KR102634529B1 - Agricultural Price Prediction Apparatus Using Multi-Step Time Series Forecasting and Method for Predicting Agricultural Product Price Using the Same - Google Patents

Abstract

This specification relates to a method for predicting prices of agricultural products using an agricultural product price prediction device. The agricultural product price prediction method using the agricultural product price prediction device includes selecting one time series prediction model from a plurality of time series prediction models; Reading price data for a plurality of agricultural products from a database and performing preprocessing for learning in the selected time series prediction model; Generating at least one price data cluster by clustering the price data using a clustering model that groups at least one agricultural product into agricultural products; A plurality of agricultural products that tune hyper-parameters for the selected time series prediction model, learn the tuned time series prediction model for each of the at least one price data cluster, and calculate the predicted price for each agricultural product at a future point in time. generating price prediction models; Performing a performance evaluation for each agricultural product among the plurality of agricultural product price prediction models and determining the agricultural product price prediction model with the best performance as the final agricultural product price prediction model for the agricultural product; and calculating a predicted price for the target agricultural product on the target date using a final agricultural product price prediction model corresponding to the target agricultural product.

Description

Agricultural Price Prediction Apparatus Using Multi-Step Time Series Forecasting and Method for Predicting Agricultural Product Price Using the Same}

This specification relates to an agricultural product price prediction device using a multi-level time series forecasting method and a method for predicting agricultural product prices using the same.

Recently, the price volatility of agricultural products has become more volatile due to the rapidly changing climate and international situation, and if a natural disaster occurs, the prices may skyrocket uncontrollably. Stabilizing prices of agricultural products is an important issue being addressed at the national level because sudden price changes have a significant impact on both producers and consumers. Producers must carefully consider the amount and area planted each year, which are closely related to price. However, it is difficult to control production depending on the harvest period, causing prices to repeatedly rise and fall. In addition, there are various reasons why supply and demand are out of balance, such as climate, pests, difficult storage, and inelastic consumption patterns. Unstable supply and demand of agricultural products affects prices, affecting producers and consumers, and makes it difficult to revitalize agricultural trade.

In order to respond to this problem, the government distributed a manual for regulating the supply and demand of agricultural products, which consists of three levels: Caution, Warning, and Serious. Additionally, as part of price observation, the 'Agricultural Observation Headquarters' established by the Korea Rural Economic Institute collects various statistical data, We analyze and publish monthly observation reports every month. Korea Agro-Fisheries and Food Trade Corporation aT has been discovering models that can predict agricultural product prices through contests starting in 2020, and Gyeongsangnam-do is providing social support for agricultural product price forecasting, such as providing price forecasting information through the “Main Agricultural Product Price Prediction System.” Demand is on the rise. However, these services only provide price predictions for at most 7 days out. Short-term price prediction information can be helpful in managing the shipment of agricultural products. However, for farmers who want to decide which crops to sow or fintech operators who want to provide financial services based on agricultural product prices, mid- to long-term agricultural product price forecast information is needed.

Korean Patent Publication No. 10-2137583 discloses a method for predicting the price and sales volume of agricultural products using LSTM.

The content described above is only intended to help understand the background technology of the technical ideas of the present invention, and therefore, it cannot be understood as content corresponding to prior art known to those skilled in the art of the present invention.

Korean Patent Publication No. 10-2137583, 2020.07.24.

The purpose of this specification is to solve the above-mentioned problems, and an embodiment of this specification aims to provide a model for predicting mid- to long-term prices of agricultural products using an agricultural product price prediction model based on a multi-level time series forecasting method.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned can be clearly understood by those skilled in the art from the description below.

This specification presents a method for predicting prices of agricultural products using an agricultural product price prediction device. The agricultural product price prediction method includes selecting one time series prediction model among a plurality of time series prediction models; Reading price data for a plurality of agricultural products from a database and performing preprocessing for learning in the selected time series prediction model; Generating at least one price data cluster by clustering the price data using a clustering model that groups at least one agricultural product into agricultural products; A plurality of agricultural products that tune hyper-parameters for the selected time series prediction model, learn the tuned time series prediction model for each of the at least one price data cluster, and calculate the predicted price for each agricultural product at a future point in time. generating price prediction models; Performing a performance evaluation for each agricultural product among the plurality of agricultural product price prediction models and determining the agricultural product price prediction model with the best performance as the final agricultural product price prediction model for the agricultural product; And it may include calculating a predicted price for the target agricultural product on the target date using a final agricultural product price prediction model corresponding to the target agricultural product.

The agricultural product price prediction method and other embodiments using the agricultural product price prediction device may include the following features.

Depending on the embodiment, the clustering model may cluster the price data so that agricultural product items that satisfy a predetermined tuning pattern for hyper-parameter tuning are in the same cluster when applying the clustered price data to the selected time series prediction model. You can.

Depending on the embodiment, the clustering model may cluster the price data into the number of clusters that produce the best performance evaluation result when applying the clustered price data to the selected time series prediction model.

Depending on the embodiment, hyper-parameter tuning for the selected time series prediction model may be performed by changing the value of each hyper-parameter within a predetermined category and training the selected time series prediction model for each of the at least one price data cluster. , It may be a process of determining the value of the hyper-parameter that allows the selected time series prediction model to produce the best performance as the value of each hyper-parameter.

According to the embodiment, the final agricultural product price prediction model may include a first agricultural product price prediction model generated for each future time point based on the selected time series prediction model, and a prediction result of the previous time point by the selected time series prediction model at the next time point. A second agricultural product price prediction model, which is input to the time series prediction model for prediction, is created by learning the prediction results of the time series prediction model predicting the price at the previous point in time by including it in the input data of the time series prediction model predicting the price at the next point in time. It may be an agricultural product price prediction model selected among a third agricultural product price prediction model and a fourth agricultural product price prediction model that calculates predicted prices for each agricultural product at all future points in time using a single time series prediction model.

According to the embodiment, the step of reading price data for the plurality of agricultural products from a database and performing preprocessing for learning in the selected time series prediction model includes loading the price data from the database and removing missing data in advance. Reinforcing with the determined estimation method; And it may include configuring the price data augmented with the missing data into a format for learning in the selected time series prediction model.

Meanwhile, this specification presents a computer program stored in a computer-readable recording medium, and the computer program is combined with hardware to execute each step included in the agricultural product price prediction method by the agricultural product price prediction device. .

