Spring tea price real-time prediction method and system

Abstract

The invention provides a spring tea price real-time prediction method and system, and belongs to the technical field of computers, and the method comprises the steps: obtaining spring tea historical price data and influence factor data in a selected time period before prediction, and dividing the data into time series data and non-time series data; extracting a dependency relationship among the multiple pieces of time sequence data to obtain association features of the time sequence data under different time steps; performing multi-layer linear transformation calculation on the multiple pieces of non-time-series data to obtain a feature vector reflecting the comprehensive influence of the non-time-series data on the price; and carrying out feature fusion on the associated features of the time sequence data under different time steps and the feature vectors reflecting the comprehensive influence of non-time sequence data on the price, and carrying out multilayer linear transformation and activation function calculation processing on the fused features to obtain a predicted spring tea price fluctuation trend. The dynamic price of the tea market is closely tracked, picking, storage and selling of tea are reasonably arranged, and therefore the economic benefits of the tea industry are improved.

Description

Technical Field

[0001] The invention belongs to the field of computer processing, and in particular relates to a method and system for real-time prediction of spring tea prices. Background Art

[0002] As one of China's top ten famous teas, Xinyang Maojian's price is influenced by a variety of factors, including its growing environment, production process, quality, market demand and supply, market competition, and economic factors. Xinyang Maojian spring tea generally refers to the new tea leaves roasted from the tea plant between late March and April. The price of spring tea is directly related to the income of tea growers. When spring tea prices are high and yields are good, growers' incomes naturally increase; otherwise, their incomes decrease. The supply and demand conditions in the tea market also affect the price of Xinyang Maojian. When market demand exceeds supply, tea prices may rise; conversely, when supply exceeds demand, prices may fall. Tea prices are also an important factor influencing growers' adjustments to their planting structure and variety selection. If a certain type of tea has a high price, growers may prefer to plant that type to meet market demand and achieve higher economic returns.

[0003] For tea growers and tea dealers, closely tracking tea market price fluctuations is essential to developing effective planting and sales strategies and mitigating market risks. Therefore, predicting spring tea price fluctuations can help growers and tea dealers proactively sell tea at the right time or store it until the market recovers, minimizing losses. However, existing methods lack direct predictions of spring tea price fluctuations. Summary of the Invention

[0004] In order to solve the problem that the existing technology lacks real-time prediction of spring tea prices, the present invention provides a real-time prediction method, system, device and storage medium for spring tea prices.

[0005] In order to achieve the above object, the present invention provides the following technical solutions: A method for real-time prediction of spring tea prices comprises the following steps: Obtaining historical price data and influencing factor data of spring tea within a selected time period before the prediction; dividing the historical price data and influencing factor data of spring tea into multiple time series data and multiple non-time series data according to data categories; Extracting the dependency relationships between the multiple time series data to obtain correlation features of the time series data at different time steps; performing multi-layer linear transformation calculations on the multiple non-time series data to obtain a feature vector reflecting the comprehensive impact of the non-time series data on prices; The correlation features of the time series data at different time steps and the feature vectors reflecting the comprehensive impact of non-time series data on prices are fused, and the fused features are subjected to multi-layer linear transformation and activation function calculation processing to obtain the predicted price fluctuation trend of spring tea.

[0006] Preferably, after obtaining the historical price data and influencing factor data of spring tea in the selected time period before the prediction, the method further includes preprocessing the historical price data and influencing factor data of spring tea, specifically including the following steps: Dividing the spring tea historical price data and influencing factor data into multiple time series data and multiple non-time series data; Calculating the mean of the multiple time series data and the multiple non-time series data, filtering out abnormal data, and replacing them with the mean; and then using interpolation to fill in the missing data; The completed data is normalized using the minimum-maximum normalization method; and the normalized data is smoothed using a set time window; Dependency extraction is performed on the smoothed time series data, and linear transformation is performed on the non-time series data.

