Method for realizing spot market day-ahead electricity price prediction based on hybrid model

Abstract

The invention discloses a method for realizing spot market day-ahead electricity price prediction based on a hybrid model, and the method comprises the following steps: 1, carrying out the modal decomposition of electricity price data to be predicted through employing a CEEMDAN algorithm, and obtaining electricity price prediction reference data; 2, extracting and predicting time-space characteristics of the electricity price prediction reference data by using a CNN prediction model and an LSTM prediction model, and outputting a high-frequency IMF component prediction result; 3, realizing time sequence characteristic extraction and prediction of the non-high-frequency IMF component through an ARIMA prediction model, and outputting a prediction result for the non-high-frequency IMF component; and 4, respectively reading prediction results in the step 2 and the step 3, and integrating and evaluating the prediction results through the hybrid model. Respective operation rules of different markets can be adapted, and respective prediction effects are improved; prediction results are combined, the final prediction effect is improved through the combined model, stable operation and effect guarantee of the model are facilitated, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electricity price prediction, and in particular relates to a method for realizing day-ahead electricity price prediction in a spot market based on a hybrid model. Background Art

[0002] At present, there are two common methods for predicting electricity prices: conventional statistical methods and artificial intelligence methods. The advantage of statistical methods is that they are more stable, but they are not easy to deal with the situation of sharp fluctuations in spot market prices during the operation of the electricity market, that is, the "price nail" phenomenon. They are suitable for describing the long-term universal trend of price changes; while artificial intelligence methods have a strong ability to describe nonlinear laws, but their network structure is not interpretable, and they are prone to overfitting during training. Complex experimental processes are often required for parameter optimization, and they are suitable for predicting short-term fluctuations in electricity prices. There are two main difficulties at present: one is model training. Electricity trading data has the characteristics of strong volatility and time-series dependence. There is also a requirement for stability in the prediction algorithm, and the model parameters need to be fully and effectively trained; the other is that the reference data of the electricity trading market in different regions are quite different. Most of the existing algorithms are operated under laboratory simulation conditions, and access to real data may not be applicable. Different markets have their own operating rules, such as differences in meteorological conditions, and the reference market public data is different, resulting in poor transferability of existing algorithms, and special data mining is required based on the prediction target and reference data.

[0003] Therefore, how to deal with the sharp fluctuations in spot market prices during the operation of the electricity market, simplify the experimental process, and be able to access real data while effectively reducing the impact of non-stationarity of predicted data under the operating rules of different markets has become a difficult problem that needs to be urgently solved by technicians in the field of electricity price forecasting. Summary of the invention

[0004] In view of the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a method for predicting the day-ahead electricity price in the spot market based on a hybrid model, which can adapt to the operating rules of different markets, effectively reduce the impact of non-stationarity of forecast data, and improve forecast accuracy.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for realizing the day-ahead electricity price forecasting in the spot market based on a hybrid model, comprising the following steps: Step 1: Use the CEEMDAN algorithm to perform modal decomposition on the electricity price data to be predicted to obtain reference data for electricity price prediction; Step 2: Use the CNN prediction model and LSTM prediction model to extract and predict the spatiotemporal characteristics of the electricity price prediction reference data, and output the high-frequency IMF component prediction results; Step 3: Extract and predict the time series characteristics of non-high-frequency IMF components through the ARIMA prediction model, and output the prediction results for non-high-frequency IMF components; Step 4: Read the prediction results of step 2 and step 3 respectively, and integrate and evaluate the prediction results through the hybrid model.

[0006] In the above-mentioned method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model, in the step 1, the specific process of modal decomposition of the forecasted electricity price data is as follows: (1) Adaptively decompose the historical curve of the electricity price data to be predicted. The decomposition result is a series of signal features with different frequencies, called IMF components, which form the basis for subsequent predictions. (2) Based on the IMF component, the zero-crossing rate is evaluated, the high frequency is judged as high quality, and the high-frequency IMF components are merged and reconstructed into an overall high-frequency component.

