Electric power spot price prediction method, medium and system

Abstract

The invention discloses an electric power spot price prediction method, medium and system, and the method comprises the steps: carrying out the reduction of an original electric power spot market operation scene, and obtaining a reduced electric power spot market operation scene; a power price prediction model formed by sequentially connecting a long short-term memory network LSTM and an improved XGBoost model is constructed, the collected historical power market spot price is screened, and the screened historical power market spot price is used for training the power price prediction model; and inputting the acquired evaluation indexes of the reduced electric power spot market operation scene into the trained electric power price prediction model, and outputting a predicted electric power spot price. According to the invention, scene reduction and electricity price prediction are combined, electric power spot price prediction with both accuracy and high efficiency is realized, and the accuracy and speed of electric power spot market price prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity spot price forecasting, and in particular, to a method, medium and system for electricity spot price forecasting. Background Art

[0002] Electricity spot price forecasting is a key technology in the operation of the electricity market, which involves various methods and systems. The spot market is an important way to discover the price of electricity commodities, and price signals are of great significance for formulating trading strategies. Traditional electricity spot price forecasting methods often rely on historical data and statistical models. However, with the continuous development and changes of the electricity market, these methods may not be able to accurately capture the dynamic characteristics of the market. Therefore, researchers have proposed a method for electricity spot price forecasting based on scenario reduction to improve the accuracy and efficiency of forecasting.

[0003] In traditional electricity spot price forecasting, technologies such as time series analysis, machine learning, and deep learning are usually adopted, such as methods like XGBoost and polynomial regression. However, when facing the problem of long-term multi-step forecasting in the future, the forecasting accuracy may decrease significantly as the forecasting period increases. In addition, due to the large price fluctuations in the electricity market, a single forecasting model is difficult to adapt to the changes in the market. Therefore, a method of using a combined model to improve forecasting accuracy has also been proposed. Although the above methods have solved some problems in electricity spot price forecasting, the data is too redundant, there are too many uncertain factors in the market, and the probability of some scenarios occurring is extremely low. Using all data as input values will reduce the efficiency of the model training and forecasting process, and at the same time increase the difficulty of model selection and optimization. Most of the previous scenario reductions only consider the uncertainty of a certain factor, such as the uncertainty of wind power and new energy photovoltaic output. The reduced scenarios are only based on the uncertainty of a single factor, which has great limitations and the scenarios are still complex. Summary of the Invention

[0004] Embodiments of the present invention provide a method, medium and system for electricity spot price forecasting to solve the problem that the scenarios used in the prior art to forecast electricity spot prices are complex, resulting in reduced forecasting efficiency and increased difficulty in model selection and optimization.

[0005] In a first aspect, a method for electricity spot price forecasting is provided, including: Reducing the original operation scenarios of the electricity spot market to obtain the reduced operation scenarios of the electricity spot market; Constructing an electricity price forecasting model composed of a long short-term memory network (LSTM) and an improved XGBoost model connected in sequence, screening and processing the collected historical electricity market spot prices, and using the screened historical electricity market spot prices to train the electricity price forecasting model; Input the evaluation metrics of the reduced operation scenarios of the electricity spot market collected into the trained electricity price prediction model, and output the predicted electricity spot price.

[0006] In a second aspect, a computer-readable storage medium is provided, on which computer program instructions are stored; when the computer program instructions are executed by a processor, the electricity spot price prediction method as described in the embodiments of the first aspect is implemented.

[0007] In a third aspect, an electricity spot price prediction system is provided, including: the computer-readable storage medium as described in the embodiments of the second aspect.

[0008] In this way, the embodiments of the present invention combine scenario reduction and electricity price prediction to achieve electricity spot price prediction that takes into account both accuracy and efficiency, improving the accuracy and speed of electricity spot market price prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0010] Figure 1 is a flowchart of the electricity spot price prediction method according to the embodiments of the present invention; Figure 2 is a schematic diagram of the evaluation metrics according to the embodiments of the present invention; Figure 3 is a schematic diagram of the long short-term memory network LSTM according to the embodiments of the present invention; Figure 4 is a schematic diagram of the convergence curves of the XGBoost model with and without adding residuals between leaf nodes according to the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0012] The embodiments of the present invention disclose an electricity spot price prediction method. As Figure 1 shown, the method of the embodiments of the present invention includes the following steps: Step S101: Reduce the original operation scenario of the electricity spot market to obtain the reduced operation scenario of the electricity spot market.

