Medium-and-long-term electricity price prediction method and device suitable for electricity market and computer equipment

Abstract

The invention relates to the field of electricity price prediction, and discloses a medium-and-long-term electricity price prediction method and device suitable for an electricity market and computer equipment, and the method comprises the steps: obtaining operation-related time sequence data of the electricity market; obtaining first time sequence data on the first type of feature dimensions from the operation related time sequence data according to the first type of feature dimensions; wherein the first type of feature dimensions are feature dimensions related to the bidding space; according to the first time sequence data, performing time sequence prediction on the feature representation on the first feature dimension to obtain bidding related time sequence data; and performing medium-and-long-term electricity price prediction based on the bidding-related time sequence data and the operation-related time sequence data to obtain a predicted electricity price. Therefore, by performing staged detailed prediction on the bidding space related features and other operation related features, various factors influencing the electricity price can be captured more comprehensively, accumulation and amplification of prediction errors are reduced, and the final electricity price prediction result is more accurate.

Description

Technical Field

[0001] The present application relates to the field of electricity price forecasting, and in particular, to a medium- and long-term electricity price forecasting method, device, and computer device suitable for the electricity market. Background Art

[0002] With the integration and development of the electricity market, the electricity volume of market-based transactions is continuously increasing, and the development of the electricity spot market is also accelerating. Spot transactions help to reveal electricity prices, while medium- and long-term transactions can provide price protection for market participants. Therefore, it is crucial to predict the floating electricity prices in the market for a period of time in the future.

[0003] However, the methods for predicting the floating electricity prices in the market in the related art still have limitations, and a method that can more accurately predict electricity prices, especially the electricity volume prices in medium- and long-term transactions, is needed. Summary of the Invention

[0004] The present application aims to at least partly solve one of the technical problems in the related art. For this purpose, the present application provides a medium- and long-term electricity price forecasting method, device, and computer device suitable for the electricity market. The main technical solutions adopted in the present application include:

[0005] In a first aspect, an embodiment of the present application provides a medium- and long-term electricity price forecasting method suitable for the electricity market. The method includes: obtaining time-series data related to the operation of the electricity market; obtaining first time-series data on a first type of feature dimension from the time-series data related to the operation according to the first type of feature dimension, where the first type of feature dimension is a feature dimension related to the bidding space; performing time-series forecasting on the feature performance on the first type of feature dimension according to the first time-series data to obtain bidding-related time-series data; and performing medium- and long-term electricity price forecasting based on the bidding-related time-series data and the time-series data related to the operation to obtain a predicted electricity price.

[0006] Optionally, second time-series data on a second type of feature dimension is obtained from the time-series data related to the operation according to the second type of feature dimension, where the second type of feature dimension is a feature dimension unrelated to the bidding space. Correspondingly, performing medium- and long-term electricity price forecasting based on the bidding-related time-series data and the time-series data related to the operation to obtain a predicted electricity price includes: performing medium- and long-term electricity price forecasting based on the bidding-related time-series data, the first time-series data, and the second time-series data to obtain a predicted electricity price.

[0007] Optionally, performing time-series forecasting on the feature performance on the first type of feature dimension according to the first time-series data to obtain bidding-related time-series data includes: inputting the first time-series data into an LSTM model; and performing time-series forecasting on the feature performance on the first type of feature dimension through the LSTM model to obtain bidding-related time-series data.

[0008] Optionally, the second type of feature dimension includes any one of economic development, coal price, liquefied natural gas price, and climate change situation.

[0009] Optionally, perform medium- and long-term electricity price prediction based on bidding-related time series data and operation-related time series data to obtain a predicted electricity price, including: performing feature dynamic enhancement processing on the bidding-related time series data and operation-related time series data to obtain a predicted feature vector corresponding to the bidding-related time series data and a historical feature vector corresponding to the operation-related time series data; using the predicted feature vector, historical feature vector, bidding-related time series data, and operation-related time series data to perform medium- and long-term electricity price prediction to obtain a predicted electricity price.

[0010] Optionally, the operation-related time series data further includes second time series data on the second type of feature dimension; wherein the second type of feature dimension is a feature dimension unrelated to the bidding space; performing feature dynamic enhancement processing on the bidding-related time series data and operation-related time series data to obtain a predicted feature vector corresponding to the bidding-related time series data and a historical feature vector corresponding to the operation-related time series data, including: performing feature dynamic enhancement processing on the bidding-related time series data, first time series data, and second time series data to obtain a first feature vector corresponding to the first time series data, a second feature vector corresponding to the second time series data, and a predicted feature vector; correspondingly, using the predicted feature vector, historical feature vector, bidding-related time series data, and operation-related time series data to perform medium- and long-term electricity price prediction to obtain a predicted electricity price, including: performing medium- and long-term electricity price prediction based on the first feature vector, second feature vector, predicted feature vector, first time series data, second time series data, and bidding-related time series data to obtain a predicted electricity price.

