Pepper price prediction method

Abstract

The invention discloses a pepper price prediction method. The method comprises the following steps: constructing a database of meteorological data, historical price data and market supply and demand relationships; constructing a polynomial cross feature regression (PCFR) model; and a pepper price index is determined, a deep neural network is trained and generated, and then a pepper price prediction model is formed. Based on historical price data of a plurality of important pepper varieties in Guizhou province, a big data analysis technology is applied to research and build pepper price indexes, and a correlation model between pepper prices and meteorological elements is researched and built; and carrying out grey prediction on the prices of the five pepper varieties under the comprehensive influence of the current weather condition, the weather forecast and other price influence factors to realize pepper monthly price prediction and generate a dynamic trend chart. Through price prediction, theoretical support and data support are provided for macroscopic regulation and control of the government, so that a competent department reasonably plans or adjusts production of the hot peppers, and a market balance mechanism of the hot pepper price is improved by strengthening social government.

Description

Technical Field

[0001] This invention relates to the field of agricultural product information technology, specifically a method for predicting chili pepper prices. Background Technology

[0002] Chili peppers are one of the twelve major characteristic industries in Guizhou Province. Fluctuations in chili pepper prices directly affect chili pepper production and distribution, increasing the risk of fluctuations in fresh chili pepper production and sales as well as overall price levels. The price trend of chili peppers is an important part of Guizhou's agricultural product price monitoring and analysis.

[0003] A variety of factors influence chili price fluctuations, including long-term market supply and demand relationships as well as short-term non-market factors. Weather, policy, price expectations, and market speculation all play key roles. For example, weather conditions such as sunshine, precipitation, and climate disasters directly affect the yield and quality of chili peppers, thus impacting prices. Policy adjustments and market expectations can also cause short-term price fluctuations. These intertwined factors make chili prices volatile and inherently uncertain.

[0004] Globally, fluctuations in chili prices play a crucial indicative role in agricultural markets, especially in major chili-producing regions like Guizhou Province. Establishing a scientific and reasonable price index is vital for monitoring and predicting market trends. In recent years, while research on chili price indices has begun to emerge both domestically and internationally, standards are not yet unified, and different index systems have varying focuses, such as the National Chili Daily Base Production Area 116 Index, the Chili Wholesale 150 Index, and the Chili Import 166 Index. These indices primarily calculate a comprehensive weighted average of market prices for various chili varieties, aiming to provide a general analysis of market prices. Despite the existence of these mature indices, such research remains nonexistent in Guizhou Province. This not only limits the ability of local governments and agricultural departments to accurately regulate the market and optimize resource allocation but also affects farmers' strategies for responding to market changes. Therefore, our research team utilized historical price data from more than ten major chili varieties in Guizhou Province and employed advanced big data analytics to construct a chili price index. Through this index, we conducted grey market price predictions and generated dynamic price trend charts. These trend charts clearly demonstrate the historical trends and future forecasts of chili prices, providing strong data support for policymakers.

[0005] Furthermore, the research findings have extremely broad application prospects. Firstly, they can help government departments conduct more accurate market forecasting and regulation at the macro level, optimizing the balance between chili production and market supply. Through predictive models, the government can anticipate the risk of significant price fluctuations and adjust relevant policies in a timely manner to stabilize market expectations and farmers' incomes. Secondly, this research also contributes to the protection and rational development of chili resources, promoting sustainable resource utilization and comprehensive environmental governance through scientific data analysis. Finally, by improving the market equilibrium mechanism for chili prices, not only can the healthy development of the chili industry be guaranteed, but the stability of the entire socio-economic system can also be promoted. Therefore, establishing and improving the chili price index in Guizhou Province has immeasurable value for achieving the sound and sustainable development of the chili industry. Summary of the Invention

[0006] Therefore, in order to overcome the above-mentioned shortcomings, the present invention provides a method for predicting chili prices.

[0007] This invention is implemented by constructing a method for predicting chili pepper prices, characterized by: building a database of meteorological data, historical price data, and market supply and demand relationships;

[0008] Construct a multinomial cross-feature regression (PCFR) model;

[0009] A chili price index is determined, and a deep neural network is trained to generate a chili price prediction model.

[0010] According to the chili price prediction method described in this application, the database construction includes: periodically reading chili price data from the Guizhou Agricultural Economy Network database interface, periodically crawling chili price data from a designated website, cleaning the acquired data, and importing it into the database; and importing meteorological data from the cimiss interface into the background database.

[0011] According to the chili price prediction method described in this application, the method is characterized in that the construction of the multinomial cross-feature regression (PCFR) model includes:

[0012] The algorithm model involved in this application will be introduced below.

[0013] ① Linear Regression (LR)

[0014] This model assumes a linear relationship between the independent and dependent variables, meaning the dependent variable can be represented by a linear combination of the independent variables, as shown in the following formula:

[0015] y=β0+β1x1+β2x2+β3x3+...+β n x n +∈

[0016] y is the dependent variable (prediction target), β0 is a constant term, and β1, β2, ..., β n These are the coefficients of the independent variable, x1, x2, ..., x. n ∈ represents the independent variable (feature); ∈ represents the error term, indicating the portion that the model cannot explain. The goal of linear regression is to find a best-fitting linear function that minimizes the error between the predicted and observed values. This is typically achieved by minimizing the sum of squared residuals (i.e., the sum of the squares of the differences between the true and predicted values);

[0017] In agricultural product price forecasting, factors such as weather may have a linear relationship with prices; therefore, linear regression can provide a basic model. It solves the problem of establishing a linear relationship between the target variable and explanatory variables, thus providing a foundational model.

[0018] ② Multinomial Regression (PR)

[0019] Given a single explanatory variable x and a target variable y, their relationship can be expressed as:

[0020] y = β0 + β1x + β2x 2 +β3x 3 +…+β n x n +∈

[0021] Where β0, β1, β2…β n y is the regression coefficient, and x is the error term. By fitting such a polynomial, we can better describe the relationship between the target variable y and x.

[0022] The weather forecast is transformed using a multinomial regression method. By introducing higher-order terms for the explanatory variables, multinomial regression can better handle the non-linear relationship between the target and explanatory variables. In agricultural product price forecasting, there may be some non-linear trends; multinomial regression helps to fit these relationships more flexibly. It resolves the non-linear relationship between the target and explanatory variables by introducing higher-order terms to fit the data more flexibly.