[0013] The operation scenario of the electricity spot market consists of multiple parameters, which can be selected according to the actual situation. For example, the parameters include: power load data, weather data, seasonal factors, holiday information, and other data.

[0014] Specifically, this step includes the following process: 1. Use the analytic hierarchy process to determine the first weight of each type of evaluation index.

[0015] The specific steps of the analytic hierarchy process are as follows: (1) Establish a judgment matrix composed of the relative importance between any two types of evaluation indexes.

[0016] The evaluation indexes can be selected according to the actual situation. In a specific embodiment, the evaluation indexes are as Figure 2 shown, including 31 evaluation indexes such as the declared space concentration degree.

[0017] The judgment matrix is as follows: .

[0018] Among them, A is the judgment matrix, a ij is the evaluation index Y i and Y j the relative importance between them, n is the number of types of evaluation indexes.

[0019] This relative importance can be determined by experts. Specifically, a ij taking the value of 1 means that the evaluation index Y i is equally important relative to the evaluation index Y j , a ij taking the value of 2 means that the evaluation index Y i is slightly more important relative to the evaluation index Y j , a ij taking the value of 3 means that the evaluation index Y i is relatively more important relative to the evaluation index Y j , a ij taking the value of 1 means that the evaluation index Y iRelative evaluation index Y j is very important. a ij A value of 1 indicates that the evaluation index Y i is the relative evaluation index Y j is absolutely important. When the value is the reciprocal of the above number, it represents the evaluation index Y j is the relative evaluation index Y i importance.

[0020] (2) Normalize each element in each column of the judgment matrix to obtain the value of each element after normalization.

[0021] The calculation formula for normalization is as follows: .

[0022] Where x ij is the value of the element a ij after normalization.

[0023] (3) Sum up the values of each element in each row after normalization to obtain the sum value of the elements in that row.

[0024] The calculation formula for the sum value of the elements in that row is as follows: .

[0025] Where X i is the sum value of the elements in the i th row.

[0026] (4) Calculate the average value of the sum values of the elements in each row.

[0027] The calculation formula for the average value of the sum values of the elements in each row is as follows: .

[0028] Where is the average value of the sum values of the elements in the i th row.

[0029] (5) Calculate the random consistency ratio of the judgment matrix.

[0030] The calculation formula for the random consistency ratio is as follows: .

[0031] Where R CRis the random consistency ratio of the judgment matrix, R CI is the consistency index of the judgment matrix, R RI is the average random consistency index of the judgment matrix, which can be obtained by looking up the table.

[0032] Among them, . λ max is the maximum eigenvalue of the judgment matrix.

[0033] Among them, . .

[0034] (6) When the random consistency ratio of the judgment matrix is less than the preset ratio threshold, the average value of the sum of each row element is used as the first weight of the evaluation index corresponding to the row of the current category. Otherwise, the relative importance between any two categories of evaluation indexes is readjusted, and the above steps are repeated until the random consistency ratio of the judgment matrix is less than the preset ratio threshold.

[0035] The preset ratio threshold can be set to 0.10. When R CR <0.10, it is considered that the result of the hierarchical single sorting has good consistency, and the calculated weights are reasonable. Otherwise, the values of each element of the judgment matrix need to be readjusted.

[0036] In this way, the set of the first weights of each category of evaluation indexes finally obtained is denoted as: .

[0037] Among them, is the set of the first weights, is the i first weight of the

[0038] 2. The entropy weight method is used to determine the second weight of each category of evaluation indexes.

[0039] The specific steps of the entropy weight method are as follows: (1) Establish a feature matrix composed of each category of evaluation indexes under each power spot market operation scenario.

[0040] Specifically, the feature matrix is as follows: .