[0011] Optionally, the first type of feature dimension includes any one of power load, wind power output, photovoltaic power output, and tie line.

[0012] In a second aspect, an embodiment of the present application provides a medium- and long-term electricity price prediction device suitable for the electricity market. The device includes: a time series data acquisition module for acquiring operation-related time series data of the electricity market; a dimension data acquisition module for obtaining first time series data on the first type of feature dimension from the operation-related time series data according to the first type of feature dimension; wherein the first type of feature dimension is a feature dimension related to the bidding space; a time series prediction module for performing time series prediction on the feature performance on the first type of feature dimension according to the first time series data to obtain bidding-related time series data; and an electricity price prediction module for performing medium- and long-term electricity price prediction based on the bidding-related time series data and operation-related time series data to obtain a predicted electricity price.

[0013] In a third aspect, the present application further provides a computer device including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method in any one of the above are implemented.

[0014] Fourthly, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method according to any one of the above are implemented.

[0015] Fifthly, the present invention provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the method according to any one of the above are implemented.

[0016] In the above embodiments, firstly, the time-series data related to the operation in the power market is dimensionally divided, and the time-series prediction is performed on the feature data of the feature dimensions related to the bidding space. Then, based on the bidding-related time-series data obtained by performing the time-series prediction on the feature performance on the first type of feature dimension and other operation-related time-series data in the power market, the secondary medium- and long-term electricity price prediction is performed, and finally the predicted electricity price is obtained. Thus, through the phased refinement prediction method of first accurately predicting the key feature related to the bidding space and then combining these prediction results with other operation-related features, not only can various factors affecting the electricity price be captured more comprehensively, but also the accumulation and amplification of prediction errors are reduced, so that the final electricity price prediction result is more accurate and better meets the demand for medium- and long-term electricity price prediction in the power market. Description of the Drawings

[0017] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of a medium- and long-term electricity price prediction method according to an embodiment of the present application;

[0019] Figure 2 It is a flowchart of a time-series prediction method according to an embodiment of the present application;

[0020] Figure 3 It is a flowchart of an electricity price prediction method according to another embodiment of the present application;

[0021] Figure 4 It is a flowchart of an electricity price prediction method according to still another embodiment of the present application;

[0022] Figure 5 It is a structural block diagram of a medium- and long-term electricity price prediction device according to an embodiment of the present application;

[0023] Figure 6Internal structure diagram of a computer device according to an embodiment of the present application. Detailed implementation manners

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0025] With the integration and development of the power market, the electricity volume of market-based transactions has been increasing continuously, and the development of the power spot market is also accelerating. Spot transactions help to reveal power prices, while medium- and long-term transactions can provide price protection for market participants. In this way, the power market trading system can be more perfect. Medium- and long-term transactions have many advantages: it can stabilize market prices to a certain extent and enhance the predictability of power sellers for future market changes. At the same time, power buyers can ensure a certain amount of power supply in advance through medium- and long-term contracts, thereby effectively controlling costs and reducing operating risks. In addition, multiple trading windows for medium- and long-term transactions also provide arbitrage opportunities for market participants. Therefore, it is crucial to predict medium- and long-term electricity prices, because it can help power market participants make more reasonable investment and operation decisions, reduce market risks, and optimize resource allocation. Accurate electricity price prediction enables power generation companies to accurately grasp the market trend and construct the optimal electricity volume and electricity price bidding strategies to obtain the maximum profit; power users can formulate reasonable electricity consumption plans based on electricity price prediction to reduce costs. At the same time, electricity price prediction provides a scientific basis for market regulators to promote the healthy, stable, and orderly competitive development of the market.

[0026] However, compared with spot price prediction, medium- and long-term electricity price prediction faces more challenges because of greater price fluctuations and more complex influencing factors, including changes in fuel costs, increases in new energy installed capacity, and the impact of economic development on industrial electricity demand.

[0027] Based on this, according to an embodiment of the present application, an embodiment of a medium- and long-term electricity price prediction method suitable for the power market is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from that here.

[0028] In this embodiment, a medium- and long-term electricity price prediction method suitable for the power market is provided. Figure 1It is a flowchart of a medium- and long-term electricity price prediction method suitable for the electricity market according to an embodiment of the present application. As Figure 1 shown, the process includes the following steps:

[0029] S110. Obtain the time-series data related to the operation of the electricity market.

[0030] Among them, the time-series data related to the operation can refer to various time-series data directly related to the operation and management of the power system in the electricity market. The time-series data related to the operation can be used to reflect the operation status and change trend of the electricity market at different time points. Exemplarily, the time-series data related to the operation can include power load, wind power output, photovoltaic power output, tie lines, meteorological-related features, installed capacity, economic development, coal price, liquefied natural gas price, climate change, relevant policy updates, etc. Specifically, the time-series data related to the operation of the electricity market can be obtained from power companies and grid companies through an automated data acquisition system in real time for power load, power generation, market transactions, etc.