On the other hand, this specification presents an agricultural product price prediction device that performs an agricultural product price prediction method. The agricultural product price prediction device includes a communication unit that communicates with an external agricultural product price data providing server to collect price data for a plurality of agricultural products; A storage unit that stores the collected price data in a database; and a control unit functionally connected to the communication unit and the storage unit, wherein the control unit selects one time series prediction model among a plurality of time series prediction models and reads the price data for a plurality of agricultural products from the database. Perform preprocessing for learning in the selected time series prediction model, cluster the price data with a clustering model that groups at least one agricultural product by agricultural product, and generate at least one price data cluster, A plurality of agricultural product prices that tune hyper-parameters for a selected time series prediction model, learn the tuned time series prediction model for each of the at least one price data cluster, and calculate a predicted price for each agricultural product at a future point in time. Generate prediction models, perform a performance evaluation for each agricultural product among the plurality of agricultural product price prediction models, and determine the agricultural product price prediction model with the best performance as the final agricultural product price prediction model for the agricultural product. , the predicted price for the target agricultural product on the target date can be calculated using the final agricultural product price prediction model corresponding to the target agricultural product.

The agricultural product price prediction device and other embodiments may include the following features.

Depending on the embodiment, the clustering model may cluster the price data so that agricultural product items that satisfy a predetermined tuning pattern for hyper-parameter tuning are in the same cluster when applying the clustered price data to the selected time series prediction model. You can.

Depending on the embodiment, the clustering model may cluster the price data into the number of clusters that produce the best performance evaluation result when applying the clustered price data to the selected time series prediction model.

Depending on the embodiment, hyper-parameter tuning for the selected time series prediction model may be performed by changing the value of each hyper-parameter within a predetermined category and training the selected time series prediction model for each of the at least one price data cluster. , It may be a process of determining the value of the hyper-parameter that allows the selected time series prediction model to produce the best performance as the value of each hyper-parameter.

According to the embodiment, the final agricultural product price prediction model may include a first agricultural product price prediction model generated for each future time point based on the selected time series prediction model, and a prediction result of the previous time point by the selected time series prediction model at the next time point. A second agricultural product price prediction model, which is input to the time series prediction model for prediction, is created by learning the prediction results of the time series prediction model predicting the price at the previous point in time by including it in the input data of the time series prediction model predicting the price at the next point in time. It may be an agricultural product price prediction model selected among a third agricultural product price prediction model and a fourth agricultural product price prediction model that calculates predicted prices for each agricultural product at all future points in time using a single time series prediction model.

Depending on the embodiment, the control unit reads price data for a plurality of agricultural products from a database, performs preprocessing for learning in the selected time series prediction model, and retrieves the price data from the database to remove missing data. The price data augmented with a predetermined estimation method and the missing data augmented can be configured into a format for learning in the selected time series prediction model.

The embodiment disclosed in this specification has the effect of providing a model for predicting the price of agricultural products using an agricultural product price prediction model based on a multi-level time series prediction method.

Meanwhile, the effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below. You will be able to.

The following drawings attached to this specification illustrate preferred embodiments of the present invention, and serve to further understand the technical idea of the present invention along with specific details for carrying out the invention. Therefore, the present invention is described in such drawings. It should not be interpreted as limited to the specific details.
Figure 1 is a diagram briefly explaining the configuration of a framework for deriving a method for predicting prices of agricultural products according to an embodiment.
Figure 2 shows the results of descriptive statistics analysis of prices for each variety of agricultural products during a specific period.
Figures 3 to 5 show prediction results of various price prediction models in the agricultural product price prediction method.
Figure 6 shows the best combination of strategies and forecasting techniques for each period of agricultural product varieties in a framework for an agricultural product price prediction method according to an embodiment.
Figure 7 shows a flow chart explaining a price prediction method according to an embodiment.
Figure 8 explains a brief configuration of an agricultural product price prediction device that performs an agricultural product price prediction method according to an embodiment.
Figure 9 is a block diagram of an AI device included as part of an agricultural product price prediction device according to an embodiment.

The technology disclosed in this specification can be applied to price prediction of agricultural products. However, the technology disclosed in this specification is not limited to this and can be applied to all devices and methods to which the technical idea of the technology can be applied.

It should be noted that the technical terms used in this specification are only used to describe specific embodiments and are not intended to limit the spirit of the technology disclosed in this specification. In addition, the technical terms used in this specification, unless specifically defined in a different way in this specification, should be interpreted as meanings generally understood by those skilled in the art in the field to which the technology disclosed in this specification belongs. It should not be interpreted in a very comprehensive sense or in an excessively reduced sense. In addition, if the technical term used in this specification is an incorrect technical term that does not accurately express the idea of the technology disclosed in this specification, it is a technical term that can be correctly understood by a person with ordinary knowledge in the field to which the technology disclosed in this specification belongs. It should be understood and replaced with . Additionally, general terms used in this specification should be interpreted as defined in the dictionary or according to the context, and should not be interpreted in an excessively reduced sense.

Terms containing ordinal numbers, such as first, second, etc., used in this specification may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component without departing from the scope of the present invention.

Hereinafter, embodiments disclosed in the present specification will be described in detail with reference to the attached drawings. However, identical or similar components will be assigned the same reference numerals regardless of the reference numerals, and duplicate descriptions thereof will be omitted.

Additionally, when describing the technology disclosed in this specification, if it is determined that a detailed description of a related known technology may obscure the gist of the technology disclosed in this specification, the detailed description will be omitted. In addition, it should be noted that the attached drawings are only intended to facilitate easy understanding of the spirit of the technology disclosed in this specification, and should not be construed as limiting the spirit of the technology by the attached drawings.

Throughout the specification, a device or terminal includes a communication terminal or communication device capable of wired or wireless communication with a server or other device. The form of the device or terminal may be various, such as a mobile phone, smartphone, smart pad, laptop computer, desktop computer, smart TV, wearable device, mirror-shaped display device, smart mirror, etc. Wearable devices can be diverse, such as watch-type terminals, glass-type terminals, HMDs, etc. Additionally, the terminal is not limited to this form and can be implemented with various electronic devices.

Multi-step Time Series Forecasting

In this specification, a multi-step time series forecasting approach (Multi-step Time Series Forecasting) is applied to predict agricultural product prices at various future points in time. Commonly used approaches for multi-step time series forecasting include Direct Multi-step Time Series Forecasting and Recursive Multi-step Time Series Forecasting. Among them, the recursive multi-step time series forecasting approach is It is generally the most used. However, the recursive multi-step time series forecasting approach uses the next day's forecast value to predict the value two days later and includes this value to predict the next day's price, so the forecast error increases in the mid- to long-term, that is, as the time to predict becomes farther away, the forecast error increases. There is no choice but to do so. Accordingly, in the embodiment of this specification, to build a mid- to long-term agricultural product price prediction model, (a) Direct Multi-step Time Series Forecasting, (b) Recursive Multi-step Time Series Forecasting ) strategy, (c) Direct-Recursive Hybrid Multi-step Time Series Forecasting strategy, and (d) Multiple Outputs strategy are used to predict mid- to long-term agricultural product prices. However, due to the error problem of the above-mentioned recursive multi-level time series forecasting approach, three strategies other than this are used to predict mid- to long-term agricultural product prices.