[0041] Among them, Z is the feature matrix, z ji is the j th category of evaluation indexes under the i th power spot market operation scenario. The number of power spot market operation scenarios is m, and the number of types of evaluation indexes is n.

[0042] (2) Normalize each element in each column of the feature matrix to obtain the value after normalization for each element.

[0043] The calculation formula for normalization is as follows: .

[0044] Where, p ji is the value of the element z ji after normalization.

[0045] The feature matrix after normalization can be denoted as P = ( p ji ) m×n .

[0046] (3) Calculate the relative entropy value of each type of evaluation index using the value after normalization for each element.

[0047] The calculation formula for the relative entropy value of each type of evaluation index is as follows: .

[0048] Where, E i is the relative entropy value of the i -th type of evaluation index.

[0049] (4) Calculate the entropy weight of each type of evaluation index using the relative entropy value of each type of evaluation index.

[0050] The calculation formula for the entropy weight of the evaluation index is as follows: .

[0051] Where, is the entropy weight of the i -th type of evaluation index.

[0052] Take the entropy weight of each type of evaluation index as the second weight of each type of evaluation index. Thus, the set of the second weights of each type of evaluation index finally obtained is denoted as: .

[0053] Where, is the set of the second weights, is the second weight of the i -th type of evaluation index.

[0054] 3. Calculate the mean of the first weight and the second weight of each type of evaluation index to obtain the weight of each type of evaluation index.

[0055] The calculation formula for the weight of each type of evaluation index is as follows: 。

[0056] Among them, is the weight of the i type of evaluation index.

[0057] 4. Sum the products of the values of each type of evaluation index at each moment in each original electricity spot market operation scenario and the weight of this type of evaluation index, to obtain the set of renewable energy output prediction values at each moment in each original electricity spot market operation scenario.

[0058] The calculation formula for the renewable energy output prediction value at each moment in the electricity spot market operation scenario is as follows: 。

[0059] Among them, is the renewable energy output prediction value at moment v in the t th electricity spot market operation scenario, is the value of the v th type of evaluation index at moment t in the i th electricity spot market operation scenario.

[0060] 5. Adopt the set of renewable energy output prediction values at each moment in all original electricity spot market operation scenarios, and perform scenario reduction through the backward substitution reduction method (SBR) to obtain the reduced electricity spot market operation scenario.

[0061] Specifically, the set of renewable energy output prediction values at each moment in the electricity spot market operation scenario is denoted as: , among which, is the renewable energy output prediction value at moment v in the t th electricity spot market operation scenario. The total number of electricity spot market operation scenarios is V. The sum of the occurrence probabilities of V electricity spot market operation scenarios is 1.

[0062] The set of deleted scenarios is denoted as M, and the initialized set of deleted scenarios is an empty set, that is, no scenario is deleted. The set of retained scenarios is denoted as S, and the initialized set of retained scenarios includes all scenarios, denoted as , among which, s i is the i th electricity spot market operation scenario.

[0063] The distance between any two electricity spot market operation scenarios can be calculated by the following formula: 。

[0064] Among them, is the distance between the i th power spot market operation scenario and the j th power spot market operation scenario.

[0065] The principle of scenario reduction established by SBR is that the probability distance between the scenario set before reduction and the scenario set after reduction is the smallest, which can be expressed by the following formula: .

[0066] Among them, p i is the i th power spot market operation scenario s i occurring probability.

[0067] The number of iterations is represented by k , initialized k = 0. In the process of each iteration, when the above formula is satisfied, the i th power spot market operation scenario s i will be deleted.

[0068] Suppose the scenario to be deleted in the k th iteration is γ k . Move γ k into the set M. Then, the probability of the scenario γ k nearest to the deleted scenario s j is: .

[0069] Repeat the iteration until the set M of deleted scenarios contains a specified number of scenarios.

[0070] Through the above steps, compare the probability distance between the generated scenarios and the deterministic scheme, select the scenario with the highest similarity to the deterministic scheme, and retain it to obtain the reduced power spot market operation scenarios. By simplifying the complex prediction problem into several key scenarios, the effective reduction of scenarios is realized, unnecessary scenario numbers are reduced, and at the same time, key market dynamic information is retained. Each scenario represents a possible market state or trend, so as to reduce the complexity of the model and improve the prediction efficiency and accuracy.