[0031] S120. Obtain the first time-series data on the first type of feature dimension from the time-series data related to the operation according to the first type of feature dimension.

[0032] Among them, the first type of feature dimension is a feature dimension related to the bidding space. Specifically, the first type of feature dimension can refer to a dimension mainly involving factors affecting the electricity market bidding process and electricity price formation. Exemplarily, the first type of feature dimension can include the power load reflecting the total electricity demand of all users in the power system during a specific period; it can also include the output of various types of generator sets such as wind power output and photovoltaic power output that affect the electricity market supply capacity; it can also include the tie line transmission data connecting different power systems or regions and reflecting the flow of electricity between different regions.

[0033] Furthermore, the first time-series data can refer to the time-series data corresponding to the first type of feature dimension extracted from the time-series data related to the operation. Specifically, the first time-series data can reflect the change trend and law of the first type of feature over time. Exemplarily, taking the power load as an example, the first time-series data of the power load can show the load change situation in different time periods of a week, such as the load gradually rising during the week, reaching a peak on weekends, and starting to drop to a low point again next week.

[0034] Specifically, the method of focusing on key features can be used to determine the feature dimension directly related to the bidding space from a large amount of time-series data related to the operation, so as to further extract and screen out the first time-series data on the first type of feature dimension.

[0035] S130. Perform time-series prediction on the feature performance on the first type of feature dimension according to the first time-series data to obtain the time-series data related to bidding.

[0036] Among them, time series prediction is a data processing method that predicts data values at a future time point or within a time period based on historical time series data. Specifically, time series prediction can predict the characteristic performance of specific numerical values or states of a specific feature in the first type of feature dimension at different time points. Exemplarily, taking the power load as an example again, if historical power load data is obtained, then by combining historical meteorological-related features and historical data such as installed capacity, analyzing the autocorrelation, trend, seasonality, etc. of the time series data based on the power load data of last week, and performing time series prediction, the future power load time series data on the feature dimension of power load at present or in the future can be obtained. This power load time series data is the time series data related to bidding. That is, the time series data related to bidding can be time series data used to reflect the predicted performance of features related to the bidding space within a future period of time.

[0037] Optionally, when performing time series prediction, a time series prediction model can be pre-trained. This time series prediction model can be an ARIMA model, a deep learning model, a long short-term memory network (LSTM), etc. And data processing, model training, and prediction can be performed using data analysis and machine learning tools such as Python's Pandas, NumPy, Scikit-learn, TensorFlow, or Keras, etc. Further, during the training process, a part of the historical data can be used in advance as a validation set to evaluate the prediction performance of the constructed time series prediction model. By comparing the difference between the model prediction value and the actual value, calculating error metrics (such as mean square error, mean absolute error, etc.), and repeatedly training and validating, to ensure that the model has good prediction ability and generalization performance.

[0038] S140. Perform medium- and long-term electricity price prediction based on the time series data related to bidding and the time series data related to operation to obtain the predicted electricity price.

[0039] Optionally, before performing medium- and long-term electricity price prediction, the time series data related to bidding and the time series data related to operation can be aligned and synchronously integrated to form a comprehensive data set, so as to ensure the temporal consistency of different data sources. Specifically, during the data fusion process, features helpful for electricity price prediction can be extracted and constructed. Exemplarily, interaction features or derivative features between different features can be calculated, such as the interaction feature between power load and temperature (used to analyze the impact of air-conditioning electricity consumption on the load), or the moving average of fuel price, etc.

[0040] Further, after preprocessing all the bidding-related time series data and operation-related time series data, a pre-trained prediction model can be used to comprehensively consider the impacts of various factors and output the predicted value of the future electricity price, that is, to conduct medium- and long-term electricity price prediction based on the bidding-related time series data and operation-related time series data, and obtain the predicted electricity price.

[0041] It can be understood that the pre-trained prediction model can be a machine learning model (such as random forest, gradient boosting tree, or support vector machine, etc.) and a deep learning model (such as fully connected neural network, convolutional neural network, or recurrent neural network, etc.). Exemplarily, considering the interpretability, prediction ability, and computing resources of the model, the prediction model can include a gated residual network, Self-Attention (self-attention mechanism), and a fully connected layer. During training, by continuously learning the relationships between different features and the electricity price, the prediction model can capture the interactions and impacts between features, and continuously adjust the model parameters and structure based on the training results, so that the prediction model can accurately predict the future medium- and long-term electricity price.