[0023] ③ Polynomial Cross Regression (PCFR)

[0024] Polynomial Cross Features Regression (PCFR) is a further improvement on polynomial regression. By introducing cross terms between features, it better captures the interactions between different features. In agricultural product price forecasting, there may be complex correlations between different weather factors, and PCFR helps to capture these relationships more comprehensively. It solves the problem of interactions between features and captures the correlations between different features more comprehensively.

[0025] In multinomial regression, we assume the dependent variable y and the independent variables x1, x2, ..., xn. n The relationship between them can be represented by a polynomial function. Its basic form is:

[0026]

[0027] Where, β i ∈ represents the coefficient to be estimated, and ∈ represents the error term.

[0028] In multinomial cross-feature regression, in addition to the multinomial terms of individual features, the interactions between features are also considered. Suppose we have two features x1 and x2, the introduction of the cross term can be expressed as:

[0029]

[0030] In this formula, β 12 x1x2 represents the interaction between features x1 and x2. By introducing these interaction terms, PCFR can capture more complex relationships, especially in agricultural product price forecasting, where complex correlations may exist between different weather factors.

[0031] According to the chili price forecasting method described in this application, the method is characterized in that determining the chili price index includes:

[0032] Determining the factors influencing chili prices

[0033] The classification of the current year's anomaly is based on the previous year's temperature anomaly and the previous year's precipitation anomaly percentage. The precipitation anomaly percentage is calculated as follows:

[0034] The formula for calculating the percentage of precipitation during a certain period relative to the average precipitation for the same period in previous years is:

[0035]

[0036] P represents the annual precipitation of the current year, and Pi represents the average precipitation over the same period over many years, taking the average precipitation over 5 years.

[0037] Precipitation anomalies are classified according to the percentage of precipitation anomalies.

[0038] The temperature anomaly is calculated as follows;

[0039] (1) First calculate the average value.

[0040]

[0041] Where: Yi—temperature (°C) value over the past 5 years; Yp—average temperature (°C) value over the corresponding period; n—number of years in the data sequence sample, taken as 5 years;

[0042] (2) Then calculate the standard deviation s

[0043]

[0044] Finally, the anomaly degree C is calculated.

[0045]

[0046] Then, the annual type is divided by combining the temperature anomaly and precipitation anomaly. The annual type is used as the input of the neural network model and is represented by a two-dimensional vector year_type=[Y1,Y2]. A poor harvest year is represented as [1,0], a good harvest year is represented as [0,1], and a normal year is represented as [0,0].

[0047] The chili price index is determined according to;

[0048] First, calculate the average price of the same type of chili pepper across different markets nationwide. Then, calculate the price index based on the average price. Set the price of a certain period as the base period price, and the index value of that month, i.e., the base period value indexj, is 1000. If the average price of chili peppers on a certain day is Pi, then the price index indexi for that day is:

[0049] Indexi = Pi / Pj × 1000

[0050] Through correlation analysis between the price change of a certain type of chili pepper and the national price index of similar chili peppers, it was found that the national price index of similar chili peppers reflects the overall market conditions and has a certain predictive ability for the price changes of chili peppers in Guizhou. Therefore, the national price index of similar chili peppers was directly used as one of the inputs to the neural network model.

[0051] Feature selection and feature engineering problems in chili price forecasting;

[0052] Analysis revealed weak or no correlation between prices and individual meteorological indices reflecting short- to medium-term weather conditions. However, historical climate factors were considered, and annual characteristics were used as one of the inputs. A price index system was constructed to reflect the comprehensive impact of meteorological factors, industry supply and demand, national similar prices, prices of the same period last year, market factors, and policy factors on chili prices. The price index was also used as one of the inputs to the neural network model. To incorporate the impact of regional meteorological factors on chili production and transportation into the model, a convolutional neural network was also used to extract features from gridded high-dimensional meteorological data during modeling.

[0053] Weekly price forecasting employs three models: ① Linear Regression (LR), which predicts weekly prices for the next four weeks based on historical price data; ② Multinomial Regression (PR), which uses the price curves of the previous three weeks, annual trends, and price indices as features to predict weekly prices for the next four weeks; and ③ Multinomial Cross-Feature Regression (PCFR), whose input consists of a vector composed of prices from the past three weeks, annual trends under the influence of climate factors, price indices, and gridded meteorological features extracted by a convolutional neural network, and whose output is the weekly price for the next four weeks. Monthly price forecasting uses the PCFR model and a deep neural network to predict monthly prices for the next 12 months. The LR input consists of a vector composed of monthly price curves from the past 12 months, annual trends under the influence of climate factors, price indices, and gridded meteorological features extracted by a convolutional neural network, and whose output is the monthly price for the next 12 months.

[0054] This invention has the following advantages: Based on historical price data of several important chili pepper varieties in Guizhou Province, this application studies and constructs a correlation model between chili pepper prices and meteorological factors. It then performs grey prediction on the prices of five chili pepper varieties under the combined influence of current weather conditions, weather forecasts, and other price-influencing factors, achieving monthly chili pepper price forecasting and generating dynamic trend charts. Its benefits include providing theoretical and data support for government macro-control through price prediction, enabling relevant departments to rationally plan or adjust chili pepper production, improve the market equilibrium mechanism for chili pepper prices through strengthened social governance, and utilize policies to regulate the relationship between the protection and development of chili pepper resources, thereby ensuring the healthy and sustainable development of the chili pepper industry. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating the implementation of this technology;

[0056] Figure 2 Distribution chart of various types of chili peppers;

[0057] Figure 3 Map showing the regional distribution of chili pepper index data;

[0058] Figure 4 There is a chart predicting the annual price of chili peppers;

[0059] Figure 5 There is a quarterly price forecast chart for chili peppers;

[0060] Figure 6 There is a monthly price forecast chart for chili peppers;

[0061] Figure 7 There is a weekly price forecast chart for chili peppers;

[0062] Figure 8 Annual price forecast chart for chili peppers without stems;

[0063] Figure 9Quarterly price forecast chart for chili peppers without stems;

[0064] Figure 10 Monthly price forecast chart for chili peppers without stems;

[0065] Figure 11 Annual price forecast chart for peppers without stems;

[0066] Figure 12 Quarterly price forecast chart for bell peppers without stems;

[0067] Figure 13 Monthly price forecast chart for stemless red peppers;

[0068] Figure 14 Weekly price forecast chart for bell peppers without stems. Detailed Implementation

[0069] The following will be combined with the appendix Figures 1-14 This invention will be described in detail, and the technical solutions in the embodiments of this invention will be clearly and completely described. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0070] This invention provides a method for predicting chili prices, which includes:

[0071] Construct a database of meteorological data, historical price data, and market supply and demand.