[0007] Preferably, the historical price data and influencing factor data of spring tea in a selected time period before prediction are processed by a spring tea price prediction model to obtain a predicted spring tea price fluctuation trend, and the spring tea price prediction model includes a long short-term memory network LSTM module, a fully connected neural network FCNN module and a price prediction module; the long short-term memory network LSTM module is used to capture the long-term dependency of the multiple time series data and obtain the correlation features of the time series data at different time steps; the fully connected neural network FCNN module is used to perform multi-layer linear transformation calculations on the multiple non-time series data to obtain a feature vector reflecting the price impact; the price prediction module is used to perform feature fusion on the correlation features of the time series data at different time steps and the feature vector reflecting the comprehensive impact of the non-time series data on the price, and perform multi-layer linear transformation and activation function calculation processing on the fused features to obtain a predicted spring tea price fluctuation trend.

[0008] Preferably, the price prediction module includes multiple fully connected layers, and each fully connected layer is connected through a nonlinear activation function ReLU.

[0009] Preferably, the method further comprises constructing a training set to train the spring tea price prediction model, wherein the training set construction comprises the following steps: Obtain historical price data and influencing factors of spring tea; Enhance the spring tea historical price data and influencing factor data by performing data enhancement through disturbance factors to obtain new sample data; The new sample data is combined with the spring tea historical price data and influencing factor data to form a training set.

[0010] Preferably, the data enhancement by using the perturbation factor specifically includes the following steps: Obtaining an original sample vector based on the historical price data of spring tea and the influencing factor data, and constructing a disturbance factor vector of the same dimension; For each position of the perturbation factor vector, generate a random number between 0 and 1, and set the value greater than 0.9 to 1 and the value less than 0.9 to 0; Iterate the original sample vector and the perturbation factor vector in sequence to generate new sample data.

[0011] Preferably, the multiple time series data specifically include historical prices, natural climate, competitor prices, picking and processing costs, market sales volume and speculative atmosphere scores; the multiple non-time series data specifically include initial fixed investment and brand effect scores.

[0012] Preferably, the method further comprises using a population distribution test method to test the rationality of the new sample data.

[0013] Preferably, the method further comprises using root mean square error and Pearson correlation coefficient as evaluation criteria to evaluate the spring tea price prediction model.

[0014] The present invention also provides a real-time prediction system for spring tea prices, which specifically includes: The data processing module is used to obtain the historical price data and influencing factor data of spring tea in the selected time period before the prediction; and divide the historical price data and influencing factor data of spring tea into multiple time series data and multiple non-time series data according to the data category.

[0015] A trend prediction module is used to extract the dependency relationships between the multiple time series data to obtain the correlation features of the time series data at different time steps; perform multi-layer linear transformation calculations on the multiple non-time series data to obtain a feature vector reflecting the comprehensive impact of the non-time series data on prices; The correlation features of the time series data at different time steps and the feature vectors reflecting the comprehensive impact of non-time series data on prices are fused, and the fused features are subjected to multi-layer linear transformation and activation function calculation processing to obtain the predicted price fluctuation trend of spring tea.

[0016] The method for real-time prediction of spring tea prices provided by the present invention has the following beneficial effects: The present invention obtains price information and influencing factor information of spring tea, divides the data into time series data and non-time series data, considers the characteristic variables affecting the price of spring tea in multiple dimensions, and enhances the reliability of the data. The dependency relationship between multiple time series data is captured to obtain the correlation characteristics of multidimensional data, and multiple non-time series data are subjected to multi-layer linear transformation calculations. The non-time series data are abstracted to reflect the characteristic vector of the comprehensive impact of the non-time series data on the price, and the nonlinear relationship in the data is considered. The correlation characteristics and the characteristic vector of the comprehensive impact on the price are further integrated and calculated to obtain the predicted spring tea price fluctuation trend. Real-time tracking and prediction of spring tea price trends are achieved. It can not only provide decision-making solutions for tea growers and tea dealers in picking, frying and selling new tea, but also contribute to the dynamic stability of the Xinyang Maojian spring tea market price and scientifically adjust the planting structure and variety selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the embodiments of the present invention and its design, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort.

[0018] Figure 1 This is an overall flow chart of a method for real-time prediction of spring tea prices according to an embodiment of the present invention.

[0019] Figure 2 2 is a model architecture diagram of the spring tea price prediction model in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of generating a sample vector from an original sample vector according to a disturbance factor vector in an embodiment of the present invention.