[0007] In the above-mentioned method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model, in the step 2, the specific process of extracting and forecasting the spatiotemporal characteristics of the electricity price forecast reference data using the CNN forecasting model and the LSTM forecasting model is as follows: (1) Collect publicly disclosed market data and weather data from cities in the corresponding provinces, aggregate and analyze the data, and use them as input data for the prediction algorithm. (2) For high-frequency IMF components, deep neural network prediction models, CNN prediction models and LSTM prediction models are built based on the keras library; (3) Use the CNN model to extract the spatial characteristics of the input data, use the LSTM prediction model to extract the temporal characteristics of the input data, and output the high-frequency IMF component prediction results.

[0008] In the above-mentioned method for realizing the day-ahead electricity price forecasting of the spot market based on the hybrid model, in the step 2, the function expression for constructing the CNN forecasting model is: ; in, To extract the features, is the sigmoid activation function, is the weight matrix, is a time series, is the bias vector.

[0009] In the above-mentioned method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model, in the step 2, the specific process of constructing the LSTM forecasting model is as follows: (1) (2) (3) (4) (5) (6) in, is the weight matrix, is the corresponding bias vector, is the hyperbolic tangent function, is the dot product, is the output at the previous moment, To retain the degree value, is the memory state of the previous moment, is the added degree value of the current state, is the intermediate state, is the current state, is the output degree value, is the output at the current moment, The input at the current moment.

[0010] In the above-mentioned method for realizing the day-ahead electricity price forecasting of the spot market based on the hybrid model, in step 3, the specific process of realizing the time series characteristic extraction and forecasting of the high-frequency IMF component through the ARIMA forecasting model is as follows: (1) Construct ARIMA forecasting model; (2) For non-high-frequency IMF components, the constructed ARIMA forecasting model is used for fitting, and the grid search method is used to determine the autoregressive, differential and moving average hyperparameters of each IMF component; (3) Use Python to search and predict the parameters of the ARIMA prediction model, use a timing trigger mechanism, and use the trained hyperparameters to output prediction results for non-high-frequency IMF components based on the latest saved CEEMDAN decomposition results.

[0011] The above-mentioned method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model, the function expression for constructing the ARIMA forecasting model is: ; Among them, in the formula: is the predicted target value; is the value of the previous period; is the mean of the time series; , is white noise, that is, the error between the observed value and the predicted value, which is usually assumed to be normally distributed; is the parameter of each white noise, which is used to measure the influence of this white noise on the current time point; , The model order needs to be specified manually; is the autocorrelation coefficient.

[0012] In the above-mentioned method for realizing the day-ahead electricity price forecasting of the spot market based on the hybrid model, the specific process of integrating and evaluating the forecast results through the hybrid model in step 4 is as follows: (1) Use Python to write a data processing program, adopt a timed trigger mechanism, read the prediction results of step 2 and step 3 respectively, and evaluate the accuracy and error rate of the prediction results for the part of the electricity price data that has been obtained. After meeting the standards, merge the results to obtain the overall electricity price prediction results; (2) If the standard is not met, the model with the latest updated date and the prediction result that meets the standard is used to execute the prediction program again, output the results and evaluate them, and push the information of unqualified prediction results to the system operation and maintenance personnel for troubleshooting.

[0013] In the above-mentioned method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model, the zero-crossing rate is the number of times the data crosses the X-axis per unit time.

[0014] The advantages of the method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model of the present invention are as follows: by means of modal decomposition, the characteristics of the electricity price series at different frequencies are extracted, thereby effectively reducing the non-stationary influence of the forecasted data; multiple curves of the modal decomposition results are differentiated according to the characteristics to input into different models, so as to find the forecasting model applicable to the corresponding data rules to the maximum extent, which can adapt to the respective operating rules of different markets and improve the respective forecasting effects; the forecasting results are merged and the final forecasting effect is improved by means of the combined model, which is conducive to the stable operation and effect guarantee of the model and improves the accuracy of the forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a method for predicting the day-ahead electricity price in the spot market based on the CEEMDAN-CNN-LSTM-ARIMA hybrid model of the present invention; Figure 2 The figure is a calculation principle diagram of the CEEMDAN algorithm in step 1 of the present invention; Figure 3 Schematic diagram of the decomposition results of signal features of different frequencies obtained by the CEEMDAN algorithm in step 1 (I); Figure 4Schematic diagram of the decomposition results of signal features of different frequencies obtained by the CEEMDAN algorithm in step 1 (II); Figure 5 Schematic diagram of the decomposition results of signal features of different frequencies obtained by the CEEMDAN algorithm in step 1 (III); Figure 6 Schematic diagram of the decomposition results of signal features of different frequencies obtained by the CEEMDAN algorithm in step 1 (IV); Figure 7 Schematic diagram of the decomposition results of signal features of different frequencies obtained by the CEEMDAN algorithm in step 1 (V); Figure 8 Schematic diagram of the decomposition results of signal features of different frequencies obtained by the CEEMDAN algorithm in step 1 (six). DETAILED DESCRIPTION