(a) Direct Multi-step Time Series Forecasting strategy

The direct multi-level time series forecasting approach strategy is defined as an approach that independently builds a forecast model for each time point, as shown in Equation (1) below.

This approach shows acceptable performance, but requires the development of multiple prediction models, consuming a lot of computing resources, and predicting results at each time point are generated independently from each other, resulting in poor connectivity.

(b) Recursive Multi-step Time Series Forecasting strategy

The recursive multi-stage time series forecasting approach refers to a strategy of predicting the value at a specific point in time by recursively using the prediction results of the previous stage to predict the next stage, as expressed in Equation (2) below.

This approach has the advantage of being able to produce forecast results for multiple future points in time with a single model. However, as predictions are made far into the future, that is, as k increases, the prediction error difference accumulates, and a clear decrease in performance appears. In the embodiment of this specification, it was confirmed that errors increased rapidly as a result of the pilot experiment, so it was excluded from the comparison strategy.

(c) Direct-Recursive Hybrid Multi-step Time Series Forecasting strategy

The direct-recursive hybrid multi-step time series forecasting approach is a hybrid strategy multi-step time series forecasting method that combines the advantages of the direct multi-step time series forecasting approach and the recursive multi-step time series forecasting approach. If you want to obtain the predicted value at time t+2, first create a model to obtain the predicted value at time t+1, then add the predicted value through this model to learn the model to obtain the predicted value at time t+2. It refers to the method of obtaining the result, and the formula is as formula (3) below.

(d) Multiple Outputs Strategy

The Multiple Outputs strategy refers to a strategy that can calculate the entire future forecast value at once. Representative analysis techniques that can be applied to this strategy include LSTM and GRU, and the equation is shown in Equation (4) below.

While the multi-output method can obtain predicted values up to the desired point with a single model, it has the disadvantage of making the model overly complex, learning takes a long time, and a large amount of training data is required to avoid overfitting.

Below, among the four multi-step time series forecasting approach strategies described above, price forecast modeling is performed focusing on the remaining three strategies, excluding the Recursive Multi-step Time Series Forecasting strategy, which amplifies errors.

Figure 1 is a diagram briefly explaining the configuration of a framework for deriving a method for predicting prices of agricultural products according to an embodiment.

Figure 2 shows the results of descriptive statistics analysis of prices for each variety of agricultural products during a specific period.

Referring to Figure 1, the artificial neural network learning model according to the embodiment includes the agricultural product variety (onion, cucumber, cabbage), period, prediction strategy (direct strategy, hybrid strategy, multiple output strategy), and prediction technique (MLP, LGBM, LSTM). , GRU), etc. are combined to derive the optimal price prediction model for each agricultural product type and time point.

MAPE (Mean Absolute Percentage Error) is used as a performance evaluation indicator for the price prediction model. MAPE is a comparison of the error between predicted values and actual measurements, and the formula is as equation (5) below. The lower this number, the better the model’s performance. Because there is a large variation in unit price for each agricultural product variety, MAPE was selected as the evaluation index for comprehensive comparison.

Below, the method for predicting prices of agricultural products shown in FIG. 1 will be described in detail.

Price prediction modeling according to the embodiment was conducted in Python version 3.6, using libraries such as Google Tensorflow, matplotlib, and statsmodels. Price prediction modeling was conducted on a total of three agricultural products (onion, radish, and cabbage). Agricultural price data required for model learning and prediction were collected from Nongnet. Using data from January 3, 2014 to December 31, 2021, there are five time points in total: 30 days (Monday), 90 days (quarterly), 180 days (semi-annual), 270 days, and 365 days (one year). Future price prediction was performed. Missing values among the collected daily data were supplemented using the pandas module and asfreq function provided in Python.

The prices of agricultural products differ for each variety, so caution is needed in comparative analysis. Specifically, the descriptive statistical analysis results of the prices for each variety are shown in Table 1 in Figure 2. The training data set is price data for each variety from January 3, 2014 to December 31, 2020. Since the agricultural harvest cycle is one year, the test data set is from January 1, 2021 to December 31, 2021 to check the pattern of price changes during the year from spring to winter. It was set as price data.

In the embodiment, how effective the Direct prediction strategy and the Hybrid prediction strategy are compared to Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), which are common time series analysis techniques used in the Multiple Outputs strategy. I looked. At this time, MLP (Multi-Layer Perceptron) and LGBM (Light Gradient Boosting Machine) were used in the direct and hybrid prediction strategies.

Figures 3 to 5 show prediction results of various price prediction models in the agricultural product price prediction method.

Table 2 in FIG. 3 shows the prediction results of the price prediction model using the LGBM analysis technique, and Table 3 in FIG. 4 shows the prediction results of the price prediction model using the MLP analysis technique. As a result of examining the predicted prices at 5 time points for 3 varieties, it can be seen that the Direct prediction strategy is a more appropriate strategy for onions, and the Hybrid prediction strategy is more appropriate for cucumbers and cabbage. In the MLP model, the Hybrid prediction strategy for onions and the Direct prediction strategy for cucumbers and cabbage show lower MAPE on average. In particular, in the case of onions, it can be seen that excellent performance is shown when used in combination with a hybrid prediction strategy.

Table 4 in Figure 5 shows the prediction results of LSTM and GRU. The prediction value of the Multiple Outputs strategy, which was tested as a comparison group of the Direct prediction strategy and the Hybrid prediction strategy, shows an overall higher MAPE compared to the optimal models of LGBM and MLP. In particular, among the detailed forecast results at five points in time, forecasts after 90 days do not show discriminatory forecasts.

Figure 6 shows the best combination of strategies and forecasting techniques for each period of agricultural product varieties in a framework for an agricultural product price prediction method according to an embodiment.

Referring to Figure 6, Table 5 shows the best combination of strategies and forecasting techniques for each variety period. For onion varieties, the combination of MLP and Hybrid prediction strategy showed the best prediction ability, and for cucumber and cabbage varieties, LGBM and Hybrid strategy showed the highest performance. From a model perspective, among the 15 best combinations, a model combining MLP and LGBM with a hybrid prediction strategy was observed 12 times depending on the breed, showing that the hybrid prediction strategy is generally superior.