[0021] Figure 4 Schematic diagram of generating new samples based on anchor point samples in an embodiment of the present invention.

[0022] Figure 5 3. This is a schematic diagram comparing the root mean square errors of price prediction values ​​of the method of the present invention, the multiple linear regression method, and the multiple nonlinear regression method on different dates in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the technical solution of the present invention and to be able to implement it, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] Example The present invention provides a method for real-time prediction of spring tea prices. Figure 1As shown, the specific steps include: Step 1: Collect factors influencing the price of Xinfa Maojian spring tea and obtain relevant data.

[0025] Through the analysis of the price of Xinyang Maojian spring tea, it is determined that the factors affecting the price of spring tea are mainly basic price, natural climate, price of competing products, picking and frying level and price, market sales, speculative atmosphere, initial fixed input cost, and brand effect.

[0026] (1) Basic price factors: Collect the price of Xinyang Maojian spring tea in the past five years. Including the price data of Xinyang Maojian spring tea from different production areas every day from March 20 to April 30. (2) Natural climate factors: Collect the temperature data of Xinyang area every day from March 20 to April 30, including the highest temperature, lowest temperature, precipitation, and humidity. (3) Competitive product price factors: Collect the price data of Jiangsu Biluochun, Enshi Yulu, West Lake Longjing, Huangshan Maofeng, Baihao Yinzhen, and Lu'an Guapian spring tea every week from March 20 to April 30. (4) Picking and roasting cost factors: Collect the market data of Xinyang Maojian roasting level from March 20 to April 30. According to the quality evaluation of professional tea roasting masters, buyers and consumers on the newly roasted Maojian green tea, set three evaluation standards: poor, ordinary, and good; Collect the labor cost price of picking and roasting in Xinyang area every day from March 20 to April 30. (5) Market sales: Daily Xinyang Maojian green tea sales data from March 20 to April 30 were collected. (6) Speculation atmosphere: Market speculation atmosphere data from March 20 to April 30 were collected, and three evaluation criteria were set: weak, average, and strong, based on the reluctance of green tea growers to sell and the rush of buyers to buy. (7) Initial fixed input factors: Initial labor costs for weeding, pruning, tilling, and fertilizing were collected from November of the previous year to March of the current year. (8) Brand effect factors: New Xinyang Maojian tea from well-known brands tends to be more expensive, and brand effect can increase the popularity and recognition of tea. Brand effect was set as average, good, and excellent.

[0027] Step 2: Data preprocessing: Verify and preprocess the data collected in step 1.

[0028] For abnormal data (data that is obviously inconsistent with the actual situation), the mean is used to replace it. If some data in a certain time period are missing, these missing data are supplemented by interpolation. If all data in a certain time period are missing, they are supplemented by using data from the corresponding time period in previous years. The evaluation standards for the spring tea frying level are set to 0.2, 0.5 and 0.8 respectively for poor, average and good; the evaluation standards for the speculative atmosphere are set to 0.2, 0.5 and 0.8 respectively; the evaluation standards for the brand effect are set to 0.5, 0.7 and 0.9 respectively for average, good and excellent. For the data (features) obtained in each dimension, the Min-Max Normalization method is used. The calculation formula is:

[0029] ; in, Represents the raw data of a certain dimension, and Represent the maximum and minimum values ​​of the dimension data respectively. Represents the data after min-max normalization.

[0030] To fully statistically analyze and describe the time series data involved in this paper, we calculated the mean and variance of all time series data. We also used time windows of 3 days (representing the short term), 5 days (representing the medium term), and 9 days (representing the long term) to smooth the data.

[0031] Step 3: Data enhancement based on random perturbations. In order to simulate the dynamic changes of factors in real scenarios and make up for the insufficient number of original samples, the present invention adopts a data enhancement algorithm based on random perturbations, which specifically includes the following steps:

[0032] S31. Obtain background information on the price and influencing factors of the original collected sample data.

[0033] S32. For each original sample vector, which includes price information and influencing factor information, construct a perturbation factor vector of the same dimension (the default value is 0). Then, for each position in the perturbation factor vector, generate a random number between 0 and 1, and set values ​​greater than 0.9 to 1 and values ​​less than 0.9 to 0. That is, each position in the original sample vector is perturbed with a probability of 10%.