[0072] Construct a multinomial cross-feature regression (PCFR) model;

[0073] A chili price index is determined, and a deep neural network is trained to generate a chili price prediction model.

[0074] This application comprehensively considers various factors such as historical prices, seasonal factors, climate conditions, market supply and demand, market competition and trading environment, policy factors, and international trade. Through big data analysis and price prediction model algorithm design, and verification of the effectiveness of the prediction model, a reliable chili price prediction model for Guizhou Province is established. This model enables the prediction and visualization of chili price fluctuation ranges and areas, providing a basis for government macro-control, providing chili price forecasts for chili production and sales enterprises, and offering decision-making references for chili production and sales. This aims to improve market sensitivity and responsiveness, reduce market risks, increase profits, and thus promote the sustainable development of the chili industry.

[0075] This application establishes a correlation between chili prices and meteorological data. Historical data on the prices of stemmed and stemless chilies, including round, small, bright red, bullet, and fresh long chilies, were collected, encompassing market prices, seasonal variations, and climatic factors. Then, big data correlation analysis and visualization analysis, such as linear regression and machine learning algorithms, were used to process and analyze this data. A chili price index prediction system for the Guizhou Agricultural Economy Network was built. The system enables real-time price updates for major chili varieties across the province, as well as time and regional price predictions. The system has been promoted and applied in chili planting bases and cooperatives in southwestern Guizhou, achieving good results.

[0076] When this application is implemented, a meteorological and price information database will be constructed:

[0077] Regularly read chili price data from the Guizhou Agricultural Economy Network database interface, regularly crawl chili price data from designated websites, clean the acquired data and import it into the database; obtain meteorological data from the cimiss interface and import it into the backend database.

[0078] The algorithm model involved in this application will be introduced below.

[0079] ① Linear Regression (LR)

[0080] Linear regression is a statistical model used to establish a linear relationship between variables. In linear regression, one or more independent variables (also called features) are used to predict a continuous dependent variable (also called the target variable). This model assumes a linear relationship between the independent and dependent variables; that is, the dependent variable can be represented by a linear combination of the independent variables. The formula is as follows:

[0081] y=β0+β1x1+β2x2+β3x3+...+β n x n +∈

[0082] y is the dependent variable (prediction target), β0 is a constant term, and β1, β2, ..., β n These are the coefficients of the independent variable, x1, x2, ..., x. n ∈ represents the independent variable (feature); ∈ represents the error term, indicating the portion that the model cannot explain. The goal of linear regression is to find a best-fitting linear function that minimizes the error between the predicted and observed values. This is typically achieved by minimizing the sum of squared residuals (i.e., the sum of the squares of the differences between the true and predicted values).

[0083] In this project, we used price as the target variable and weather as the explanatory variable. This module uses the least squares method for model training by default. Linear regression is a simple and effective algorithm suitable for establishing a linear relationship between the target variable and the explanatory variables. In agricultural product price forecasting, factors such as weather may have a linear relationship with prices, therefore linear regression can provide a basic model. It solves the problem of establishing a linear relationship between the target variable and the explanatory variables, providing a basic model.

[0084] ② Multinomial Regression (PR)

[0085] Polynomial regression is an extension of linear regression, used to handle non-linear relationships between target and explanatory variables. Therefore, polynomial regression does not simply fit the data in a linear space, but rather fits complex non-linear relationships through a higher-dimensional feature space. Suppose we have a single explanatory variable x and a target variable y, their relationship can be expressed as:

[0086] y = β0 + β1x + β2c 2 +β3c 3 +…+β n x n +∈

[0087] Where β0, β1, β2…β n y is the regression coefficient, and x is the error term. By fitting such a polynomial, we can better describe the relationship between the target variable y and x.

[0088] In this project, we will perform a multinomial transformation on the weather data. Multinomial regression, by introducing higher-order terms for the explanatory variables, can better handle the nonlinear relationship between the target variable and the explanatory variables. In agricultural product price forecasting, there may be some nonlinear trends, and multinomial regression helps to fit these relationships more flexibly. It resolves the nonlinear relationship between the target variable and the explanatory variables by introducing higher-order terms to fit the data more flexibly.

[0089] Multinomial regression expands the expressive power of linear regression by introducing higher-order terms, enabling it to more flexibly adapt to the changing trends of nonlinear data. For example, when predicting agricultural yields, weather patterns, or stock prices, the data may exhibit seasonal fluctuations or periodic variations. These nonlinear relationships cannot be effectively fitted using only linear regression. Compared to multinomial cross-feature regression, multinomial regression has significant advantages in terms of computational complexity and model stability. Therefore, when dealing with practical problems with nonlinear trends, multinomial regression is a highly effective modeling method that can improve model accuracy and provide strong support for decision-making.

[0090] Compared to traditional linear regression and multinomial cross-feature regression, multinomial regression typically offers higher accuracy when handling nonlinear data. Multinomial regression not only captures the curve relationships within the data but also balances fitting accuracy and model complexity by appropriately selecting the polynomial order, effectively avoiding overfitting and improving prediction accuracy. In fields such as agricultural price forecasting and meteorological data modeling, by setting the polynomial order appropriately, more accurate prediction results can be obtained than with traditional linear regression.

[0091] ③ Polynomial Cross Regression (PCFR)

[0092] Polynomial Cross Features Regression (PCFR) is a further improvement on polynomial regression. By introducing cross terms between features, it better captures the interactions between different features. PCFR further considers the interactions between explanatory variables, improving the model's expressive power. In agricultural product price forecasting, there may be complex correlations between different weather factors, and PCFR helps to capture these relationships more comprehensively. It solves the problem of interactions between features, capturing the correlations between different features more comprehensively.

[0093] In multinomial regression, we assume the dependent variable y and the independent variables x1, x2, ..., xn. n The relationship between them can be represented by a polynomial function. Its basic form is:

[0094]

[0095] Where, β i ∈ represents the coefficient to be estimated, and ∈ represents the error term.

[0096] In multinomial cross-feature regression, in addition to the multinomial terms of individual features, the interactions between features are also considered. Suppose we have two features x1 and x2, the introduction of the cross term can be expressed as:

[0097]

[0098] In this formula, β 12 x1x2 represents the interaction between features x1 and x2. By introducing these interaction terms, PCFR can capture more complex relationships, especially in agricultural product price forecasting, where complex correlations may exist between different weather factors.