[0016] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] In the present invention, unless otherwise stated, directional words such as "upper" and "lower" generally refer to the upper and lower parts of the device in actual use or working state, specifically the drawing direction in the attached drawings; while "inside" and "outside" refer to the outline of the device. In addition, in the description of this application, the term "including" means "including but not limited to". The terms first, second, third, etc. are used only as labels and do not impose numerical requirements or establish an order. The term "plurality" means "two or more".

[0018] The invention discloses a method for realizing day-ahead electricity price forecasting in a spot market by using a hybrid model method. The method comprises: realizing a mode decomposition method of electricity price sequence data based on a fully adaptive noise integrated empirical mode decomposition (CEEMDAN) algorithm, extracting and forecasting spatial characteristics of electricity price forecast reference data based on a CNN forecast model, extracting and forecasting time series characteristics of electricity price forecast reference data based on an LSTM forecast model, and extracting and forecasting time series characteristics of low-frequency IMF components based on an ARIMA forecast model.

[0019] The details are as follows: 1. Implementing the modal decomposition method of electricity price series data based on the fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) algorithm; Through the CEEMDAN method, the historical curve of the electricity price data to be predicted is adaptively decomposed, and the decomposition result is a series of signal features with different frequencies, which form the basis for subsequent predictions.

[0020] 2. Extract and predict the spatiotemporal characteristics of electricity price prediction reference data based on CNN prediction model and LSTM prediction model; This method collects parameter data related to electricity prices as the input of the prediction algorithm, uses the CNN prediction model and the LSTM prediction model to extract the spatial characteristics and temporal characteristics of the first step results respectively, and constructs a deep neural network prediction model.

[0021] 3. Extract and predict the time series characteristics of low-frequency IMF components based on the ARIMA forecasting model; This method is based on the ARIMA forecasting model. It performs parameter fitting of autoregression, difference and moving average analysis on the results of the first step and outputs the forecast results.

[0022] 4. Integration and evaluation of prediction results based on hybrid models; The second and third step forecast results are evaluated separately and combined into the overall electricity price forecast result.

[0023] like Figure 1-3 As shown, a method for predicting the day-ahead electricity price in the spot market based on the CEEMDAN-CNN-LSTM-ARIMA hybrid model includes the following steps: Step 1: Use the CEEMDAN algorithm to perform modal decomposition on the electricity price data to be predicted to obtain the reference data for electricity price prediction; the specific process is as follows: (1) Adaptively decompose the historical curve of the electricity price data to be predicted. The decomposition result is a series of signal features with different frequencies, called IMF components, which form the basis for subsequent predictions. (2) Based on the IMF component, the zero-crossing rate is evaluated, and the high-frequency judgment is set as the best. The high-frequency IMF components are merged and reconstructed into an overall high-frequency component, where the zero-crossing rate is the number of times the data crosses the X-axis per unit time.

[0024] This step uses Python to write a data processing program and a timing trigger mechanism to perform CEEMDAN decomposition based on the electricity price data from the historical date to the latest disclosure date, generate modal decomposition results, and update the generated result data in full to the IMF result table of the day. After completion, the thread is destroyed.