[0034] S33. Generate a new sample vector based on the original sample vector and the perturbation factor vector. Figure 3As shown in the figure. When the perturbation factor at a position is 0, the value at that position remains unchanged; when the perturbation factor is 1, the value at that position changes. Specifically, in the feature space constructed from all original samples, with the current sample as the anchor point, the six samples with the closest Euclidean distance are selected as learning samples. The data values ​​of these six learning samples at the current position are obtained. The average of these six data values ​​is used as the new data value for that position.

[0035] S34: Iterate the original data vector and the perturbation factor vector in sequence until new data values ​​are generated at all positions where the perturbation factor is 1. At this point, a new sample data vector is obtained.

[0036] S35. Use the Mann-Whitney U test, which is a population distribution test method, to evaluate the distribution difference and check the rationality of the generated sample. The Mann-Whitney U test is used to infer whether there is a difference in the population distribution of the original sample (i.e., the anchor sample) and the generated sample. When the p-value obtained by the hypothesis test is > 0.05, that is, there is no significant difference between the two samples, the generated sample is retained. Otherwise, the generated sample is discarded. Figure 4 shown.

[0037] Step 4: The newly generated sample data from Step 3 is confounded with the original sample data to create a confounded dataset. 80% of the confounded dataset is randomly selected as the training set, and the remaining 20% ​​is the validation set. The training set is primarily used for model building, while the validation set is used to test the model's accuracy and generalization performance.

[0038] Step 5: Build a price prediction model based on the LSTM model, specifically including a long short-term memory (LSTM) module, a fully connected neural network (FCNN) module, and a price prediction module. The LSTM module is trained using the time series data in the training set to capture the long-term dependencies between multiple time series data. The FCNN module is trained using the non-time series data in the training set, performing multi-layer linear transformations on the non-time series data. The price prediction module is trained using the feature values ​​output by the first two modules as input, resulting in an optimized model, the spring tea price prediction model.

[0039] The price prediction model is constructed using a Long Short-Term Memory (LSTM) network. LSTM is a variant of recurrent neural networks that effectively handles the vanishing and exploding gradient problems that occur in traditional recurrent neural networks. Furthermore, LSTM incorporates storage units and gating mechanisms, making it highly effective at capturing and processing dependencies in time series data.

[0040] Time series data (including basic prices, natural climate, competitor prices, picking and processing levels and prices, market sales, and speculative atmosphere) are input into the LSTM module as features.

[0041] The core component of the LSTM module is the basic unit. Each basic unit includes a forget gate, an update gate, and an output gate. Specifically, the formula for the forget gate is as follows:

[0042] ; in, is the weight matrix, is the offset. The output of the forget gate Perform element-wise multiplication with the state value of the previous unit. The calculation formula of the update gate is:

[0043] ; Here, the output of the update gate is a vector between 0 and 1. In order to obtain new state information , the output of the update gate will be the same as Perform element-wise multiplication:

[0044] ; ; Current output value and the hidden state value passed to the next unit : ; ; Non-time-series data (including pre-investment and brand effect factors) is input as features into the Fully Connected Neural Network (FCNN) module. The FCNN module consists of multiple stacked fully connected layers. The neurons in each fully connected layer are connected to all neurons in the previous layer via a linear transformation function, calculated as follows:

[0045] ; in, and Represent the neurons of the previous layer and the current layer respectively; is the weight coefficient; is the offset; As the activation function, the present invention uses the nonlinear activation function ReLU.

[0046] The features output by the LSTM and FCNN modules serve as input features for the price prediction module. Similar to the FCNN module, the price prediction module also includes multiple fully connected layers, with interlayer activation using the nonlinear ReLU activation function. The output of the final layer of the price prediction module is the predicted real-time price of Xinyang Maojian green tea.

[0047] The root mean square error (RMSE) and the Pearson correlation coefficient (PCC) are used as evaluation criteria. The calculation formulas for RMSE and PCC are as follows:

[0048] ; ; in and Respectively represent the Actual and predicted Xinyang Maojian green tea prices for each date. and Respectively represent the corresponding average values. The total number of days.

