support information includes, for example, information indicating the pruning period of the tea tree, the period when the tea tree is covered with a light-shielding material, etc.

[0026] The output mode by the output device 50 is not particularly limited. The output device 50 may be, for example, a display device that displays the information as an image, a printing device that prints the information, or an alarm device that outputs the information as sound. Also, the output device 50 may be a display of a mobile device such as a smartphone that displays the result predicted by the prediction device 20, the harvesting work support information and cultivation management support information based on the result, etc.

[0027] (Model Generation Device 10) The model generation device 10 is a device that generates a prediction model for predicting the degree of growth of tea leaves. The model generation device 10 generates a plurality of prediction models for predicting the degree of growth of tea leaves from the integrated values of meteorological data for a predetermined period from an arbitrary day of the target year, according to the combination of data to be referred to. Here, the case where the prediction model generated by the model generation device 10 predicts the appropriate picking time of tea leaves as an example of the degree of growth of tea leaves will be described. Also, "an arbitrary day of the target year" may be set as appropriate according to the purpose of prediction and the like. For example, it may be the budding day of the tea tree in the target year. In the following embodiments, the case where the budding day is used as an example of "an arbitrary day of the target year" will be described.

[0028] The model generation device 10 includes a control unit 11. The control unit 11 comprehensively controls each part of the model generation device 10 and is realized by, for example, a processor and a memory. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions included in the program. Thereby, each part of the control unit 11 is configured. As each of these parts, the control unit 11 includes a data acquisition unit 12, a data conversion unit 13, a model generation unit 14, and a model selection unit 15.

[0029] <Data acquisition unit 12> The data acquisition unit 12 acquires data for generating a prediction model. The data acquisition unit 22 reads data from the storage device 40 based on an input signal representing an instruction to start generating a prediction model from the input device 30. Also, the data acquisition unit 12 may acquire data input via the input device 30. The data acquisition unit 12 outputs the acquired data to the data conversion unit 13.

[0030] The data acquired by the data acquisition unit 12 may be, as an example, data in which the measured values of past meteorological data of the tea plantation where tea trees grow are associated with the appropriate picking period of tea leaves in that year. The meteorological data is, for example, data of variables such as temperature, humidity, rainfall, sunshine hours, etc., and is available from the website of the Meteorological Agency, etc. for each past year. The past tea leaf picking periods are, as an example, data of the four-leaf stage for each past year, and for example, data obtained in the past in the field where prediction is to be performed may be used.

[0031] <Data conversion unit 13> The data conversion unit 13 converts the data acquired by the data acquisition unit 12 into data suitable for generating a prediction model. The conversion of data by the data conversion unit 13 can be executed using a known data transformer or the like. The data conversion unit 13 outputs the converted data to the model generation unit 14.

[0032] The data conversion unit 13, as an example, calculates an integrated value obtained by integrating meteorological data for each variable over a predetermined period. The data conversion unit 13 calculates, for example, the integrated values of the minimum temperature, rainfall, and sunshine hours respectively. Also, the data conversion unit 13 calculates the integrated values obtained by integrating meteorological data for each variable for each number of accumulated days. The data conversion unit 13 calculates, for example, integrated values obtained by integrating meteorological data for n days (n is an integer of 1 or more), such as integrated values obtained by integrating meteorological data for 4 days and integrated values obtained by integrating meteorological data for 5 days, for each number of accumulated days.

[0033] The data conversion unit 13 generates conversion data in which the integrated values of meteorological data and the number of accumulated days calculated as described above are associated with the germination date and the appropriate picking period of that year for each year. The data conversion unit 13 can generate a plurality of such conversion data according to the combination of variables of the meteorological data. Also, the data conversion unit 13 can generate a plurality of conversion data according to the number of accumulated days of the integrated meteorological data.

[0034] An example of data conversion by the data conversion unit 13 will be described with reference to FIG. 3. FIG. 3 is a diagram for explaining data conversion processing in the model generation apparatus 10 according to an aspect of the present invention. As shown in FIG. 3, the data acquired by the data acquisition unit 12 includes, as an example, weather data for each past day for each variable. Further, the data may include information regarding the budding dates of each past year. Using such data, for example, when the number of input days is set to 4 days from April 6, 1999, conversion data obtained by integrating the weather data for 4 days from April 6 to April 9, 1999 for each variable is obtained by a data transformer.