In this example, in order to verify how effective the new strategies are in predicting mid- to long-term agricultural product prices compared to the general time series prediction strategy in a recursive form, price prediction was made using the various strategies described above for each price prediction model for three varieties, including onions, cucumbers, and cabbage. proceeded. Referring to Table 5, which summarizes the analysis results, there are differences by breed and period, but overall, the multi-step forecasting method, especially the hybrid forecasting strategy, is relatively better than the multiple outputs strategy. It can be seen that it shows a better combination of performance.

In the above-described embodiment, it can be seen that the strategy and price prediction technique for predicting the best price are applied differently for each type of agricultural product and for each prediction time point. In other words, it can be seen that the agricultural product prediction model that produces the best performance price prediction results may be different at each future point in time to be predicted for each agricultural product variety/item.

The agricultural product prediction model uses various time series prediction models (time series prediction networks) to learn each price data to achieve the best performance depending on the variety of agricultural products, and calculates the hyper-parameter values of the time series prediction network used. It is tuned to produce the best price prediction results through various changes.

Among the time series prediction models used in the above-described embodiments, when creating an agricultural product price prediction model that predicts the price of agricultural products at a specific point in the future through an LSTM network, in order to obtain a high-performance prediction model, the values of the parameters of the LSTM network Settings are important. Parameters such as learning rate, number of iterations of the learning process, mini-batch size, etc. are called hyper-parameters to distinguish them from parameters within the network. . The learning rate represents the distance a value moves at one time on a graph drawn by a cost function that represents the difference between the predicted value and the actual value. If you move too much distance at once, the value of parameter W that minimizes the result of the loss function may be exceeded, and if you move too small a distance, it will take too much time to find and learn, making it inefficient. The learning rate is generally set to 0.001. In deep learning, a mini-batch method is used to determine the size of learning data used at once, unlike the full batch method used in existing machine learning. The full-batch method is a method of creating a learning model by using all data for one learning, while the mini-batch method uses the entire data for learning, but divides the entire data into sub-datasets within one iteration to create a learning model. learn Since the data used for learning cannot represent the entire population being analyzed, randomly dividing it into sub-datasets and learning them multiple times has the advantage of better representing the entire population. There is no absolute mini-batch size, but usually about 1 to 2% of the total training data is set as the mini-batch size.

However, because time series prediction models (time series prediction networks) are diverse and price data is different for each type of agricultural product, hyper-parameters must be tuned for each type of agricultural product in the process of creating an agricultural product price prediction model. In addition, since new price data is added to the agricultural product price prediction model once created every day, it is necessary to prevent price prediction performance from deteriorating by learning and tuning it again over time. This process requires an investment of a lot of time and resources, which leads to an increase in costs. In an embodiment of the present specification, in tuning hyper-parameters among various agricultural product varieties, varieties with the same tuning pattern (tuning value) are grouped into one cluster (may be called a group or cluster) and learned using price data within the same cluster. We seek to solve this problem by tuning time series prediction models at once with the same hyper-parameter values. In other words, agricultural product price data included in the same cluster show the same performance characteristics with the same hyper-parameter values. Therefore, by clustering price data and tuning the hyper-parameters of the time series prediction network, resources required for learning and tuning work can be saved.

Below, an example of a method for predicting prices of agricultural products is explained through simplification of the time series prediction model tuning process through price data clustering.

Figure 7 is a flowchart for explaining a method for predicting prices of agricultural products according to an embodiment.

First, the agricultural product price prediction method according to the embodiment collects price data of agricultural products for price prediction (S710). Price data for agricultural products is collected from the server of an organization that provides agricultural product price data (e.g., Korea Agro-Fisheries and Food Trade Corporation) through API and stored in a database. The stored price data is first processed and the transaction volume (tone) and price (won/kg) data for 10 items and 29 varieties are created and stored in one table. The price data of agricultural products collected from the server of the Korea Agro-Fisheries and Food Trade Corporation consists of real-time meridian price, wholesale price, and retail price information of agricultural products, and each price information includes item name, wholesale market name (in case of meridian, wholesale), and market name. (for retail), transaction date and time, and transaction price.

Next, the agricultural product price prediction method selects one time series prediction model among a plurality of time series prediction models (S720).

Next, the agricultural product price prediction method reads price data for a plurality of agricultural products from the database and performs preprocessing for learning in the selected time series prediction model (S730). Here, the data preprocessing method loads price data from the database, augments the missing data with a predetermined estimation method, and configures the price data augmented with the missing data into a format for learning in the selected time series prediction model. The agricultural product price prediction method calls price data stored in a database and organizes it into a data format for input into a time series prediction model, that is, processing the data format. Price data is organized into the required format for each column. For data on days when there were no transactions (missing values), a row is created, the average value of the values of the day before and after the blank data is calculated, and this is assigned to the blank data. At this time, if the price data is classified by item rather than variety, the data is organized by assigning the average value of the variety price for each item as the average value of the corresponding item. Organized data is organized into a data frame by specifying items as row names and dates as column names. The constructed data frame is converted to a numpy array to facilitate data calculation.

Next, the agricultural product price prediction method generates at least one price data cluster by clustering the price data with a clustering model that groups at least one agricultural product by agricultural product unit (S740), and then uses the selected time series prediction model Tuning hyper-parameters and learning the tuned time series prediction model for each of the at least one price data cluster to generate a plurality of agricultural product price prediction models that calculate the predicted price for each agricultural product at a future point in time. You can do it (S750). When applying the clustered price data to the selected time series prediction model, the clustering model clusters the price data so that agricultural product items that satisfy a predetermined tuning pattern for hyper-parameter tuning are in the same cluster. In other words, the clustering model learns with a time series prediction model selected for the same group of 10 items or 29 varieties of agricultural products into several groups, and tunes the same hyper-parameters to the same values when creating an agricultural product price prediction model. When the price data of these grouped agricultural products is tuned to the same value for the same hyper-parameter, optimal performance is achieved. In addition, the clustering model clusters the price data of agricultural products into the number of clusters that yields the best performance evaluation results when applying the clustered price data to the selected time series prediction model. For example, if there are 12 types of agricultural products, you can make 2 clusters by grouping 6 into one cluster, 3 clusters by grouping 4 into one cluster, and 4 by grouping 3 into one cluster. You can create a colony of dogs. In this case, the performance of the price prediction model varies depending on the number of clusters. You can create a price prediction model by clustering with the number of clusters that produces the best performance, tuning the hyper-parameters of the time series prediction model, and learning the prediction model. . As indicators to evaluate performance, MAPE (Mean Absolute Percentage Error), RMSE (Root Mean Squared Error), or R2 score can be used.