[0099] The process and results of this technology are explained below;

[0100] Factors affecting chili prices

[0101] Climate Impact: Considering the long-term impact of climate on chili pepper cultivation, which affects the yield at harvest and consequently price changes after harvest, this study analyzes the meteorological factors of the previous year (data source: meteorological data for various chili pepper producing areas obtained from CIMISS) to classify the current year's chili pepper price into different types (different from traditional crop climate year types; here, the year type refers to the yield of chili peppers in the harvest year, i.e., bumper harvest, poor harvest, or average harvest). Price change patterns differ across year types. Based on the "Agricultural Climate Impact Assessment: Crop Climate Year Type Classification Method (GB / T 21986-2008)," this study modifies the method slightly, classifying the current year's year type based on the previous year's temperature anomaly and the previous year's annual precipitation anomaly percentage.

[0102] The percentage of precipitation anomaly is calculated as follows:

[0103] The formula for calculating the percentage of precipitation during a certain period relative to the average precipitation for the same period in previous years is:

[0104]

[0105] In this technique, P represents the annual precipitation of the current year. i The average precipitation for the same period over many years is the average precipitation over 5 years (2018-2023).

[0106] The following table shows the classification of precipitation anomalies based on the percentage of precipitation anomalies.

[0107] Precipitation anomalies are classified according to the percentage of precipitation anomalies.

[0108]

[0109] The temperature anomaly is calculated as follows.

[0110] (1) First calculate the average value.

[0111]

[0112] In the formula:

[0113] Yi – Temperature (°C) values ​​over the past 5 years;

[0114] Yp—The average value of the temperature (°C) element within the corresponding time period;

[0115] n—The number of years in the data sequence sample. This paper uses 5 years.

[0116] (2) Then calculate the standard deviation s

[0117]

[0118] Finally, the anomaly degree C is calculated.

[0119]

[0120] The classification of temperature anomalies is shown in the table below.

[0121] The following table shows the classification of temperature anomalies.

[0122]

[0123] Then, the annual type is determined by combining temperature anomalies and precipitation anomalies, as shown in the table below:

[0124] The table below classifies year types based on both temperature and precipitation anomalies.

[0125]

[0126] The year type is used as the input to the neural network model and is represented by a two-dimensional vector year_type = [Y1, Y2]. A poor harvest year is represented as [1, 0], a good harvest year is represented as [0, 1], and a normal harvest year is represented as [0, 0].

[0127] The innovation of this technology lies in:

[0128] ① The technological research results fill the gap in chili price index research and chili price forecasting in our province. ② Technologically, the project has three innovations:

[0129] Firstly, when analyzing the chili price index, the index system not only comprehensively considers market factors and other factors, but also takes into account long-term meteorological factors, namely the influence of climate, so that the constructed system can more comprehensively and accurately reflect the market conditions of chili.

[0130] Secondly, when establishing a chili price forecasting model, not only the impact of market supply and demand on chili prices is considered, but also the impact of regional meteorological factors on chili production and transportation. This allows for a more comprehensive summary of influencing characteristics during the modeling process, making the model's forecast results more reasonable and accurate.

[0131] Third, when building a prediction model, deep learning algorithms, such as convolutional neural networks and deep neural networks, are used. Convolutional neural networks can effectively extract high-dimensional features of the data, reduce the complexity of the model structure, and more accurately predict future chili prices.

[0132] 2. Key technical issues to be addressed:

[0133] ① Issues related to cleaning, organizing, and synchronizing historical data.

[0134] The project team cleaned and organized historical price data for the past five years for several important chili varieties, including stem-less chili, stemless small chili, stemless bright chili, and fresh long chili, from the Guizhou Agricultural Economy Network database. They also obtained and synchronized chili price data from several important national chili websites, such as the Guizhou Agricultural Economy Network, Guizhou Chili Industry Information Network, Chili World Network, and Oriental Chili Network, through data interfaces and web scraping. At the same time, they obtained meteorological information data such as temperature, precipitation, and air humidity from the CIMISS interface and imported it into the backend database.

[0135] ② How is the chili price index determined?

[0136] The project team first analyzed the correlation between the daily, weekly, and monthly average prices of five types of chili peppers and meteorological indicators such as average temperature, precipitation, air humidity, and high-temperature heat damage, and found that the two were either uncorrelated or weakly correlated.

[0137] The project team further analyzed the meteorological factors of the previous year, mainly based on the previous year's temperature anomaly and the percentage of annual precipitation anomaly, to classify the current year's crop year type (different from traditional crop climate year types; here, the year type refers to the yield of chili peppers in the harvest year, i.e., bumper harvest, poor harvest, or average harvest; price change patterns differ between different year types. This study slightly modified the method according to "Agricultural Climate Impact Assessment: Crop Climate Year Type Classification Method (GB / T 21986-2008)"). The year type serves as the input to the neural network model. Furthermore, considering the combined effects of industry supply and demand, prices in the same period of the previous year, national prices for similar products, policy factors, and sudden epidemics on price changes, a price index system is constructed to reflect the comprehensive impact of various factors on prices.

[0138] First, calculate the average price of the same type of chili pepper across different markets nationwide. Then, calculate the price index based on the average price. Set the price of a certain period as the base period price, and the index value of that month, i.e., the base period value indexj, is 1000. If the average price of chili peppers on a certain day is Pi, then the price index indexi for that day is:

[0139] Indexi = Pi / Pj × 1000

[0140] By analyzing the correlation between the price change of a certain type of chili pepper and the national price index of similar chili peppers, it was found that the national price index of similar chili peppers reflects the overall market conditions and has a certain predictive power for the price changes of chili peppers in Guizhou. Therefore, the national price index of similar chili peppers was directly used as one of the inputs to the neural network model.

[0141] ③ Feature selection and feature engineering problems in chili price forecasting.

[0142] Analysis revealed weak or no correlation between prices and individual meteorological indices reflecting short- to medium-term weather conditions. However, historical climate factors were considered, and annual characteristics were used as one of the inputs. A price index system was constructed to reflect the comprehensive impact of meteorological factors, industry supply and demand, national similar prices, prices in the same period of the previous year, market factors, and policy factors on chili prices. The price index was also used as one of the inputs to the neural network model. In order to incorporate the impact of regional meteorological factors on chili production and transportation into the model, a convolutional neural network was also used to extract features from gridded high-dimensional meteorological data during modeling.