[0035] Further, the data conversion unit 13 may calculate the difference between the integrated value of weather data for an arbitrary past year and the integrated value of weather data for another arbitrary past year. Furthermore, the data conversion unit 13 may calculate the difference between the picking appropriate period for the arbitrary year and the picking appropriate period for the other arbitrary year. Then, the data conversion unit 13 can generate conversion data in which the calculated difference in weather data is associated with the difference in the picking appropriate period.

[0036] <Model Generation Unit 14> The model generation unit 14 generates a prediction model by performing machine learning using the conversion data converted by the data conversion unit 13 as learning data. The model generation unit 14 generates a prediction model in which the integrated value of weather data for a predetermined period from the budding date of the tea tree in the target year is input information and information representing the degree of growth of tea leaves is output information.

[0037] As an example, the model generation unit 14 generates a prediction model using a known machine learning method such as a neural network, a decision tree, a random forest, or a support vector machine. The model generation unit 14 outputs the generated prediction model to the model selection unit 15. Further, the model generation unit 14 may store the generated prediction model in the storage device 40.

[0038] The model generation unit 14 generates a plurality of prediction models according to combinations of meteorological data included in the learning data. A plurality of conversion data are generated in the data conversion unit 13 according to combinations of variables of the meteorological data. That is, certain conversion data includes integrated values of the minimum temperature, rainfall amount, and sunshine duration as variables of the meteorological data, and other conversion data includes integrated values of the maximum temperature, humidity, and sunshine duration. Thus, the generated prediction models can be different due to different combinations of variables of the meteorological data included in the conversion data.

[0039] In addition, the model generation unit 14 generates a plurality of prediction models for each accumulation number of days of the integrated values of the meteorological data included in the learning data. A plurality of conversion data are generated in the data conversion unit 13 for each accumulation number of days of the integrated values of the meteorological data. That is, certain conversion data includes an integrated value obtained by integrating the meteorological data for 4 days, and other conversion data includes an integrated value obtained by integrating the meteorological data for 5 days. Thus, the generated prediction models can be different due to different accumulation numbers of days of the meteorological data included in the conversion data.

[0040] The model generation unit 14 performs machine learning using each of the plurality of conversion data as learning data, thereby generating a plurality of prediction models for each learning data. Since machine learning can be parallelized to generate prediction models using each of the plurality of conversion data as learning data, the calculation speed is increased and the generation speed of the prediction models is increased. Also, if the number of learning data increases due to an increase in the number of years of measurement of the meteorological data, the generated prediction models are updated, and the prediction accuracy can be further improved.

[0041] Regarding the plurality of prediction models generated by the model generation unit 14, a description will be given with reference to FIG. 4. FIG. 4 is a diagram showing an example of a plurality of prediction models generated by the model generation device 10 according to one aspect of the present invention. In the model group shown in FIG. 4, the models numbered No. 1 to No. 7 are integrated values obtained by integrating meteorological data for 4 days, the model numbered No. 8 is an integrated value obtained by integrating meteorological data for 11 days, and the models numbered No. 9 to No. 11 are generated using learning data including integrated values obtained by integrating meteorological data for 12 days, respectively. Also, the models numbered No. 9 and No. 10 are generated using learning data including the minimum temperature, humidity, and rainfall as variables of meteorological data, and the model numbered No. 12 is generated using learning data including the minimum temperature, rainfall, and sunshine duration as variables of meteorological data, respectively.

[0042] Further, the model generation unit 14 may generate a prediction model by performing machine learning using conversion data in which the differences in meteorological data for any past year are associated with the differences in the picking appropriate periods of those years. The prediction model generated in this way has, as input information, the difference in the integrated value of meteorological data between any past year and the target year of the tea field where the tea trees grow, and the information representing the difference in the degree of growth of tea leaves between any past year and the target year as output information.