Meanwhile, the hyper-parameter tuning process trains the selected time series prediction model for each of the at least one price data cluster by changing the value of each hyper-parameter within a predetermined category, so that the selected time series prediction model produces the best performance. This is performed by determining the value of the hyper-parameter to be determined from the value of each hyper-parameter. For example, after setting the range of applicable values for each hyper-parameter, tuning the prediction model with a combination of all values that can be selected within the set range to derive the result value, and then deriving the hyper-parameter of the combination that satisfies the predetermined criteria. Automatically finds parameter values.

On the other hand, if new agricultural products are added and the data processing pattern changes significantly after clustering the price data is performed again, tuning of the subsequently tuned agricultural product price prediction model must also be performed again.

The above-mentioned processes can be performed one by one for all of the plurality of time series prediction models, so that an agricultural product price prediction model based on each time series prediction model can be generated.

Here, the plurality of agricultural product price prediction models are derived by applying the previously described multi-step time series forecasting approach and agricultural product price prediction methods and strategies that show the best performance for each agricultural product with reference to Figures 1 to 6. . That is, the plurality of agricultural product price prediction models are agricultural product price prediction models (direct models) generated for each future point in time based on the selected time series prediction model for each agricultural product, and the prediction results at the previous time point by the selected time series prediction model are as follows. An agricultural product price prediction model (recursive model) that is input to a time series prediction model for prediction at a point in time, and the prediction results of a time series prediction model predicting the price at a previous point in time are included in the input data of a time series prediction model for predicting the price at the next point in time. It may include an agricultural product price prediction model (hybrid model) created by learning, and an agricultural product price prediction model (multiple output model) that calculates the predicted price for each agricultural product at all future points in time with a single time series prediction model.

Next, the agricultural product price prediction method performs a performance evaluation for each agricultural product among the plurality of agricultural product price prediction models generated through the above-mentioned processes, and selects the agricultural product price prediction model with the best performance for the corresponding agricultural product. It is decided using the final agricultural product price prediction model (S760). That is, as for the agricultural product price prediction model for a specific agricultural product, the model showing the best performance among the direct model, recursive model, hybrid model, and multi-output model created for the specific agricultural product is determined as the final agricultural product price prediction model.

Finally, the predicted price for the agricultural product on the date for which the predicted price is to be known is calculated using the final agricultural product price prediction model corresponding to the agricultural product for which the price is to be predicted (S770).

Meanwhile, when new price data is added, forecast values can be continuously derived by simply preprocessing the added price data. However, if the accuracy of the derived price prediction value has decreased significantly, that is, if the performance of the price prediction model has decreased significantly, the price prediction model is tuned by renewing tuning or starting from clustering again.

In the above description, the steps, processes or operations S710 to S770 may be further divided into additional steps, processes or operations, or may be combined into fewer steps, processes or operations, depending on the implementation of the present invention. . Additionally, some steps, processes, or operations may be omitted, or the order between steps or operations may be switched, as needed. Additionally, each step or operation included in the above-described agricultural product price prediction method may be implemented as a computer program and stored in a computer-readable recording medium, and each step, process, or operation may be executed by a computer device.

Figure 8 explains a brief configuration of an agricultural product price prediction device that performs an agricultural product price prediction method according to an embodiment.

Referring to FIG. 8, the agricultural product price prediction device 100 according to an embodiment includes a control unit 110, a communication unit 120, and a storage unit 130. The illustrated components are not essential, and an agricultural product price prediction device may be implemented with more components or fewer components. These components may be implemented in hardware or software, or through a combination of hardware and software.

The control unit 110 is functionally connected to the communication unit 120 and the storage unit 130 and can control them.

The control unit 110 collects price data of agricultural products for price prediction. The control unit 110 controls the communication unit 120 to connect to the external agricultural product price data server 200, collects agricultural product price data through data crawling or API, converts it into a database, and stores it in the storage unit 130. do.

The control unit 110 selects one time series prediction model among a plurality of time series prediction models, reads the price data for a plurality of agricultural products from the database, and performs preprocessing for learning in the selected time series prediction model. The control unit 110 reads price data for a plurality of agricultural products from a database, performs preprocessing for learning in the selected time series prediction model, and retrieves the price data from the database to reinforce missing data using a predetermined estimation method. And, the price data augmented with the missing data is configured into a format for learning in the selected time series prediction model.

Additionally, the control unit 110 generates at least one price data cluster by clustering the price data using a clustering model that groups at least one agricultural product into agricultural products. When applying the clustered price data to the selected time series prediction model, the clustering model clusters the price data so that agricultural product items that satisfy a predetermined tuning pattern for hyper-parameter tuning are in the same cluster. Additionally, the clustering model clusters the price data into the number of clusters that yields the best performance evaluation result when applying the clustered price data to the selected time series prediction model.

In addition, the control unit 110 tunes hyper-parameters for the selected time series prediction model, trains the tuned time series prediction model for each of the at least one price data cluster, and predicts each agricultural product at a future point in time. Create multiple agricultural product price prediction models that calculate prices. The control unit 110 changes the value of each hyper-parameter within a predetermined category and trains the selected time series prediction model for each of the at least one price data cluster, so that the selected time series prediction model produces the best performance. The value of the parameter is determined by the value of each hyper-parameter.

In addition, the control unit 110 performs a performance evaluation for each agricultural product among the plurality of agricultural product price prediction models and determines the agricultural product price prediction model with the best performance as the final agricultural product price prediction model for the corresponding agricultural product. do. The final agricultural product price prediction model is an agricultural product price prediction model (direct model) generated for each future time point based on the selected time series prediction model for each agricultural product, and the prediction result at the previous time point by the selected time series prediction model is the prediction at the next time point. An agricultural product price prediction model (recursive model) that is input to the time series prediction model for, is created by training the prediction results of the time series prediction model predicting the price at the previous point in time by including it in the input data of the time series prediction model predicting the price at the next point in time. It can be selected as the best performing agricultural product price prediction model (hybrid model), and an agricultural product price prediction model (multiple output model) that calculates the predicted prices for each agricultural product at all future points in time with a single time series prediction model.

Additionally, the control unit 110 calculates the predicted price for the target agricultural product on the target date using a final agricultural product price prediction model corresponding to the target agricultural product.