[0143] Weekly price forecasting employs three models: ① Linear Regression (LR), which predicts weekly prices for the next four weeks based on historical price data; ② Multinomial Regression (PR), which uses the price curves of the previous three weeks, annual trends, and price indices as features to predict weekly prices for the next four weeks; and ③ Multinomial Cross-Feature Regression (PCFR), whose input consists of a vector composed of prices from the past three weeks, annual trends under the influence of climate factors, price indices, and gridded meteorological features extracted by a convolutional neural network, and whose output is the weekly price for the next four weeks. Monthly price forecasting uses the PCFR model and a deep neural network to predict monthly prices for the next 12 months. The LR input consists of a vector composed of monthly price curves from the past 12 months, annual trends under the influence of climate factors, price indices, and gridded meteorological features extracted by a convolutional neural network, and whose output is the monthly price for the next 12 months.

[0144] The following section explains the determination of policy factors, market factors, and other factors;

[0145] Supply and Demand: A bumper or poor harvest of chili peppers directly determines price fluctuations. If the current chili pepper harvest is bountiful, exceeding the output of the same period last year, the relatively stable market demand will put downward pressure on prices, resulting in farmers not seeing increased income despite a good harvest, which in turn will reduce production for the next planting season. Conversely, rising chili pepper prices will lead farmers to increase chili pepper planting in their subsequent planting decisions.

[0149] Nationwide similar prices: Although the price of the same type of chili pepper varies in different places, the price fluctuations of similar chili peppers can be used as a reference, considering the demand for the same type of chili pepper.

[0150] Last year's price: Under policy regulation, the phenomenon of drastic price fluctuations in chili peppers has improved. Therefore, last year's price can be used as a benchmark to predict this year's price.

[0151] The combined influence of meteorological factors and other factors

[0152] In summary, the relationship between the constructed single meteorological index and chili pepper prices shows a weak correlation. Most of the constructed meteorological indices are real-time and short-to-medium-term meteorological factors, while the impact of weather on chili pepper prices is often long-term. Therefore, the above study also divided the year into annual patterns based on long-term climate conditions, using the annual pattern as one of the inputs to the neural network model.

[0153] Chili price fluctuations are influenced not only by weather factors, but also by industry supply and demand, prices in the same period last year, national prices of similar products, policy factors, and sudden outbreaks of disease. These factors combined affect price changes, so a price index system is constructed to comprehensively reflect the impact of various factors on prices.

[0154] Research and determination of the chili price index;

[0155] Chili price index data distribution:

[0156] A chili price index needs to reflect the overall price level of the chili market. A rise in the chili price index indicates an increase in the average price level of chilies; conversely, a fall indicates a decrease. There is a certain correlation between the price fluctuations of chilies nationwide and those in Guizhou province.

[0157] By collecting and organizing relevant data on different categories of chili peppers, we aim to demonstrate the coverage and breadth of data collection for these categories in actual market research. The charts, from left to right, show various categories of chili peppers, with the data volume for each category represented by a vertical bar chart. It is evident from the charts that the seventh category of chili peppers has a significantly higher data volume than other categories, reaching 94,100 data records, which may indicate a higher level of market attention and in-depth research on this category. In contrast, other categories have lower data volumes, with most categories having fewer than 15,000 records. The first and second categories have the fewest data records, at only 5,100 and 5,400 respectively, which may reflect a smaller scale of cultivation or sales, or limited market circulation for these categories. Figure 2 Distribution of data for various types of chili peppers

[0158] Data distribution is crucial for understanding the performance of different chili pepper varieties in the Guizhou market and growers' preferences. Categories with abundant data may indicate high market demand and significant commercial value, while categories with less data may require further marketing or adjustments to planting area. The data collected in this study not only allows us to analyze market dynamics and grower preferences for various chili pepper types but also provides decision-making support for the government and related enterprises based on data volume, enabling more scientific and rational planning in areas such as agricultural policy formulation, marketing strategies, and resource allocation. In the future, we plan to expand the scope and depth of data collection to provide more comprehensive market analysis and forecasts.

[0159] Regional distribution of chili price index:

[0160] Figure 3 Regional distribution of chili index data; when studying market data for various types of chili peppers in Guizhou Province, it is very important to consider different regions as influencing factors.

[0161] Guizhou Province boasts diverse geographical and climatic conditions, ranging from mountainous to plains, which directly impact the growing environment and yield of chili peppers. For instance, Guiyang, as the provincial capital, has a more concentrated economic activity and potentially greater market demand, influencing the market price and planting scale of chili peppers. Meanwhile, areas surrounding Guiyang may become hotspots for chili pepper cultivation and sales due to their convenient transportation and logistics.

[0162] By analyzing chili pepper cultivation conditions and market data in different regions, we can better understand the differences in market demand, growers' planting preferences, and the regional characteristics of price fluctuations. Such analysis can help governments and agricultural enterprises formulate more precise market strategies and agricultural policies, promote balanced regional economic development, and provide consumers with more diversified product choices. Furthermore, by studying regional factors, we can also predict future chili pepper market trends, such as identifying potential growth areas or anticipating possible supply shortages. This in-depth regional analysis not only improves the economic benefits of chili pepper cultivation but also strengthens the rational allocation and management of agricultural resources.

[0163] Example: Prediction results for various types of chili peppers

[0164] 1) Price prediction for chili peppers:

[0165] Figure 4 There is a forecast of the annual price of chili peppers. Figure 4The graph shows the predictive analysis results of three different models—linear regression (LR), multinomial regression (PR), and PCR (principal component regression)—on the average annual price of chili peppers, spanning from 2021 to 2026. As can be seen from the graph, the prediction results of the three models are similar in overall trend, all showing an upward trend in the average annual price of chili peppers year by year. The starting price in 2021 was 19, and the price increased year by year thereafter. By 2026, the predicted price of all three models is close to 21, with linear regression (LR) showing a simple linear trend of price increases year by year. Its prediction result is significantly higher than the other two methods, reaching nearly 24.5 units in 2026, which may be an overestimation. The multinomial regression (PR) curve shows a certain non-linear fluctuation trend, slightly declining in 2024 and then rebounding, with the prediction result being more balanced among the three models, reaching approximately 23 units in 2026. Similar to PR, the Polynomial Cross-Regression Factual (PCFR) also exhibits a fluctuating trend, but it provides a more conservative forecast of future price increases, projecting approximately 22.5 for 2026, which is more likely to be close to the actual figure. This price increase may reflect increased demand in the chili pepper market or rising production costs. Analyzing these forecasts can help growers and market analysts better plan future planting and sales strategies. For example, if the forecast indicates continued price increases, growers may increase the planting area for chili peppers, while merchants may seek to stockpile in advance to avoid purchasing when costs are higher in the future.