[0043] Here, the input information input to the prediction model generated by the model generation unit 14 can be the measured value or the forecast value of the meteorological data of the year to be predicted. The measured value of meteorological data is the actually measured value, and the forecast value can be a forecast value such as a 2-week weather forecast or a 50m mesh weather forecast provided by the Japan Meteorological Agency or the like. When using the forecast value as the input information, the degree of growth of tea leaves can be predicted earlier. <Model Selection Unit 15> The model selection unit 15 selects a prediction model suitable for prediction from a model group including a plurality of prediction models generated by the model generation unit 14. The model selection unit 15 selects the optimal model based on rules from the above model group. As the rule-based criteria, the cumulative number of days of meteorological data from the germination date of the target year which is the input information, the AIC criterion, the error (MAE) criterion, etc. can be set. For example, it can be a rule such that after selecting the variables of meteorological data based on the AIC criterion, an appropriate model is selected based on the cumulative number of days and the MAE criterion.

[0044] Regarding the selection of the prediction model by the model selection unit 15, with reference to FIG. 4 described above, FIG. 5 will be referred to for explanation. FIG. 5 is a diagram for explaining the process of selecting a suitable model from a plurality of prediction models in the model generation device 10 according to an aspect of the present invention.

[0045] In the model group shown in FIG. 4, each of the models No. 1 to No. 11 is a model selected for each number of days of accumulating the integrated value of the meteorological data of the target year input to the prediction model. That is, the model No. 1 is selected as the model with the highest accuracy when the integrated value of the meteorological data for 4 days from the germination date is used as the input information. Also, the model No. 11 is selected as the model with the highest accuracy when the integrated value of the meteorological data for 14 days from the germination date is used as the input information.

[0046] As shown in FIGS. 4 and 5, when the input information is the integrated value of the meteorological data for 4 to 10 days from the germination date, the prediction model (4-day model) created using the integrated value of the meteorological data for the past 4 days is the optimal model with the highest accuracy. On the other hand, when the input information is the integrated value of the meteorological data for 11 days from the germination date, the prediction model (11-day model) created using the integrated value of the meteorological data for the past 11 days is the optimal model. Further, when the input information is the integrated value of the meteorological data on and after the 12th day from the germination date, the prediction model (12-day model) created using the integrated value of the meteorological data for the past 12 days is the optimal model. Thus, depending on the number of days elapsed from the germination date in the input information, the variables of the meteorological data that affect the harvesting period can change. Therefore, by selecting an optimal prediction model according to the number of days elapsed from the germination date of the input information, the prediction accuracy can be improved over time. Also, contrary to what might be generally considered, it is not the case that the greater the number of days elapsed or the closer the elapsed days are to the harvesting period, the higher the prediction accuracy of the generated model. Therefore, by selecting an optimal prediction model according to the number of days elapsed from the germination date, accurate predictions can always be made.

[0047] Figure 4 shows the results of Leave One Out cross-validation using the prediction model actually generated and selected by the model generation device 10. As shown in Figure 4, the error (MAE) was 2.96 days for the cumulative value up to the 4th day from the germination date and 2.51 days for the cumulative value up to the 12th day. There is an empirical rule that the quality of tea changes significantly after 3 days from the harvesting date. However, with the prediction model generated and selected by the model generation device 10, it is possible to achieve an accuracy within an error of 3 days, which is essential for prediction.

[0048] The model selection unit 15 selects a prediction model from a plurality of prediction models according to the number of days in a predetermined period from the germination date of the target year, which is the input information. Depending on the number of days from the germination date of the input information, the variables that affect the prediction of the degree of growth of tea leaves can differ. Therefore, by selecting an optimal model according to the number of days from the germination date of the input information, the degree of growth of tea leaves can be accurately predicted.

[0049] The model selection unit 15 stores in the storage device 40 the prediction model data in which the result of selecting a suitable prediction model for each number of days from the germination date is associated with the selected prediction model. An example of such prediction model data is the data shown in Figure 4.

[0050] According to the model generation device 10, by generating a plurality of prediction models and selecting a prediction model suitable for prediction, accurate prediction can be realized. Further, according to the model generation device 10, the cumulative number of days of past weather data used for generating the prediction model is updated daily, and by selecting an optimal prediction model according to the number of days from the germination date of the input information, the optimal prediction model is updated daily. Therefore, the degree of growth of tea leaves can be accurately predicted.