[0071] Step S102: construct an electricity price prediction model composed of a long short-term memory network LSTM and an improved XGBoost model connected in sequence, and screen the collected historical electricity market spot prices, and use the screened historical electricity market spot prices to train the electricity price prediction model.

[0072] Specifically, the long short-term memory network LSTM is an existing network, such as Figure 3 As shown in the figure, independent storage units are configured based on RNN. Through the configuration of LSTM, the gradient diffusion problem of RNN is solved and long-term memory management is realized. LSTM configures three control units based on RNN, namely the forget gate, input gate, and output gate. The mathematical logic controlled by the gate unit can effectively avoid the shortcomings of RNN network operation, thereby promoting the accuracy and analysis speed of electricity price prediction.

[0073] Compared with the XGBoost model in the prior art, the improved XGBoost model adds residual connections between leaf nodes. In addition, the improved XGBoost model in the embodiment of the present invention is the same as the existing XGBoost model. The performance of the model is improved by introducing residual connections. The residual connection helps to alleviate the gradient vanishing problem in the deep model, so that the information in the deep tree structure can be more effectively propagated to the front layer, thereby improving the training efficiency and performance of the model. Figure 4 As shown in the convergence curve of , the model with residual connection converges better than the model without residual connection.

[0074] There is a certain volatility in the spot price of the electricity market. For example, abnormal weather, some rare events including natural disasters, power grid failures, major accidents, and data errors lead to abnormal fluctuations in the spot price of the electricity market. There is a certain randomness in this part of the data, which affects the model training effect and should be eliminated first.

[0075] The specific steps include: 1. Determine the appropriate number of cluster centers using the Elbow method. Draw Elbow curves for different numbers of cluster centers and select the point where the curve begins to flatten as the number of cluster centers.

[0076] 2. Use the K-means clustering algorithm to cluster the collected historical electricity price data. The goal of K-means is to minimize the sum of squared errors within the cluster, that is, the sum of the squares of the distances from each point to the center of the cluster to which it belongs. The objective function of clustering is as follows: .

[0077] in, K is the number of clusters, C i It is i The point set of a cluster,x belongs to C i data points μ i is the i centroid of the cluster, representing the x Euclidean distance squared between the data point μ i and the cluster center.

[0078] Each data point will be assigned to the vicinity of the nearest clustering center. Next, measure the clustering of each point to the clustering center, and use the Euclidean distance method to measure the distance between each point and the cluster center. , n where

[0079] is the dimension of the data. The data points at each clustering center are sorted in descending order according to the values calculated by the Euclidean distance method.

[0080] In the case of a standard normal distribution, data beyond 3 standard deviations is generally considered an outlier, and data within 3 standard deviations contains more than 99% of the data in the dataset. The remaining 1% of the data can be regarded as outliers. Therefore, referring to common outlier handling methods, set a proportion of outliers at 1%, that is, select the top 1% of each clustering center's ranking as outliers, and finally remove the outliers.

[0081] The electricity price prediction model is pre-trained. During the process of training the electricity price prediction model, the training samples input to the long short-term memory network LSTM are the evaluation indicators of the reduced electricity spot market operation scenario, and the training samples input to the improved XGBoost model are the predicted electricity spot prices output by the long short-term memory network LSTM and the parameters of the reduced electricity spot market operation scenario. The training process includes: 1. Input the evaluation indicators of the reduced electricity spot market operation scenario into the LSTM to output the predicted electricity spot price, and input the predicted electricity spot price output by the long short-term memory network LSTM and the parameters of the reduced electricity spot market operation scenario into the improved XGBoost model. Define the loss function used for training as follows: .

[0082] .

[0083] where y jis the output value of the prediction model, y i is the true value of the sample, f k is the k th tree model, is the total loss of each sample, and the loss function of XGBoost can be set artificially. is the regularization term, T is the number of leaf nodes, γ is the regularization term of the leaf nodes of the tree, w is the k th leaf node weight value of the tree. λ ensures that the scores of the leaf nodes do not exceed a certain range. Both parts of the penalty term can prevent overfitting.