[0049] RMSE is primarily used to quantify the difference between the predicted Xinyang Maojian green tea price and the actual price, while PCC is used to quantify the correlation between the two. Therefore, smaller RMSE values ​​indicate smaller errors between the predicted and actual values. PCC ranges between -1 and 1. When PCC = -1, the predicted and actual values ​​are opposite, and vice versa. PCC values ​​closer to 1 indicate closer predictions to the actual values.

[0050] Step 6: Obtain real-time spring tea price information and influencing factors, input them into the spring tea price prediction model, and obtain the predicted spring tea price fluctuation trend.

[0051] To further verify the accuracy of the proposed method, the method was compared with multiple linear regression and multiple nonlinear regression, where the multiple nonlinear regression used a logistic regression equation. Overall, the average root mean square error (RMS) of the proposed method over 42 calendar days (March 20 to April 30) was 1.5%, with a PCC of 0.83. The average RMS error of the multiple linear regression method was 3.4%, with a PCC of 0.66; the average RMS error of the multiple nonlinear regression method was 2.3%, with a PCC of 0.74.

[0052] like Figure 5The figure shows a comparison of the root mean square error (RMS) of price predictions for Xinyang Maojian spring tea using the method presented in this paper, multiple linear regression, and multiple nonlinear regression methods for different dates. Overall, the RMS errors for Xinyang Maojian spring tea price predictions by the three methods show a decreasing trend with increasing date. This is because in late March, due to a late spring cold snap, sudden temperature fluctuations, and varying precipitation, the growth, yield, and harvest of the first wave of Xinyang Maojian spring tea were subject to significant uncertainty. This uncertainty was reflected in growers' reluctance to sell and market speculation, further exacerbating price fluctuations. With the stabilization of the spring climate and the gradual increase in spring tea production, growers, distributors, and customers have increasingly converged on their understanding of the Xinyang Maojian spring tea market, reducing interfering factors. Consequently, price volatility has gradually decreased, and prediction errors have been significantly reduced. From March 20th to March 31st, the RMS error of the price prediction method presented in this paper ranged from 2% to 3.5%. In comparison, the root mean square error of the multivariate linear regression method is between 4% and 8%, and the root mean square error of the multivariate nonlinear regression method is between 3% and 6%. After April 16, the root mean square error of the price forecast method of the present invention was below 1%, demonstrating high accuracy and effectiveness.

[0053] The present invention also provides a real-time prediction system for spring tea prices, which specifically includes: The data processing module is used to obtain the historical price data and influencing factor data of spring tea in the selected time period before the prediction; and divide the historical price data and influencing factor data of spring tea into multiple time series data and multiple non-time series data according to the data category.

[0054] The trend prediction module is used to extract the dependencies between multiple time series data and obtain the correlation features of time series data at different time steps; it performs multi-layer linear transformation calculations on multiple non-time series data to obtain feature vectors that reflect the comprehensive impact of non-time series data on prices; The correlation features of time series data at different time steps and the feature vectors reflecting the comprehensive impact of non-time series data on prices are fused, and the fused features are subjected to multi-layer linear transformation and activation function calculation to obtain the predicted price fluctuation trend of spring tea.

[0055] It should be pointed out that the specific implementation methods described above can enable those skilled in the art to understand the invention more comprehensively, but do not limit the invention in any way. Therefore, although the present specification and examples have described the invention in detail, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the scope of protection of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved. Any simple change or equivalent replacement of the technical solution that can be obviously obtained by any person familiar with the art within the technical scope disclosed in the present invention falls within the scope of protection of the present invention.

Claims

1. A method for real-time prediction of spring tea prices, characterized in that: The following steps are involved: Obtaining historical price data and influencing factor data of spring tea within a selected time period before the prediction; dividing the historical price data and influencing factor data of spring tea into multiple time series data and multiple non-time series data according to data categories; Extracting the dependency relationship between the multiple time series data to obtain correlation features of the time series data at different time steps; Performing multi-layer linear transformation calculations on the plurality of non-time series data to obtain a feature vector reflecting the comprehensive impact of the non-time series data on the price; The correlation features of the time series data at different time steps and the feature vectors reflecting the comprehensive impact of non-time series data on prices are fused, and the fused features are subjected to multi-layer linear transformation and activation function calculation processing to obtain the predicted price fluctuation trend of spring tea.