[0051] <Modification example of the model generation device 10> The model generation device 10 can also generate a prediction model for another field using the prediction model created for one field. That is, the model generation device 10 includes an acquisition step of acquiring information representing the degree of growth of tea leaves output by inputting past weather data in another field into a prediction model generated in one field, a calculation step of calculating the difference between the degree of growth of tea leaves acquired in the acquisition step and the degree of growth of past tea leaves in another field, and a re-learning step of generating a prediction model for another field by re-learning the prediction model generated in one field using the information representing the calculated difference.

[0052] In this modification example, the model generation device 10 uses, as an example, the prediction model A for field A generated using data associating the measured values of past weather data in field A with the appropriate tea leaf picking period of that year or its conversion data as learning data. The model generation unit 14 of the model generation device 10 acquires, for example, the prediction model A stored in the storage device 40. Then, the model generation unit 14 inputs the measured values of past weather data in field B as input information into the prediction model A and acquires information representing the appropriate tea leaf picking period as output information.

[0053] Next, the model generation unit 14 calculates the difference between the appropriate tea leaf picking period obtained using the prediction model A and the actual appropriate tea leaf picking period in the past in field B. The model generation unit 14 re-learns the prediction model A using the data representing the calculated difference as learning data to generate a prediction model B.

[0054] By using the prediction model created for one field to generate the prediction models for other fields, it is possible to generate a prediction model that more accurately predicts the degree of tea leaf growth.

[0055] (Prediction device) The prediction device 20 predicts the growth of tea leaves using the prediction model generated by the model generation device 10. The prediction device 20 uses a prediction model in which the integrated value of meteorological data for a predetermined period from the budding date of the tea tree in the target year is the input information and the information representing the degree of tea leaf growth is the output information, and predicts the degree of tea leaf growth in the target year. The prediction device 20 can predict the appropriate picking period of the tea leaves in the target year from the integrated value of the meteorological data for a predetermined period from the budding date of the tea tree in the target year.

[0056] The prediction device 20 includes a control unit 21. The control unit 21 comprehensively controls each part of the prediction device 20 and is realized by, for example, a processor and a memory. In this example, the processor accesses a storage (not shown), loads a program (not shown) stored in the storage into the memory, and executes a series of instructions included in the program. Thereby, each part of the control unit 21 is configured. As each of these parts, the control unit 21 includes a data acquisition unit 22, a data conversion unit 23, a model acquisition unit 24, and a prediction unit 25.

[0057] The data acquisition unit 22 acquires meteorological data for a predetermined period from the budding date of the tea tree in the target year. The meteorological data of the target year acquired by the data acquisition unit 22 may be the actually measured values or the forecast values provided by the Japan Meteorological Agency or the like. The data acquisition unit 22 reads out the meteorological data for a predetermined period from the budding date of the tea tree in the target year from the storage device 40 based on an input signal representing an instruction to start prediction from the input device 30. Further, the data acquisition unit 22 may acquire the meteorological data for a predetermined period from the budding date of the tea tree in the target year input via the input device 30. The data acquisition unit 22 outputs the acquired meteorological data to the data conversion unit 23.

[0058] The data conversion unit 23 converts the meteorological data acquired by the data acquisition unit 22 into input information to be input to the prediction model. The conversion of data by the data conversion unit 23 can be executed using a known data transformer or the like. The data conversion unit 13 outputs the converted data to the prediction unit 25.

[0059] The data conversion unit 23 calculates an integrated value obtained by integrating meteorological data for a predetermined period from the budding date of the tea tree in the target year. The data conversion unit 23 generates input information associating the calculated integrated value with the number of integrated days.

[0060] Further, the data conversion unit 23 may calculate the difference between the integrated value of the meteorological data of an arbitrary year and the integrated value of the meteorological data of the target year, and generate input information.

[0061] The model acquisition unit 24 acquires a prediction model generated by the model generation device 10, in which the integrated value of meteorological data for a predetermined period from the budding date of the tea tree in the target year is the input information, and the information representing the degree of growth of the tea leaves is the output information. The model acquisition unit 24 may acquire the prediction model generated by the model generation device 10 and stored in the storage device 40. The model acquisition unit 24 acquires the optimal prediction model selected according to the number of days from the budding date of the input information. The model acquisition unit 24 outputs the acquired prediction model to the prediction unit 25.