Meanwhile, the storage unit 130 can store not only price data but also various time series prediction models. The storage unit 130 can store programs for operating the control unit 110 and temporarily store input/output data. The storage unit 130 is a flash memory type, a hard disk type, a solid state disk type, an SDD type (Silicon Disk Drive type), and a multimedia card micro type. type), card-type memory (e.g. SD or XD memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), EEPROM (electrically erasable programmable memory) It may include at least one type of storage medium among read-only memory (PROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk.

Networks disclosed herein include, for example, wireless networks, wired networks, public networks such as the Internet, private networks, Global System for Mobile communication networks (GSM) networks, general packet wireless networks (General Packet Radio Network (GPRN), Local Area Network (LAN), Wide Area Network (WAN), Metropolitan Area Network (MAN), Cellular Network, Public Switched Telephone Network ; PSTN), Personal Area Network, Bluetooth, Wi-Fi Direct, Near Field communication, Ultra-Wide band, combinations thereof, or any It may be a different network, but is not limited to these.

Figure 9 is a block diagram of an AI device included as part of an agricultural product price prediction device according to an embodiment.

The AI device 20 may include an electronic device including an AI module capable of performing AI processing or a server including the AI module. In addition, the AI device 20 may be included as at least a part of the agricultural product price prediction device 100 shown in FIG. 8 and may be equipped to perform at least part of the AI processing process, such as learning a prediction model.

The AI processing of the AI device 20 may include all operations related to the control of the agricultural product price prediction device 100 shown in FIG. 8 and all operations for price prediction modeling through artificial intelligence learning. For example, the agricultural product price prediction device 100 may process/judge and learn the collected price data set through AI processing, and then perform an operation to predict the future price for a specific agricultural product. The AI device 20 may be included as a component of the control unit 110 of FIG. 8 or may be replaced with the control unit 110.

The AI device 20 may include an AI processor 21, memory 25, and/or a communication unit 27.

The AI device 20 is a computing device capable of learning a neural network, and may be implemented as various electronic devices such as servers, desktop PCs, laptop PCs, tablet PCs, etc., or may be implemented as a single chip. In the present invention, the AI device 20 may be an agricultural product price prediction device implemented in any one of the various electronic devices.

The AI processor 21 can learn a neural network using a program stored in the memory 25. In particular, the AI processor 21 can learn a neural network for recognizing device-related data. Here, a neural network for recognizing device-related data may be designed to simulate the human brain structure on a computer, and may include a plurality of network nodes with weights that simulate neurons of a human neural network. Multiple network modes can exchange data according to each connection relationship to simulate the synaptic activity of neurons sending and receiving signals through synapses. Here, the neural network may include a deep learning model developed from a neural network model. In a deep learning model, multiple network nodes are located in different layers and can exchange data according to convolutional connection relationships. Examples of neural network models include Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Restricted Boltzmann Machine (RBM), and Deep Trust Neural Networks. It includes various deep learning techniques such as (DBN, Deep Belief Networks) and Deep Q-Network, and can be applied to fields such as computer vision (CV), speech recognition, natural language processing, and voice/signal processing. there is.

Meanwhile, the processor that performs the above-described functions may be a general-purpose processor (e.g., CPU), or may be an AI-specific processor (e.g., GPU) for artificial intelligence learning.

The memory 25 can store various programs and data necessary for the operation of the AI device 20. The memory 25 can be implemented as non-volatile memory, volatile memory, flash-memory, hard disk drive (HDD), or solid state drive (SSD). The memory 25 is accessed by the AI processor 21, and reading/writing/modifying/deleting/updating data by the AI processor 21 can be performed. Additionally, the memory 25 may store a neural network model (eg, deep learning model 26) generated through a learning algorithm for data classification/recognition according to an embodiment of the present invention.

Meanwhile, the AI processor 21 may include a data learning unit 22 that learns a neural network for data classification/recognition. The data learning unit 22 can learn standards regarding what learning data to use to determine data classification/recognition and how to classify and recognize data using the learning data. The data learning unit 22 can learn a deep learning model by acquiring learning data to be used for learning and applying the acquired learning data to the deep learning model.

The data learning unit 22 may be manufactured in the form of at least one hardware chip and mounted on the AI device 20. For example, the data learning unit 22 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or may be manufactured as part of a general-purpose processor (CPU) or a graphics processor (GPU) to be used in the AI device 20. It may be mounted. Additionally, the data learning unit 22 may be implemented as a software module. When implemented as a software module (or a program module including instructions), the software module may be stored in a non-transitory computer readable recording medium that can be read by a computer. In this case, at least one software module may be provided by an operating system (Operating System) or an application (application program).

The data learning unit 22 may include a learning data acquisition unit 23 and a model learning unit 24.

The learning data acquisition unit 23 may acquire learning data required for a neural network model for classifying and recognizing data. For example, the learning data acquisition unit 23 may acquire image data and/or sample data for transmitters on road infrastructure to be input into a neural network model as learning data.

The model learning unit 24 can use the acquired training data to train the neural network model to have a judgment standard on how to classify certain data. At this time, the model learning unit 24 can learn a neural network model through supervised learning that uses at least some of the learning data as a judgment standard. Alternatively, the model learning unit 24 can learn a neural network model through unsupervised learning, which discovers a judgment standard by learning on its own using training data without guidance. In addition, the model learning unit 24 can learn a neural network model through reinforcement learning using feedback on whether the result of situational judgment based on learning is correct. Additionally, the model learning unit 24 may learn a neural network model using a learning algorithm including error back-propagation or gradient descent.

When the neural network model is learned, the model learning unit 24 may store the learned neural network model in memory. The model learning unit 24 may store the learned neural network model in the memory of a server connected to the AI device 20 through a wired or wireless network.

The data learning unit 22 further includes a learning data preprocessing unit (not shown) and a learning data selection unit (not shown) to improve the analysis results of the recognition model or save the resources or time required for generating the recognition model. You may.

The learning data preprocessor may preprocess the acquired data so that the acquired data can be used for learning to determine the situation. For example, the learning data preprocessor may process the acquired data into a preset format so that the model learning unit 24 can use the acquired learning data for learning to recognize image data for the transmitter.

Additionally, the learning data selection unit may select data required for learning from among the learning data acquired by the learning data acquisition unit 23 or the learning data pre-processed by the pre-processing unit. The selected learning data may be provided to the model learning unit 24. For example, the learning data selection unit may recognize a specific field among the data sets collected through the network and select only data included in the specific field as learning data.

Additionally, the data learning unit 22 may further include a model evaluation unit (not shown) to improve the analysis results of the neural network model.