[0166] Figure 5 There is a quarterly price forecast for chili peppers; from Figure 5 As can be seen, the price forecast for chili peppers exhibits significant volatility across quarters. Initially, prices gradually rose from a low point in the fourth quarter of 2021 to a high point in the second quarter of 2023, reaching approximately 27 yuan. Subsequently, prices plummeted, returning to near the initial 2021 level (approximately 18 yuan) by the fourth quarter of 2023. This trend indicates that market prices experienced significant volatility in a short period, exhibiting a pattern of "rapid rise followed by rapid fall." This price fluctuation may reflect the impact of seasonal factors, changes in market supply and demand, fluctuations in production costs, or other macroeconomic conditions. For example, seasonal price increases may be related to increased holiday demand, while price decreases may be due to increased production or market saturation. These forecasts are significant for chili pepper growers and distributors, who can use this data to optimize planting plans, adjust inventory, and develop sales strategies. For instance, they can increase production in advance during seasons with predicted higher prices or reduce inventory when prices are expected to fall.

[0167] Figure 6There are monthly price forecasts for chili peppers. As shown in the chart, all models predict relatively stable prices for chili peppers from October 2023 to June 2024, remaining at a low level of approximately 16 to 18 yuan. However, starting in July 2024, forecasts show a sharp price increase, reaching a high of approximately 24 yuan in August 2024. Afterward, prices slightly declined but remained at a high level until the end of 2024. This sharp price increase may be related to seasonal production fluctuations, increased market demand, or changes in the supply chain. Particularly in summer, reduced planting area or adverse weather conditions can lead to decreased supply and potentially higher prices. Furthermore, festivals and specific cultural events may also drive demand growth in certain months, thus affecting prices.

[0168] Figure 7 There is a weekly price forecast for chili peppers. From September 2021 to December 2022, actual data showed a gradual downward trend. The price gradually decreased from approximately 24 yuan to approximately 20 yuan, with overall stability but slight rebounds in certain periods (such as late October to early November). This trend may be influenced by market factors, policy intervention, or fluctuations in supply and demand. The LR model shows a single linear downward trend, ignoring fluctuations in the actual data. Linear regression is suitable for long-term trend analysis, but it cannot accurately capture short-term changes when faced with complex fluctuations. The PR model predicted the downward price trend and partially captured fluctuations, such as the price rebound in early November. Compared to the LR model, PR is closer to the actual data changes, but there is still some deviation in the magnitude prediction. The PCFR model captures the main price trends and fluctuation characteristics, and its curve is closest to the actual data. During the local rebound in late October and the price stabilization period in December, PCFR outperformed PR and LR. The price increase may be related to the late summer and early autumn harvest season, when the supply of chili peppers in the market may be relatively tight, especially if seasonal weather affects production. In addition, holidays and special events may also drive up demand during certain weeks, thus pushing up prices.

[0169] 2) Price prediction for chili peppers without stems:

[0170] Figure 8This paper presents three models—linear regression (LR), multinomial regression (PR), and principal component regression (PCR)—to predict the average annual price of small red chili peppers, spanning from 2021 to 2026. It can be seen that the predicted trajectories of the average annual price of small red chili peppers differ slightly among the models, but they generally exhibit similar trends. Starting in 2021, the price gradually declined from 24 yuan to a low of approximately 22 yuan in 2023, before rebounding from 2024 and is expected to continue rising until 2026. The linear regression (LR) model shows a relatively gentle price decline and rise, reflecting a stable long-term trend. Compared to LR, the PR model considers the non-linear changes in prices, resulting in a gentler upward trend that is more likely to closely approximate the actual market. The PCFR model predicts a strong upward momentum after 2024, which may indicate that after considering the principal components, market sentiment or other external economic factors are expected to have a positive impact on the price of small red chili peppers. Figure 8 Annual price forecast for chili peppers without stems.

[0171] Price fluctuations can be related to a variety of factors, including changes in agricultural production costs, market supply and demand, and macroeconomic conditions. For example, an oversupply of chili peppers or a decline in consumer demand in 2023 could lead to a temporary price decrease. Subsequent price rebounds could be due to market adjustments, such as reduced supply or increased demand.

[0172] exist Figure 9 The price forecasts for millet peppers show significant volatility. Prices remained stable (around 21 yuan) from Q4 2021 to Q2 2022. Subsequently, prices rose rapidly from Q3 2022 to Q1 2023, reaching a peak (around 27 yuan), before gradually declining to around 21 yuan in Q2 2023. This change reflects the market's gradual recovery after a temporary shock, possibly due to a surge in demand or supply shortages. Starting in Q4 2024, the forecasts from the three models began to diverge. The LR model predicted a stable, gradual upward trend from 2023, with prices gradually rising to around 22 yuan. The PR model predicted slight price fluctuations, with a slight rebound in 2024 followed by stabilization, and prices slightly higher than the actual data in 2025. The PCFR model was the most conservative in its price forecasts, predicting a brief decline at the beginning of 2024 followed by stabilization, which closely matches the overall trend of the actual data. Figure 9 Quarterly price forecast for chili peppers without stems.

[0173] Figure 10This paper presents three models—linear regression (LR), multinomial regression (PR), and principal component regression (PCR)—to predict monthly chili pepper prices, covering the period from November 2023 to September 2024. Prices fluctuated slightly between January and June 2024, reaching a peak in July 2024 (approximately 24 yuan), before rapidly declining to a low of approximately 20 yuan in September 2024. This trend indicates a rapid market correction after a short-term price increase. The LR model predicts prices will stabilize after mid-2024 with minimal fluctuations; the PR model shows more flexibility in predicting price changes. It captured the price fluctuation trend from June to August 2024, with predictions close to actual data at the peak; the PCFR model's price fluctuation predictions are closest to actual data. It captured the rapid changes from July to September 2024 and provides a reasonable prediction for the stabilization period after September 2024. The LR model cannot reflect sharp fluctuations in the short term and has a large gap with the actual data, while the PR and PCFR models can capture the fluctuation characteristics of the actual data better. In particular, the PCFR model performs more closely to the actual situation during periods of rapid rise and fall. Figure 10 Monthly price forecast for chili peppers without stems.