[0062] When the input information is the difference between the integrated value of the meteorological data of an arbitrary year and the integrated value of the meteorological data of the target year, the model acquisition unit 24 acquires a prediction model in which the difference between the integrated values of the meteorological data of the arbitrary year and the target year is the input information, and the information representing the difference in the degree of growth of the tea leaves between the arbitrary year and the target year is the output information.

[0063] The prediction unit 25 uses a prediction model in which the integrated value of meteorological data for a predetermined period from the budding date of the tea plants in the target year acquired by the model acquisition unit 24 is the input information and the information representing the degree of growth of the tea leaves is the output information, to predict the degree of growth of the tea leaves in the target year. The prediction unit 25 inputs the input information converted by the data conversion unit 23 into the prediction model, and acquires information representing the degree of growth of the tea leaves in the target year as the output information. The prediction unit 25 outputs the acquired information representing the degree of growth of the tea leaves in the target year to the output device 50. The prediction unit 25 may store the acquired information representing the degree of growth of the tea leaves in the target year in the storage device 40.

[0064] The prediction unit 25 may use a prediction model in which the difference in the integrated value of meteorological data between an arbitrary year acquired by the model acquisition unit 24 and the target year is the input information and the information representing the difference in the degree of growth of the tea leaves between the arbitrary year and the target year is the output information, to predict the information representing the difference in the degree of growth of the tea leaves between the arbitrary year and the target year. The prediction unit 25 inputs the input information, which is the difference between the integrated value of the meteorological data of an arbitrary year and the integrated value of the meteorological data of the target year, into the prediction model, and acquires information representing the difference in the degree of growth of the tea leaves between the arbitrary year and the target year as the output information. Based on such output information, it is also possible to obtain a prediction result of the degree of growth of the tea leaves in the target year.

[0065] According to the prediction device 20, by using the prediction model generated by the model generation device 10, it is possible to predict the degree of growth of the tea leaves in that year only by inputting the meteorological data of the target year for prediction. In addition, since the prediction model used by the prediction device 20 is the optimal prediction model selected according to the integrated number of days of meteorological data from the budding date of the target year, which is the input information, it is possible to accurately predict the degree of growth of the tea leaves. In this way, it is possible to select and predict the optimal prediction model for each integrated number of days of meteorological data from the budding date of the target year, which is the input information.

[0066] (Model Generation Process) The flow of the prediction model generation process (prediction model generation method) by the model generation device 10 will be described with reference to FIG. 6. FIG. 6 is a flowchart showing an example of the generation process executed by the model generation device 10 according to an aspect of the present invention. As shown in FIG. 6, first, the data acquisition unit 12 acquires data for generating a prediction model including measured values of past weather data and the picking suitable period of that year (step S1). Next, the data conversion unit 13 converts the data acquired by the data acquisition unit 12 into data suitable for generating a prediction model (step S2).

[0067] The model generation unit 14 uses the converted data converted by the data conversion unit 13 as learning data, performs machine learning, and generates a plurality of prediction models according to the combination of weather data and the accumulated number of days (step S3). Next, the model selection unit 15 selects a prediction model suitable for prediction from a model group including the plurality of prediction models generated by the model generation unit 14 (step S4). Then, the model selection unit 15 stores the selected prediction model in the storage device 40 (step S5) and ends the model generation process.

[0068] (Prediction process) The flow of the prediction process (prediction method) by the prediction device 20 will be described with reference to FIG. 7. FIG. 7 is a flowchart showing an example of the prediction process executed by the prediction device 20 according to an aspect of the present invention. As shown in FIG. 7, first, the data acquisition unit 22 acquires weather data for a predetermined period from the budding date of the tea tree in the target year (step S11). Next, the data conversion unit 23 converts the weather data acquired by the data acquisition unit 22 into input information to be input to the prediction model (step S12).

[0069] The model acquisition unit 24 acquires a prediction model that is generated by the model generation device 10 and selected according to the number of accumulated days of weather data from the germination date included in the input information (step S13). Then, the prediction unit 25 inputs the converted input information into the prediction model and acquires a predicted value of the harvesting suitable period for the target year output (step S14). The prediction unit 25 outputs the acquired predicted value to the output device 50 as a prediction result (step S15), and ends the prediction process.