[0084] 2. Iteratively solve, and the final predicted value is the sum of the outputs of multiple trees as follows: .

[0085] Among them, is the objective function of the tree, is the predicted output value of the previous t -1 trees, is the t th tree output value, is the regularization term.

[0086] 3. Use the second derivative Taylor rule to expand the loss function , iteratively learn to the t th tree, , among which, is the gradient (first derivative) of the i th sample, which is the first derivative of the loss function with respect to the predicted value , that is , is the Hessian (second derivative) of the i th sample, which is the second derivative of the loss function with respect to the predicted value , that is .

[0087] 4. Ignore the constant term to simplify the objective function. The first part of the function is the loss function of the previous round of the tree, which is a known constant in this round. This value does not affect the tree construction process in this round and can be ignored, thus simplifying the objective function to: .

[0088] 5. Add residual connections between leaf nodes. First, assume that the tree has T leaf nodes, and the sample set in the J-th leaf node is denoted as . Based on this, , which is the sum of the first-order derivatives corresponding to all samples falling into leaf node J, is the weight of the j-th leaf node. Therefore, can be converted to . After converting and combining the regularization terms, the output can be simplified to: .

[0089] 6. Split the nodes of the tree according to the formula and judge the iterative process according to the set conditions. When the value by which the loss function decreases is greater than γ , the tree can be split. The γ parameter added to the denominator when calculating the gain can make it smooth. At the same time γ can help prevent overfitting and is optional.

[0090] 7. Stop growing and generate a decision tree. The sum of the output values is the predicted electricity spot price.

[0091] For the above training process, the input and output descriptions of each network are shown in Table 1.

[0092] Table 1 Input and Output Data Examples

[0093] Step S103: Input the evaluation index of the reduced electricity spot market operation scenario collected into the trained electricity price prediction model, and output the predicted electricity spot price.

[0094] It should be understood that, different from the training process, in this actual prediction step, when outputting the predicted electricity spot price, the predicted electricity spot price output by the long short-term memory network LSTM is input into the improved XGBoost model, that is, in actual prediction, the parameters of the reduced electricity spot market operation scenario are no longer input into the improved XGBoost model.

[0095] It should also be understood that whether it is training or actual prediction, the evaluation index input into the electricity price prediction model is the normalized evaluation index. Similarly, during the training process, the parameters of the reduced electricity spot market operation scenario input are also normalized parameters.

[0096] The normalization process can adopt existing conventional processing methods. For example, adopt the following method: .

[0097] Among them, is the normalized value; s is the original value; , respectively represent the minimum and maximum values of different types of numerical values.

[0098] By predicting the spot price of electricity, enterprises, power plants or power supply companies can be guided to adjust the power generation volume, and then adjust the power used for power generation, the use of solar panels, the coal consumption of coal-fired power, etc., so as to reduce the loss of materials and slow down the aging of devices.

[0099] The following takes a specific embodiment to further illustrate the technical solution of the embodiment of the present invention.

[0100] Based on the data published by a certain power trading market, first obtain the weights of each evaluation index, as shown in Table 2. Represent the states of each possible scenario with indicators, and use the backward substitution reduction method (SBR) for scenario reduction. After multiple iterations and sorting according to the probability of scenario occurrence, the scenarios are reduced to 5, greatly reducing the computational complexity. Input the evaluation indicators of the 5 scenarios into the electricity price prediction model to obtain the prediction results. Table 3 shows the comparison between the price prediction results and the actual electricity price on a certain day.

[0101] Table 2 Weight Results

[0102] Table 3 Price Comparison Results

[0103] As can be seen from Table 3, among the 24 electricity price prediction results of the technical solution of the embodiment of the present invention, 23 are all kept below 3%, and the prediction compliance rate is 96.8%, with relatively high prediction accuracy.

[0104] In addition, in order to verify the advantages and disadvantages of the technical solution of the embodiment of the present invention, 2 currently widely used day-ahead clearing electricity price prediction models are selected for comparison. The commonly used indicators for verifying the accuracy of electricity price prediction are the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE), and their calculation processes are as follows: .