2. A method for real-time prediction of spring tea price according to claim 1, characterized in that: After obtaining the historical price data and influencing factor data of spring tea in the selected time period before the prediction, the process also includes preprocessing the historical price data and influencing factor data of spring tea, specifically including the following steps: Dividing the spring tea historical price data and influencing factor data into multiple time series data and multiple non-time series data; Calculating the mean of the multiple time series data and the multiple non-time series data, filtering out abnormal data, and replacing them with the mean; and then using interpolation to fill in the missing data; The completed data is normalized using the minimum-maximum normalization method; and the normalized data is smoothed using a set time window; Dependency extraction is performed on the smoothed time series data, and linear transformation is performed on the non-time series data.

3. A method for real-time prediction of spring tea price according to claim 1, characterized in that: The spring tea price prediction model is used to process the historical price data and influencing factor data of spring tea in a selected time period before prediction to obtain a predicted spring tea price fluctuation trend. The spring tea price prediction model includes a long short-term memory network LSTM module, a fully connected neural network FCNN module and a price prediction module; the long short-term memory network LSTM module is used to capture the long-term dependency of the multiple time series data and obtain the correlation features of the time series data at different time steps; the fully connected neural network FCNN module is used to perform multi-layer linear transformation calculations on the multiple non-time series data to obtain a feature vector reflecting price impact; the price prediction module is used to perform feature fusion on the correlation features of the time series data at different time steps and the feature vector reflecting the comprehensive impact of non-time series data on prices, and perform multi-layer linear transformation and activation function calculation processing on the fused features to obtain a predicted spring tea price fluctuation trend.

4. A method for real-time prediction of spring tea price according to claim 3, characterized in that: The price prediction module includes multiple fully connected layers, and each fully connected layer is connected by a nonlinear activation function ReLU.

5. A method for real-time prediction of spring tea price according to claim 3, characterized in that: The method also includes constructing a training set to train the spring tea price prediction model, wherein the training set construction includes the following steps: Obtain historical price data and influencing factors of spring tea; Enhance the spring tea historical price data and influencing factor data by performing data enhancement through disturbance factors to obtain new sample data; The new sample data is combined with the spring tea historical price data and influencing factor data to form a training set.

6. A method for real-time prediction of spring tea price according to claim 5, characterized in that: The data enhancement by the perturbation factor specifically includes the following steps: Obtaining an original sample vector based on the historical price data of spring tea and the influencing factor data, and constructing a disturbance factor vector of the same dimension; For each position of the perturbation factor vector, generate a random number between 0 and 1, and set the value greater than 0.9 to 1 and the value less than 0.9 to 0; Iterate the original sample vector and the perturbation factor vector in sequence to generate new sample data.

7. A method for real-time prediction of spring tea price according to claim 1, characterized in that: The multiple time series data specifically include historical prices, natural climate, competitor prices, picking and processing costs, market sales volume and speculative atmosphere scores; the multiple non-time series data specifically include initial fixed investment and brand effect scores.

8. A method for real-time prediction of spring tea price according to claim 5, characterized in that: It also includes using a population distribution test method to test the rationality of the new sample data.

9. A method for real-time prediction of spring tea price according to claim 3, characterized in that: It also includes using root mean square error and Pearson correlation coefficient as evaluation criteria to evaluate the spring tea price prediction model.

10. A real-time prediction system for spring tea prices, characterized in that: include: The data processing module is used to obtain the historical price data and influencing factors of spring tea in the selected time period before the forecast; Dividing the spring tea historical price data and influencing factor data into multiple time series data and multiple non-time series data according to data categories; A trend prediction module is used to extract the dependency relationship between the multiple time series data and obtain the correlation features of the time series data at different time steps; Performing multi-layer linear transformation calculations on the plurality of non-time series data to obtain a feature vector reflecting the comprehensive impact of the non-time series data on the price; The correlation features of the time series data at different time steps and the feature vectors reflecting the comprehensive impact of non-time series data on prices are fused, and the fused features are subjected to multi-layer linear transformation and activation function calculation processing to obtain the predicted price fluctuation trend of spring tea.

Images