The model evaluation unit inputs evaluation data into the neural network model, and when the analysis result output from the evaluation data does not satisfy a predetermined standard, the model learning unit 22 can perform re-training. In this case, the evaluation data may be predefined data for evaluating the recognition model. As an example, the model evaluation unit may evaluate that, among the analysis results of the learned recognition model for the evaluation data, the number or ratio of evaluation data for which the analysis result is inaccurate exceeds a preset threshold, as not satisfying a predetermined standard. .

The communication unit 27 can transmit the results of AI processing by the AI processor 21 to an external electronic device. Here, the external electronic device may be defined as the agricultural product price prediction device 100 of FIG. 8. Meanwhile, the AI device 20 may be implemented by being functionally embedded in the control unit 110 provided in the agricultural product price prediction device 100.

Meanwhile, the AI device 20 shown in FIG. 9 has been described as functionally divided into an AI processor 21, a memory 25, a communication unit 27, etc., but the above-described components are integrated into one module to form an AI module. Please note that it may also be referred to as .

The term “unit” (e.g., control unit, etc.) used herein may mean, for example, a unit including one or a combination of two or more of hardware, software, or firmware. “Part” may be used interchangeably with terms such as unit, logic, logical block, component, or circuit, for example. A “part” may be the minimum unit of an integrated part or a part thereof. “Part” may be the minimum unit or part of one or more functions. The “part” may be implemented mechanically or electronically. For example, a “part” may be an Application-Specific Integrated Circuit (ASIC) chip, Field-Programmable Gate Arrays (FPGAs), or programmable-logic device, known or to be developed in the future, that performs certain operations. It can contain at least one.

At least a portion of the device (e.g., modules or functions thereof) or method (e.g., operations) according to various embodiments may be stored in a computer-readable storage media, e.g., in the form of a program module. It can be implemented with instructions stored in . When the instruction is executed by a processor, the one or more processors may perform the function corresponding to the instruction. Computer-readable media includes all types of recording devices that store data that can be read by a computer system. Storage media that can be read by a computer include hard disks, floppy disks, magnetic media (e.g. magnetic tape), optical media (e.g. CD-ROM (compact disc read only memory), DVD (digital versatile disc), magneto-optical media (e.g., floptical disk), hardware device (e.g., read only memory (ROM), random access memory (RAM), or flash memory etc.), etc. In addition, program instructions may include not only machine language codes such as those created by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc. The above-mentioned hardware device It may be configured to operate as one or more software modules to perform the operations of various embodiments, and vice versa.

A module or program module according to various embodiments may include at least one of the above-described components, some of them may be omitted, or may further include other additional components. Operations performed by modules, program modules, or other components according to various embodiments may be executed sequentially, in parallel, iteratively, or in a heuristic manner. Additionally, some operations may be executed in a different order, omitted, or other operations may be added.

As used herein, the term “one” is defined as one or more than one. Additionally, the use of introductory phrases such as “at least one” and “one or more” in a claim may mean that the same claim contains introductory phrases such as “at least one” and “one or more” and ambiguous phrases such as “an.” The introduction of another claim element by the ambiguous phrase "a", if any, shall be construed to mean that any particular claim containing the claim element so introduced is limited to an invention containing only one such element. It shouldn't be.

In this document, expressions such as “A or B” or “at least one of A and/or B” may include all possible combinations of the items listed together.

Unless otherwise specified, terms such as “first” and “second” are used to optionally distinguish between the elements described by such terms. Accordingly, these terms are not necessarily intended to indicate temporal or other priority of such elements, and the mere fact that particular means are recited in different claims does not indicate that a combination of such means cannot be advantageously used. . Accordingly, these terms are not necessarily intended to indicate temporal or other priority of such elements. The mere fact that a particular measure is cited in different claims does not indicate that a combination of these measures cannot be used usefully.

The arrangement of components to achieve the same function is effectively “related” so that the desired function is achieved. Accordingly, any two components combined to achieve particular functionality may be considered to be “related” to each other such that the desired functionality is achieved, regardless of structure or intervening components. Likewise, two such associated components may be considered “operably connected” or “operably coupled” to each other to achieve a desired function.

Additionally, those skilled in the art will recognize that the boundaries between the functionality of the foregoing operations are illustrative only. Multiple operations may be combined into a single operation, a single operation may be distributed into additional operations, and the operations may be executed at least partially overlapping in time. Additionally, alternative embodiments may include multiple instances for a particular operation, and the order of the operations may vary in various other embodiments. However, other modifications, variations and alternatives are also possible. Accordingly, the detailed description and drawings are to be regarded in an illustrative rather than a restrictive sense.

The phrase “may be X” indicates that condition X may be met. This phrase also indicates that condition X may not be met. For example, a reference to a system containing a specific component should also include scenarios in which the system does not contain the specific component. For example, a reference to a method that includes a specific behavior should also include scenarios in which the method does not include that specific component. However, as another example, a reference to a system configured to perform a specific action should also include scenarios in which the system is not configured to perform a specific task.

The terms “comprising,” “having,” “consisting of,” “consisting of,” and “consisting essentially of” are used interchangeably. For example, any method may include at least the operations included in the drawings and/or the specification, or may include only the operations included in the drawings and/or the specification. Alternatively, the word “comprising” does not exclude the presence of elements or acts listed in a claim.

Those skilled in the art will recognize that the boundaries between logical blocks are illustrative only and that alternative embodiments may merge logical blocks or circuit elements or impose alternative decompositions of functionality on various logical blocks or circuit elements. You will recognize that it exists. Accordingly, it should be understood that the architecture shown herein is merely exemplary, and in fact many other architectures may be implemented that achieve the same functionality.

Also, for example, in one embodiment, the depicted examples could be implemented on a single integrated circuit or as circuitry located within the same device. Alternatively, the above examples could be implemented as any number of separate integrated circuits or individual devices interconnected together in any suitable manner, and other variations, modifications, variations and alternatives are also possible. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense.

Additionally, for example, the above-described examples, or portions thereof, may be implemented as a software or code representation of a physical circuit or a logical expression convertible to a physical circuit, such as any suitable type of hardware description language.

Additionally, the present invention is not limited to physical devices or units implemented in non-programmable hardware, but includes mainframes, minicomputers, servers, workstations, personal computers, notepads, etc., generally referred to herein as 'computer systems'. A programmable device that can perform a desired device function by operating in accordance with appropriate program code, such as personal digital assistants (PDAs), electronic games, automobiles and other embedded systems, cell phones, and various other wireless devices. Or it can also be applied to units.

A system, apparatus or device referred to in this specification includes at least one hardware component.