[0105] .

[0106] Among them, is the actual value, is the predicted value, and T is the total number of the test set.

[0107] Method 1 is based on a single XGBoost model for prediction, and Method 2 is based on a GRU-XGBoost combined model for prediction.

[0108] The prediction results of the three methods are shown in Table 4.

[0109] Table 4 Prediction Results of the Three Methods

[0110] As can be seen from Table 4, the average MAPE of the LSTM and XGBoost combined prediction model of the embodiments of the present invention in 30 days is 3.49, which is lower than 5.96 and 6.24 of the single prediction models XGBoost and GRU-XGBoost combined model; at the same time, the average MAE of the LSTM and XGBoost combined prediction model of the embodiments of the present invention is 0.56, which is significantly lower than 1.13 and 0.92 of the single prediction models XGBoost and GRU-XGBoost combined model; it proves that the combined model of the embodiments of the present invention is significantly better than the other two methods.

[0111] In addition, the embodiments of the present invention also provide a computer-readable storage medium, on which computer program instructions are stored; when the computer program instructions are executed by a processor, the power spot price prediction method as described in the above embodiments is implemented.

[0112] In addition, the embodiments of the present invention also provide a power spot price prediction system, including: the computer-readable storage medium as described in the above embodiments. The system integrates multiple modules such as data processing, scenario generation, reduction optimization, and price prediction, and realizes the full-process automation of power spot price prediction.

[0113] In summary, the embodiments of the present invention combine scenario reduction and electricity price prediction to achieve power spot price prediction that takes into account both accuracy and efficiency, improve the accuracy and speed of power spot market price prediction, and provide strong support for power trading decisions.

[0114] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for predicting spot electricity prices, characterized in that: include: The original electricity spot market operation scenario is reduced to obtain a reduced electricity spot market operation scenario; Constructing an electricity price prediction model composed of a long short-term memory network LSTM and an improved XGBoost model connected in sequence, and screening the collected historical electricity market spot prices, and using the screened historical electricity market spot prices to train the electricity price prediction model; The collected evaluation indexes of the reduced electricity spot market operation scenario are input into the trained electricity price prediction model to output the predicted electricity spot price.

2. The method for predicting electricity spot prices according to claim 1, characterized in that: The step of obtaining the reduced electricity spot market operation scenario includes: The first weight of each evaluation index is determined by using the analytic hierarchy process; The entropy weight method is used to determine the second weight of each type of evaluation index; Calculate the average of the first weight and the second weight of each type of evaluation index to obtain the weight of each type of evaluation index; The product of the value of each type of evaluation index at each moment in each original electricity spot market operation scenario and the weight of the evaluation index of this type is added together to obtain a set of renewable energy output forecast values ​​at each moment in each original electricity spot market operation scenario; The set of renewable energy output forecast values ​​at each moment under all original electricity spot market operation scenarios is adopted, and the scenario is reduced through backward substitutive reduction method to obtain the reduced electricity spot market operation scenario.

3. The method for predicting electricity spot prices according to claim 1, characterized in that: The improved XGBoost model adds residual connections between leaf nodes.

4. The method for predicting electricity spot prices according to claim 1, characterized in that: In the process of training the electricity price prediction model, the training samples input into the long short-term memory network LSTM are the evaluation indicators of the reduced electricity spot market operation scenario, and the training samples input into the improved XGBoost model are the predicted electricity spot prices output by the long short-term memory network LSTM and the parameters of the reduced electricity spot market operation scenario.

5. The method for predicting electricity spot prices according to claim 1, characterized in that: In the process of outputting the predicted electricity spot price, the predicted electricity spot price output by the long short-term memory network LSTM is input into the improved XGBoost model.

6. The method for predicting electricity spot prices according to claim 1, characterized in that: The evaluation index input into the electricity price prediction model is the evaluation index after normalization processing.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by the processor, the method for predicting electricity spot prices according to any one of claims 1 to 6 is implemented.

8. A power spot price forecasting system, characterized in that: include: The computer readable storage medium of claim 7.

Images