[0025] This method builds a relational database MySQL in the application server, obtains market disclosed electricity price data through an external interface, and automatically triggers the local data processing script to run. The details are as follows: from PyEMD import CEEMDAN output_table_1 = input_table_1.copy() ceemdan = CEEMDAN() IMF = ceemdan.ceemdan(input_table_1['ColumnValues'].values) for i in range(len(IMF)): output_table_1['IMF{}'.format(i)] = IMF[i] like Figure 3-Figure 8 As shown in Figure 2, the decomposition result of the signal characteristics of different frequencies obtained by using the CEEMDAN algorithm to obtain the historical curve of the electricity price data to be predicted is Figure 3-Figure 8 It can be seen that the fluctuation curve gradually tends to be stable. Due to the non-stationary characteristics of electricity price data, the algorithm will first decompose the non-stationary characteristics of the electricity price curve to be predicted to form IMF0, and decompose the electricity price curve to be predicted again based on IMF0, gradually eliminating the influence of non-stationarity. That is: add white noise to the original signal to generate a new signal sequence; then perform EMD decomposition on each new signal to obtain a series of IMFs; calculate the average value of the same order IMFs of all new signals to obtain the final IMFs; subtract the obtained IMFs from the original signal to obtain a new residual signal. Repeat steps 1-4 until the residual signal meets the stopping criteria. Finally, the following are formed in turn. Figure 3-Figure 8 The six IMF component curves that gradually tend to be stable reflect the stability characteristics of the electricity price curve to be predicted at different frequencies. Finally, the zero-crossing rate is evaluated based on the IMF component, the high-frequency judgment threshold is set, the high-frequency components are merged, and reconstructed into an overall high-frequency component.

[0026] Step 2: Use the CNN prediction model and LSTM prediction model to extract and predict the spatiotemporal characteristics of the electricity price prediction reference data, and output the high-frequency IMF component prediction results; the specific process is as follows: (1) Collect publicly disclosed market data and weather data from cities in the corresponding provinces, aggregate and analyze the data, and use them as input data for the prediction algorithm. (2) For high-frequency IMF components, deep neural network prediction models, CNN prediction models and LSTM prediction models are built based on the keras library; (3) Use the CNN prediction model to extract the spatial characteristics of the input data, use the LSTM prediction model to extract the temporal characteristics of the input data, and output the high-frequency IMF component prediction results.

[0027] This step implements the construction of deep neural network CNN prediction model and LSTM prediction model based on keras library. The data processing program is written in python, and the timing trigger mechanism is adopted to read the CEEMDAN decomposition results of the latest day. The debugged model training parameters are used as the default parameters for training, and RMSE (root mean square error) is used as the loss function. The model training is implemented according to the back propagation algorithm. During model training, the ratio of training set to validation set is 8:1. By evaluating the decrease of the loss function of the validation set, it is decided whether to stop the model training. After the model training is completed, it is saved in the model folder directory, and the latest day's data is predicted. After the prediction results are generated, they are stored in the prediction result table of the day.

[0028] Among them, the function expression for constructing the CNN prediction model is: ; In the formula, To extract the features, is the sigmoid activation function, is the weight matrix, is a time series, is the bias vector.

[0029] The specific process of building an LSTM prediction model is as follows: (1) (2) (3) (4) (5) (6) In the formula, is the weight matrix, is the corresponding bias vector, is the hyperbolic tangent function, is the dot product, is the output at the previous moment, To retain the degree value, is the memory state of the previous moment, is the added degree value of the current state, is the intermediate state, is the current state, is the output degree value, is the output at the current moment, The input at the current moment.

[0030] Step 3: Use the ARIMA model to extract and predict the time series characteristics of non-high-frequency IMF components, and output the prediction results for non-high-frequency IMF components; the specific process is as follows: (1) Construct ARIMA forecasting model; (2) For non-high-frequency IMF components, the constructed ARIMA forecasting model is used for fitting, and the grid search method is used to determine the autoregressive, differential and moving average hyperparameters of each IMF component; (3) Use Python to search and predict the parameters of the ARIMA prediction model, use a timing trigger mechanism, and use the trained hyperparameters to output prediction results for non-high-frequency IMF components based on the latest saved CEEMDAN decomposition results.