[0070] 〔Example of implementation by software〕 The functions of the model generation device 10 and the prediction device 20 (hereinafter referred to as "devices") are programs for causing a computer to function as the devices, and can be realized by programs for causing a computer to function as each control block of the devices (especially each part included in the control unit 11 and the control unit 21).

[0071] In this case, the above device includes a computer having at least one control device (for example, a processor) and at least one storage device (for example, a memory) as hardware for executing the above program. By executing the above program with this control device and storage device, each function described in the above embodiments is realized.

[0072] The above program may be recorded on one or more computer-readable recording media, not temporarily. This recording medium may or may not be provided in the above device. In the latter case, the above program may be supplied to the above device via any wired or wireless transmission medium.

[0073] Also, part or all of the functions of the above control blocks can also be realized by a logic circuit. For example, an integrated circuit in which a logic circuit functioning as each of the above control blocks is formed is also included in the scope of the present invention. In addition to this, for example, it is also possible to realize the functions of the above control blocks by a quantum computer.

[0074] In addition, each process described in each of the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI may operate in the above control device, or may operate in another device (for example, an edge computer or a cloud server, etc.).

[0075] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope indicated in the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

Example

[0076] 〔Example 1: Construction of a prediction model based on the integrated value of meteorological data from the germination date〕 A prediction model was constructed using the integrated value of meteorological data from the germination date. The constructed prediction model is shown in FIG. 4. For the construction of the prediction model shown in FIG. 4, daily meteorological data (average temperature (°C), minimum temperature (°C), maximum temperature (°C), average humidity (%), precipitation (mm), sunshine hours (hr)) measured at the tea research institute field in Kyoto Prefecture from 2000 to 2019, as well as data of eight variables of the germination date (day) and the appropriate picking date (day) obtained during the same period were used. These data are data for 13 years (2000 - 2006, 2009, 2010, 2013, 2015, 2016, and 2019) where there are no missing values in the above eight variables up to the 14th day with the germination date of each year as the starting date.

[0077] In the construction of the prediction model, as explanatory variables, the values of the above six meteorological data with the germination date as the starting date were input daily, and further, as the target variable, the difference between the number of days from the germination date to the appropriate picking date and the number of days input daily with the explanatory variables was input. With the condition that the average error is within 3 days, the prediction model was constructed for each input number of days by the method described in the embodiment. As shown in Fig. 4, until the 10th day starting from the germination date, the model created based on the data up to the 4th day (4th-day model) was adopted, and by using the respective integrated values from the 4th day to the 10th day as input data for prediction, it was possible to make a prediction with a Leave One Out (LOO) average error of 2.96 days (No. 1 to No. 7).

[0078] Also, for the 11th day starting from the germination date, by adopting the 11th-day model and using the integrated value up to the 11th day as input data for prediction, it was possible to make a prediction with an LOO average error of 2.57 days (No. 8). Furthermore, since the AIC and the LOO average error started to increase on the 14th day starting from the germination date (No. 11), after the 12th day, by adopting the 12th-day model and using the integrated value up to the 12th or 13th day as input data for prediction, it was possible to make a prediction with an LOO average error of 2.51 days (No. 9 to No. 11).

[0079] Thus, it cannot be said that the accuracy necessarily improves when the number of days to be integrated is larger. According to the present invention, it has been shown that by comparing appropriate prediction models for the days to be integrated, an optimal model for the day to be predicted can be obtained.

[0080] 〔Example 2: Construction of a prediction model using differential meteorological data with meteorological data of an arbitrary day〕 A prediction model was constructed using the integrated value of meteorological data with an arbitrary day as the starting date instead of the germination date. For the construction of the prediction model, among the data from 2000 to 2019 similar to Example 1, six variable data of the average temperature (°C), minimum temperature (°C), maximum temperature (°C), average humidity (%), precipitation (mm), and the data of the picking suitable date (days) obtained during the same period were used. These data are data for 14 years (2000 to 2006, 2009, 2010, 2012, 2013, 2015, 2016, and 2019) in the period from March 1 to April 30 of each year without missing values in the above six variables. Also, as preprocessing, the difference between the above six variables was taken between the year to be predicted and the most recent past year, and used as input data (the data for 2000 was obtained as the difference from the data for 1999).