[0174] 3) Price forecast for stemless red peppers:

[0175] Figure 11 Annual price forecast for bell peppers without stems Figure 11 The chart shows the predictive analysis results of three different models—linear regression (LR), multinomial regression (PR), and principal component regression (PCR)—on the annual average price of stemless bell peppers, spanning from 2021 to 2026. As can be seen from the chart, the actual price rose rapidly from approximately 20 yuan in 2021 to approximately 24 yuan in 2022, before declining slightly to approximately 23 yuan in 2023. The overall trend is a rapid rise followed by a slight decline, possibly reflecting a gradual stabilization of market demand after its initial growth. The LR model predicts a continued slow price increase, reaching approximately 25 yuan in 2026. The PR model predicts a slight decline in price in 2024 followed by a gradual recovery, but the predicted value for 2026 is lower than that of the LR model (approximately 24.5 yuan). The PCFR model predicts a low point in price in 2024 followed by a gradual recovery, but with a more conservative overall increase, reaching approximately 24 yuan in 2026. In practical applications, the choice of which model to use often requires comprehensive consideration of data characteristics, model predictive performance, and the need for model interpretability.

[0176] Using this system to analyze these price forecasts can help growers and market analysts better plan future planting and sales strategies. For example, when a decline in the price of stemless bell peppers is predicted, growers can take effective countermeasures in advance, such as reducing the planting area of ​​stemless bell peppers, thereby reducing potential economic losses.

[0177] Figure 12 Quarterly price forecast for bell peppers without stems, from Figure 12 As can be seen, prices experienced a significant rise from the beginning of 2022 to the beginning of 2023, reaching a peak (approximately RMB 27). From the second quarter of 2023, prices began to decline rapidly, falling to approximately RMB 18 by the fourth quarter of 2023. The LR model predicts that prices will slowly recover after reaching a low point in 2023, stabilizing in 2025 (approximately RMB 23); the PR model predicts slight fluctuations in early 2024, followed by a significant rebound in mid-2024 (approximately RMB 25), and then gradually stabilizing (approximately RMB 22); the PCFR model predicts similarities to PR, but is more conservative regarding the rebound in mid-2024, with prices gradually stabilizing after 2024 (approximately RMB 21.5).

[0178] Such price fluctuations may reflect seasonal factors, changes in market supply and demand, fluctuations in production costs, or other possible factors. For example, a price decrease in stemless chili peppers during a particular season may be due to an overproduction of stemless chili peppers that season, resulting in a supply far exceeding demand. Conversely, a seasonal increase in stemless chili pepper prices may be related to a decrease in production during those seasons. These forecasts are significant for chili pepper growers and distributors. For instance, growers can adjust their planting scale based on these forecasts to maximize profits, while distributors can adjust their purchase volume and inventory to avoid losses.

[0179] Figure 13 Monthly price forecast for bell peppers without stems, from Figure 13As can be seen, from November 2023 to April 2024, the actual price showed a slow downward trend, reaching its lowest point in May 2024 (approximately RMB 18). After May 2024, the actual price began to rebound rapidly, reaching a peak in September 2024 (approximately RMB 24), and then slightly declined and stabilized. Among the three forecasting models, the LR model failed to capture the price fluctuations in the actual data, exhibiting a simple linear upward trend. During the rapid rise and subsequent fall from May to September 2024, the LR model's predictions deviated significantly from the actual data; the PR model captured the price fluctuation characteristics better, especially predicting the rapid price rebound during July to September 2024. After September 2024, the PR model predicted that the price would gradually stabilize, which was closer to the actual data; the PCFR model was more robust in predicting the rapid rise and fall of prices in 2024, especially at the peak in September 2024, where the PCFR prediction was closer to the actual data. After October 2024, PCFR price forecasts stabilized, largely matching actual data.

[0180] Figure 14 Weekly price forecast for bell peppers without stems. Figure 14 During the period, all models predicted a similar price trend for bell peppers without stems, indicating a high degree of consistency between the models' interpretation and prediction of the data. From September 4th to December 9th, 2023, the actual price showed a continuous downward trend. The price gradually decreased from 26 yuan to 23 yuan, with a slight rebound in mid-October to early November, but then continued to fall. The LR model predicted a relatively smooth downward trend in prices, failing to capture the actual rebound in late October to early November. Because linear regression assumes a linear trend in price changes, it is insufficient in reflecting short-term non-linear fluctuations (such as rebounds). The PR model predicted a rebound within the downward price trend, and the predicted rebound magnitude was relatively close to the actual data. The PR model can capture some non-linear fluctuation trends, but its rebound magnitude was slightly higher than the actual data. The PCFR model had the highest fit to the actual data, especially during the rebound in mid-October to early November and the subsequent decline, where its predicted curve was closer to the actual price.

[0181] Application effectiveness

[0182] The research and application of chili price forecasting models have yielded significant results. They have not only improved the overall efficiency of the agricultural economy but also directly increased farmers' income and optimized the operation mechanism of the agricultural product market. This model, by integrating big data analysis, machine learning, and economic forecasting technologies, provides a powerful decision support tool for all participants in the chili industry chain.

[0183] Improving agricultural economic efficiency: This predictive model helps farmers understand real-time market data analysis, enabling them to understand immediate fluctuations and future trends in chili prices, allowing them to plant and sell at the optimal time to maximize profits. Furthermore, the model assists agricultural processing enterprises and distributors in making more accurate decisions regarding procurement and inventory management, reducing waste.

[0184] Increasing farmers' income: Accurate price forecasting enables farmers to mitigate the risks of price fluctuations. By rationally planning production and sales, farmers can avoid selling their products when prices are low, ensuring higher economic returns. At the same time, this also encourages farmers to adopt more scientific planting methods and technologies, further improving yield and quality.

[0185] Optimizing agricultural product market operations: The model provides valuable data support for government and market regulatory agencies, enabling them to understand market supply and demand in a timely manner, rationally regulate the market, and prevent price manipulation and abnormal fluctuations. This efficient market supervision not only protects the interests of consumers but also safeguards the rights and interests of farmers.