[0031] The function expression for constructing the ARIMA forecasting model is: ; Among them, in the formula: is the predicted target value; is the value of the previous period; is the mean of the time series; , is white noise, that is, the error between the observed value and the predicted value, which is usually assumed to be normally distributed; is the parameter of each white noise, which is used to measure the influence of this white noise on the current time point; , The model order needs to be specified manually; is the autocorrelation coefficient.

[0032] This step uses the constructed ARIMA forecasting model to fit the non-high-frequency IMF components, and uses the grid search method to determine the autoregressive, differential and moving average hyperparameters of each IMF component. The trained hyperparameters are used to output the results for the latest data to be predicted.

[0033] Use Python to implement the parameter search and prediction program of the ARIMA prediction model, use a timed trigger mechanism, and output the prediction results for the non-high-frequency IMF components based on the latest saved CEEMDAN decomposition results. Update the generated result data to the ARIMA result table for the day, and destroy the thread after completion.

[0034] Step 4: Read the prediction results of step 2 and step 3 respectively, and integrate and evaluate the prediction results through the hybrid model. The specific process is as follows: (1) Use Python to write a data processing program, adopt a timed trigger mechanism, read the prediction results of step 2 and step 3 respectively, and evaluate the accuracy and error rate of the prediction results for the part of the electricity price data that has been obtained. After meeting the standards, merge the results to obtain the overall electricity price prediction results; (2) If the standard is not met, the model with the latest updated date and the prediction result that meets the standard is used to execute the prediction program again, output the results and evaluate them, and push the information of unqualified prediction results to the system operation and maintenance personnel for troubleshooting.

[0035] Embodiment 1: The method of the present invention is described below by taking the Gansu spot day-ahead node electricity price forecast as an example: The present invention predicts a total of 672 data points for 7 consecutive days from July 17 to July 23, 2024, with a sampling point every 15 minutes. The sliding window used for prediction is one day, that is, the historical data is used to predict one day backward to output the prediction results, and the results are evaluated, as shown in Table 1.

[0036] The evaluation indicators for the model in this project are MAPE and R 2 Among them, MAPE (Mean Absolute Percentage Error) is a measure commonly used to evaluate the error between the predicted value and the actual value. It is an important indicator to measure the accuracy of the predicted value. MAPE appears in the form of a percentage, indicating the average percentage of the relative error between the predicted value and the actual value. The value range of MAPE is [0, +∞), and the smaller the value, the more accurate the prediction model; R 2 , namely R-Squared, is often used to measure the fit of linear regression. A value close to 1 indicates a strong positive correlation between variables, a value close to -1 indicates a strong negative correlation between variables, and a value close to 0 indicates that there is not much relationship between variables. Therefore, as shown in Table 1, the CEEMDAN-CNN-LSTM prediction model proposed in this paper has achieved the best results compared with other prediction models tested in this paper.

[0037] Table 1 Gansu spot day-ahead node electricity price forecasting performance evaluation indicators Prediction Model MAPE <![CDATA[R 2 ]]> LSTM 0.575 0.423 CNN-LSTM 0.593 0.508 CEEMDAN-LSTM 0.505 0.556 CEEMDAN-CNN 0.466 0.601 CEEMDAN-CNN-LSTM 0.325 0.675 The present invention aims at the operation rules of node electricity prices in the spot market in Gansu, conducts mining based on the data disclosed by integrated trading institutions and meteorological data, and proposes a combined model to realize the prediction of node electricity prices in Gansu, which has achieved good results. Through effective electricity price forecasting, power generation companies can be guided to predict the supply and demand relationship of the market in advance, adjust their own business behaviors, make decisions based on maximizing individual interests, thereby maximizing social welfare and promoting the continuous optimization of the operation efficiency of the power market in the entire market.