[0081] For the construction of the prediction model, data on the appropriate harvesting date that had undergone the same preprocessing as above was used as the target variable, and the above five meteorological data that had undergone the same preprocessing as above was used as the explanatory variables. As an example, in the prediction model that used the integrated values for 14 days up to March 24 with March 11 as the starting date as the input data, the LOO average error was 3.033 days, which was slightly more accurate than the target error within 3 days. On the other hand, in the prediction model that used the integrated values for 46 days up to April 25 as the input data, the LOO average error was 2.059 days.

[0082] In this way, a prediction model can be constructed even when using the difference in the integrated values with an arbitrary day as the starting date as the input data, and the prediction accuracy of the appropriate harvesting date can be improved by using the model with the highest accuracy in the number of days up to the prediction date.

Explanation of symbols

[0083] 10 Model generation device (generation device) 14 Model generation unit 20 Prediction device 25 Prediction unit 100 Prediction system

Claims

1. A method for generating a prediction model for predicting tea leaf growth, comprising: By carrying out machine learning using the accumulated value of past weather data for a certain period of time for the fields where tea plants are grown and the past growth rate of tea leaves as learning data, A generation method including a generation step of generating multiple prediction models in which the input information is the accumulated value of meteorological data for a specified period from any day of a target year, and the output information is information representing the degree of growth of tea leaves, in accordance with the accumulated number of days of the meteorological data included in the learning data.

2. The generating method according to claim 1 , wherein in the generating step, a plurality of the prediction models are generated according to combinations of the weather data included in the learning data and an accumulated number of days.

3. The method of claim 1 or 2, further comprising the step of selecting a forecast model from the plurality of forecast models based on a number of days in a predetermined period from the arbitrary date in a year of interest.

4. A method for generating a prediction model for predicting tea leaf growth, comprising: By carrying out machine learning using as learning data the difference between any year in the integrated value of past weather data of a field where tea plants are grown for a predetermined period and the difference between the growth degree of tea leaves for the same year, The method includes a generation step of generating a predictive model in which the difference in the integrated value between any past year and a target year in a field growing tea plants is input information, and information representing the difference in the degree of tea leaf growth between the any past year and the target year is output information.

5. The method according to claim 1 , wherein the weather data for the target year are measured values ​​or forecast values.

6. The method according to claim 1 , wherein the prediction model is a model for predicting the leaf opening stage of the tea leaves.

7. an acquisition step of acquiring information representing the degree of growth of tea leaves by inputting past meteorological data of another farm field into the prediction model generated in the one farm field; a calculation step of calculating a difference between the degree of growth of the tea leaves acquired in the acquisition step and the degree of growth of the tea leaves in the past in the other farm field; The method according to claim 1 , further comprising a re-learning step of re-learning the prediction model generated in the one field using information representing the calculated difference, thereby generating a prediction model for the other field.

8. A prediction method comprising a step of predicting the growth of tea leaves using a predictive model generated by the generation method according to any one of claims 1 to 7.

9. A device for generating a prediction model for predicting tea leaf growth, By carrying out machine learning using the accumulated value of past weather data for a certain period of time for the fields where tea plants are grown and the past growth rate of tea leaves as learning data, A generation device having a generation unit that generates multiple predictive models in which the input information is the accumulated value of meteorological data for a specified period from any day of a target year, and the output information is information representing the degree of growth of tea leaves, in accordance with the accumulated number of days of the meteorological data.

10. The generating device according to claim 9 , wherein the generating unit generates a plurality of prediction models for each integrated number of days of the integrated value included in the learning data.

11. A device for generating a prediction model for predicting tea leaf growth, By carrying out machine learning using as learning data the difference between any year in the integrated value of past weather data of a field where tea plants are grown for a predetermined period and the difference between the growth degree of tea leaves for the same year, A generation device having a generation unit that generates a predictive model in which the difference in the integrated value between any past year and a target year in a field where tea trees grow is input information, and information representing the difference in the degree of tea leaf growth between the any past year and the target year is output information.

12. A prediction device comprising a prediction unit that predicts the growth of tea leaves using a prediction model generated by the generation device according to any one of claims 9 to 11.

13. A prediction system for predicting the growth of tea leaves, comprising: a generating device according to any one of claims 9 to 11; and a prediction device according to claim 12.

Images