[0186] Promoting Sustainable Development: Through continuous data analysis and trend forecasting, the model helps with long-term planning and resource allocation in the chili pepper industry. For example, forecast data can guide agricultural policy formulation, such as setting reasonable subsidy policies, optimizing the allocation of agricultural resources, and supporting agricultural technological innovation, thereby promoting the sustainable development of the entire industry.

Claims

1. A method of chili price prediction, characterized by; Database construction of weather data, historical price data and market supply and demand relationship; Construction of polynomial cross feature regression (PCFR) model; Determine the chilli price index, and train the generated deep neural network to form a chilli price prediction model.

2. The method of claim 1, wherein; The construction of the database includes: periodically reading the chilli price data from the Guizhou Agricultural Economic Network database interface, periodically crawling the chilli price data from the specified website, and importing the cleaned data into the database; Obtain weather data from the cimiss interface and import it into the background database.

3. The method of claim 1, wherein the price of the pepper is predicted. The construction of the polynomial cross feature regression (PCFR) model includes: ① Linear regression (LR) This model assumes that there is a linear relationship between the independent variable and the dependent variable, that is, the dependent variable can be represented by the linear combination of the independent variable, and the formula is as follows: y = β0+ β1x1+ β2x2+ β3x3+... + β n x n + ∈ y is the dependent variable (forecast target), β0 is the constant term, β1, β2,..., β n are the coefficients of the independent variables, x1, x2,..., x n are the independent variables (features); ∈ is the error term, representing the part that the model cannot explain; in the prediction of agricultural product prices, factors such as weather may have a linear relationship with prices, so linear regression can provide a basic model; solves the problem of establishing a linear relationship between the target variable and the explanatory variable, providing a basic model; ② Polynomial regression (PR) There is a single explanatory variable x and a target variable y, and their relationship can be expressed as: y = β0+ β1x + β2x 2 + β3x 3 + … + β n x n + ∈ where β0, β1, β2... β n are regression coefficients and ∈ is the error term; by fitting such a polynomial, we can better describe the relationship between the target variable y and x; Polynomial transformation of weather; Polynomial regression can better handle the non-linear relationship between the target variable and the explanatory variable by introducing high-order terms of the explanatory variable; In agricultural product price prediction, there may be some nonlinear trends, and polynomial regression can help more flexibly fit these relationships; Solve the non-linear relationship between the target variable and the explanatory variable, and more flexibly fit the data by introducing high-order terms; ② Polynomial cross feature regression (PCFR) Polynomial cross feature regression is a further improvement of polynomial regression, which better captures the interaction between different features by introducing cross terms between features; In agricultural product price prediction, there may be complex relationships between different weather factors, and PCFR can help better capture these relationships; Solve the interaction problem between features and more comprehensively capture the relevance between different features; In polynomial regression, it is assumed that the relationship between the dependent variable y and the independent variables x1, x2, …, x n can be represented by a polynomial function; Its basic form is where β i is the coefficient to be estimated and ∈ is the error term; In polynomial cross feature regression, in addition to the polynomial terms of individual features, the interaction between features is also considered; Assuming there are two features x1 and x2, the introduction of cross terms can be expressed as: In this formula, β 12 x1x2 represents the interaction between features x1 and x2; by introducing these cross terms, PCFR is able to capture more complex relationships, especially in agricultural product price prediction, where there can be complex associations between different weather factors.

4. The method of claim 1, wherein the price of the pepper is predicted. Determination of chilli price index includes: Determination of the influencing factors of chilli price The division of the year type is based on the temperature anomaly degree of the previous year and the precipitation anomaly percentage of the previous year; The calculation method of precipitation anomaly percentage is as follows: Calculate the percentage of precipitation in a certain period to the average precipitation in the same period, the formula is: P is the annual precipitation, Pi is the average precipitation in the same period, and the average value of 5 years is taken; According to the size of precipitation anomaly percentage, the precipitation anomaly degree is divided; The calculation of temperature anomaly degree is as follows: (1) First calculate the average value In the formula: Yi—5-year average temperature (℃) value; Yp—average temperature (℃) value in the corresponding period; n—data sequence sample years, take 5 years; (2) Then calculate the standard deviation s Finally, calculate the anomaly degree C Then combine the temperature anomaly degree and the precipitation anomaly degree to divide the year type, which is used as the input of the neural network model, represented by a two-dimensional vector year_type=[Y1,Y2], indicating that the under-collection year is represented by [1,0], the bumper year is represented by [0,1], and the normal year is represented by [0,0]; The determination of the chilli price index is as follows: First, the average price of the same kind of pepper in different markets nationwide is calculated, and then the price index is calculated according to the average price. The price of a certain period is set as the base period price, and the index value of this month is indexj=1000. The average price of pepper on a certain day is Pi, and the price index of that day is indexi=Pi / Pj×1000. Through the correlation analysis of the price changes of a certain kind of pepper and the price index of the same kind of pepper nationwide, it is found that the price index of the same kind of pepper nationwide reflects the overall market situation of each market, and has a certain predictability for the price changes of pepper in Guizhou. Therefore, the price index of the same kind of pepper nationwide is directly used as one of the inputs of the neural network model; Pepper price prediction feature selection and feature engineering problems; Through analysis, the correlation between price and the constructed single meteorological index reflecting short-term meteorology is weak or non-existent, but considering historical climate factors, the annual type feature is used as one of the inputs. A set of price index system is constructed, which reflects the comprehensive influence of meteorological factors, industry supply and demand, national similar price, same period price of last year, market factors, policy factors and other factors on pepper price. The price index is also used as one of the inputs of the neural network model. In order to include the influence of regional meteorological factors on pepper production and transportation into the model, convolutional neural network is used to extract features from grid meteorological high-dimensional data during modeling; Three models are used for weekly price prediction: ① Linear regression (LR), based on past historical price data, to predict the weekly price of the next 4 weeks; ② Polynomial regression (PR), taking the price curve of the previous three weeks, annual type and price index as features, to predict the weekly price of the next 4 weeks; ③ Polynomial cross-feature regression (PCFR) input is a vector composed of the past 3 weeks' price, annual type affected by climate factors, price index and grid meteorological features extracted by convolutional neural network, and the output is the weekly price of the next 4 weeks. The PCFR model and deep neural network are used for monthly price prediction, and the monthly price of the next 12 months is predicted. LR input is a vector composed of the past 12 months' monthly price curve and annual type, price index and grid meteorological features extracted by convolutional neural network affected by climate, and the output is the monthly price of the next 12 months. ​

Images