[0038] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Any changes, modifications, additions or substitutions made by ordinary technicians in this technical field within the essential scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for predicting day-ahead electricity prices in the spot market based on a hybrid model, characterized in that: The steps include: Step 1: Use the CEEMDAN algorithm to perform modal decomposition on the electricity price data to be predicted to obtain reference data for electricity price prediction; Step 2: Use the CNN prediction model and LSTM prediction model to extract and predict the spatiotemporal characteristics of the electricity price prediction reference data, and output the high-frequency IMF component prediction results; Step 3: Extract and predict the time series characteristics of non-high-frequency IMF components through the ARIMA prediction model, and output the prediction results for non-high-frequency IMF components; Step 4: Read the prediction results of step 2 and step 3 respectively, and integrate and evaluate the prediction results through the hybrid model.

2. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 1 is characterized in that: In step 1, the specific process of modal decomposition of the predicted electricity price data is as follows: (1) Adaptively decompose the historical curve of the electricity price data to be predicted. The decomposition result is a series of signal features with different frequencies, called IMF components, which form the basis for subsequent predictions. (2) Based on the IMF component, the zero-crossing rate is evaluated, the high frequency is judged as high quality, and the high-frequency IMF components are merged and reconstructed into an overall high-frequency component.

3. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 1 is characterized in that: In step 2, the specific process of extracting and predicting the spatiotemporal characteristics of the electricity price prediction reference data using the CNN prediction model and the LSTM prediction model is as follows: (1) Collect publicly disclosed market data and weather data from cities in the corresponding provinces, aggregate and analyze the data, and use them as input data for the prediction algorithm. (2) For high-frequency IMF components, deep neural network prediction models, CNN prediction models and LSTM prediction models are built based on the keras library; (3) Use the CNN model to extract the spatial characteristics of the input data, use the LSTM prediction model to extract the temporal characteristics of the input data, and output the high-frequency IMF component prediction results.

4. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 3 is characterized in that: In step 2, the function expression for constructing the CNN prediction model is: ; in, To extract the features, is the sigmoid activation function, is the weight matrix, is a time series, is the bias vector.

5. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 3 is characterized in that: In step 2, the specific process of building the LSTM prediction model is as follows: （1） （2） （3） （4） （5） （6） in, is the weight matrix, is the corresponding bias vector, is the hyperbolic tangent function, is the dot product, is the output at the previous moment, To retain the degree value, is the memory state of the previous moment, is the added degree value of the current state, is the intermediate state, is the current state, is the output degree value, is the output at the current moment, The input at the current moment.

6. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 1 is characterized in that: In step 3, the specific process of extracting and predicting the time series characteristics of the high-frequency IMF component through the ARIMA prediction model is as follows: (1) Construct ARIMA forecasting model; (2) For non-high-frequency IMF components, the constructed ARIMA forecasting model is used for fitting, and the grid search method is used to determine the autoregressive, differential and moving average hyperparameters of each IMF component; (3) Use Python to search and predict the parameters of the ARIMA prediction model, use a timing trigger mechanism, and use the trained hyperparameters to output prediction results for non-high-frequency IMF components based on the latest saved CEEMDAN decomposition results.

7. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 6 is characterized in that: The function expression for constructing the ARIMA forecasting model is: ; Among them, in the formula: is the predicted target value; is the value of the previous period; is the mean of the time series; , is white noise, that is, the error between the observed value and the predicted value, which is usually assumed to be normally distributed; is the parameter of each white noise, which is used to measure the influence of this white noise on the current time point; , The model order needs to be specified manually; is the autocorrelation coefficient.

8. The method for realizing day-ahead electricity price forecasting in the spot market based on a hybrid model according to claim 1 is characterized in that: The specific process of integrating and evaluating the prediction results through the hybrid model in step 4 is as follows: (1) Use Python to write a data processing program, adopt a timed trigger mechanism, read the prediction results of step 2 and step 3 respectively, and evaluate the accuracy and error rate of the prediction results for the part of the electricity price data that has been obtained. After meeting the standards, merge the results to obtain the overall electricity price prediction results; (2) If the standard is not met, the model with the most recent update date and the prediction results that meet the standard will be used to execute the prediction program again, output the results and evaluate them, and push the information about the unqualified prediction results to the system operation and maintenance personnel for troubleshooting.

9. The method for realizing the day-ahead electricity price forecasting in the spot market based on the hybrid model according to claim 2 is characterized in that: The zero-crossing rate is the number of times the data crosses the X-axis per unit time.

Images