Connections as described herein may be any type of connection suitable for transmitting signals to or from each node, unit or device, for example via an intermediate device. Accordingly, unless implied or otherwise stated, a connection may be, for example, a direct connection or an indirect connection. A connection may be described or depicted with reference to being a single connection, multiple connections, one-way connection, or two-way connection. However, different embodiments may vary the implementation of the connection. For example, you could use a separate one-way connection rather than a two-way connection, or vice versa. Additionally, multiple connections can be replaced with a single connection that transmits multiple signals sequentially or in a time-multiplexed manner. Likewise, a single connection carrying multiple signals may be split into various connections carrying subsets of those signals. Therefore, many options exist for transmitting signals.

In the above, preferred embodiments of the technology of this specification have been described with reference to the attached drawings. Here, the terms or words used in this specification and claims should not be construed as limited to their usual or dictionary meanings, but should be construed as meanings and concepts consistent with the technical idea of the present invention. The scope of the present invention is not limited to the embodiments disclosed in this specification, and the present invention may be modified, changed, or improved in various forms within the scope described in the spirit and claims of the present invention.

Claims (13)

Selecting one time series prediction model from among a plurality of time series prediction models;
Reading price data for a plurality of agricultural products from a database and performing preprocessing for learning in the selected time series prediction model;
Generating at least one price data cluster by clustering the price data with a clustering model that groups at least one agricultural product by agricultural product, the clustering model applying the clustered price data to the selected time series prediction model Clustering the price data so that agricultural products that satisfy a predetermined tuning pattern for hyper-parameter tuning are in the same cluster, and clustering the price data into the number of clusters that yield the best performance evaluation results;
For the selected time series prediction model, hyper-parameters are tuned for each of the at least one price data cluster, and the tuned time series prediction model is trained for each of the at least one price data cluster, so that each agricultural product at a future point in time Generating a plurality of agricultural product price prediction models that calculate predicted prices;
Performing a performance evaluation for each agricultural product among the plurality of agricultural product price prediction models and determining the agricultural product price prediction model with the best performance as the final agricultural product price prediction model for the agricultural product; and
Comprising: calculating a predicted price for the target agricultural product on the target date using a final agricultural product price prediction model corresponding to the target agricultural product;
Agricultural product price prediction method using an agricultural product price prediction device. delete delete According to claim 1,
Hyper-parameter tuning for the selected time series prediction model is,
The selected time series prediction model is trained for each of the at least one price data cluster by changing the value of each hyper-parameter within a predetermined category, and the hyper-parameter value is set so that the selected time series prediction model produces the best performance. The process of determining the value of each hyper-parameter
Agricultural product price prediction method using an agricultural product price prediction device. According to claim 1,
The final agricultural product price prediction model is,
A first agricultural product price prediction model generated for each future time point based on the selected time series prediction model, and a second agricultural product price prediction in which the prediction result of the previous time point by the selected time series prediction model is input to a time series prediction model for prediction of the next time point. model, a third agricultural product price prediction model created by learning by including the prediction results of a time series prediction model predicting the price of the previous point in the input data of a time series prediction model predicting the price of the next point in time, and a single time series prediction model for all An agricultural product price prediction model selected among the fourth agricultural product price prediction models that calculates the predicted price of each agricultural product at a future point in time.
Agricultural product price prediction method using an agricultural product price prediction device. According to claim 1,
The step of reading price data for a plurality of agricultural products from the database and performing preprocessing for learning in the selected time series prediction model is:
Loading the price data from the database and reinforcing missing data using a predetermined estimation method; and
Containing the step of configuring the price data augmented with the missing data into a format for learning in the selected time series prediction model.
Agricultural product price prediction method using an agricultural product price prediction device. Combined with hardware,
A computer program stored in a computer-readable storage medium to execute each step included in the agricultural product price prediction method using the agricultural product price prediction device according to any one of claims 1, 4 to 6. A communication department that communicates with an external agricultural product price data provision server to collect price data for a plurality of agricultural products;
A storage unit that stores the collected price data in a database; and
It includes a control unit functionally connected to the communication unit and the storage unit, wherein the control unit includes,
Select one time series prediction model among a plurality of time series prediction models,
Reading the price data for a plurality of agricultural products from the database and performing preprocessing for learning in the selected time series prediction model,
Generating at least one price data cluster by clustering the price data using a clustering model that groups at least one agricultural product by agricultural product unit,
For the selected time series prediction model, hyper-parameters are tuned for each of the at least one price data cluster, and the tuned time series prediction model is trained for each of the at least one price data cluster, so that each agricultural product at a future point in time Create multiple agricultural product price prediction models that calculate predicted prices,
Among the plurality of agricultural product price prediction models, a performance evaluation is performed for each agricultural product for each agricultural product, and the agricultural product price prediction model with the best performance is determined as the final agricultural product price prediction model for the agricultural product,
Calculate the predicted price for the target agricultural product on the target date using the final agricultural product price prediction model corresponding to the target agricultural product,
When applying the clustered price data to the selected time series prediction model, the clustering model clusters the price data so that agricultural products that satisfy a predetermined tuning pattern for hyper-parameter tuning are in the same cluster, but achieves the best performance evaluation results. Clustering the price data by the number of clusters to be calculated
Agricultural product price prediction device. delete delete According to clause 8,
Hyper-parameter tuning for the selected time series prediction model is,
The selected time series prediction model is trained for each of the at least one price data cluster by changing the value of each hyper-parameter within a predetermined category, and the hyper-parameter value is set so that the selected time series prediction model produces the best performance. The process of determining the value of each hyper-parameter
Agricultural product price prediction device. According to clause 8,
The final agricultural product price prediction model is,
A first agricultural product price prediction model generated for each future time point based on the selected time series prediction model, and a second agricultural product price prediction in which the prediction result of the previous time point by the selected time series prediction model is input to a time series prediction model for prediction of the next time point. model, a third agricultural product price prediction model created by learning by including the prediction results of a time series prediction model predicting the price of the previous point in the input data of a time series prediction model predicting the price of the next point in time, and a single time series prediction model for all An agricultural product price prediction model selected among the fourth agricultural product price prediction models that calculates the predicted price of each agricultural product at a future point in time.
Agricultural product price prediction device. According to clause 8,
The control unit,
Reading price data for multiple agricultural products from the database and performing preprocessing for learning in the selected time series prediction model,
Retrieving the price data from the database and reinforcing missing data using a predetermined estimation method,
Configuring the price data augmented with the missing data into a format for learning in the selected time series prediction model.
Agricultural product price prediction device.

Images