[0042] Among them, the gating structure can control the degree of non-linear contribution in the model, and its formula is as follows:

[0043]

[0044] In the formula, GLU represents the output of the gated linear unit; σ represents the non-linear activation function, W and V represent different weight matrices obtained through multiple trainings for linear transformation of the input features; b and c represent different bias terms obtained through multiple trainings, also used for linear transformation.

[0045] The residual network solves the problem of gradient disappearance in the deep network through residual connections. Instead of directly mapping the input to the output, it adds the input across some network layers to the output. Specifically, the input features can be linearly transformed first. To alleviate the problem of gradient disappearance, the ELU activation function, a non-linear activation function that remains linear in the positive domain and shows exponential growth in the negative domain, is applied. Then, the GLU activation function in the gating structure is applied to the data processed by the ELU activation function. Most importantly, the input features of the original input are added to the features processed by the activation function to form a residual connection. Finally, layer normalization is performed on the result of the residual connection, and the output features processed by the residual network layer are output. Its formula is as follows:

[0046] Y = LayerNorm(a + GLU((ELU(W'a + b'))))

[0047] Wherein, Y represents the output of the residual network layer; a represents the input feature vector of the residual network layer; W′ represents the weight matrix for linearly transforming the input features; b′ represents the bias term, also for linear transformation; ELU represents the non-linear activation function; GLU represents the activation function of the gating mechanism; LayerNorm represents layer normalization.

[0048] Furthermore, after passing through the gated residual network, it can be further input into the Self-Attention mechanism, which can learn global dependencies and filter out the importance of different features. Specifically, in the calculation process, for each token (i.e., each element in the input sequence), three vectors are first generated: query (query vector Q), key (key vector K), and value (value vector V), all of which are obtained from the product of the input matrix X and the matrix (these vectors are obtained by multiplying the input matrix X with different weight matrices). It can be understood that they are essentially all linear transformations of the input matrix X, but contain different data information. Among them, the query vector is analogous to asking, the key vector is analogous to indexing, and the value vector is analogous to answering. The Attention calculation formula in the form of key-value pairs is as follows:

[0049]

[0050] Wherein, Attention(Q, K, V) represents the output of the self-attention mechanism; Q represents the query vector for retrieving relevant information from all key vectors; K represents the key vector for comparing with the query vector to determine the relevance of each value vector; V represents the value vector, which contains the actual information; softmax represents the probability distribution function; d represents the dimension of the key vector, represents the scaling factor, which is used to prevent the dot product result from being too large, thereby avoiding the vanishing gradient when applying the softmax function.

[0051] Specifically, during the calculation of Self-Attention, the query vector of each token will perform a dot product operation with the key vectors of all tokens, obtaining multiple attention scores representing the contribution degree of each token to the current token. Then these scores are normalized through a scaling factor and the softmax function to obtain the attention weights of different features. Further, these attention weights are also used to weight the corresponding value vectors, ultimately obtaining the comprehensive representation of the current token. It can be understood that through the Self-Attention mechanism, the prediction model can learn the mutual relationships and importance between different time steps or features. For example, the prediction model may find that during the winter peak period, certain features (such as heating demand) have a greater impact on electricity prices, so it will assign higher attention weights to these features. In this way, the prediction model can dynamically adjust the degree of attention to different time steps and features to more accurately predict future electricity price changes.

[0052] Optionally, during the training process, a part of the first-time series data can also be used as a validation set to evaluate the prediction performance of the prediction model. By repeatedly comparing the differences between the model prediction values and the actual values, error metrics (such as mean square error or mean absolute error, etc.) are calculated, and thus the model is continuously optimized according to the evaluation results. At the same time, methods such as cross-validation or hyperparameter tuning can also be adopted to improve the generalization ability and prediction accuracy of the model.

[0053] In the above embodiment, first, the time-series data related to the operation in the electricity market is dimensionally divided, and the time-series prediction is performed on the feature data of the feature dimensions related to the bidding space. Then, based on the bidding-related time-series data obtained from the time-series prediction of the feature performance on the first type of feature dimension and other operation-related time-series data in the electricity market, the secondary medium- and long-term electricity price prediction is carried out, and finally the predicted electricity price is obtained. Thus, through the phased refinement prediction method of first accurately predicting the key bidding space-related features and then combining these prediction results with other operation-related features, not only can various factors affecting electricity prices be more comprehensively captured, but also the accumulation and amplification of prediction errors are reduced, so that the final electricity price prediction result is more accurate and better meets the requirements of medium- and long-term electricity price prediction in the electricity market.

[0054] In some embodiments, according to the second type of feature dimension, the second time-series data on the second type of feature dimension is obtained from the operation-related time-series data.

[0055] Among them, the second type of feature dimension is a feature dimension that has nothing to do with the bidding space. Specifically, the second type of feature dimension is opposite to the first type of feature dimension, and can refer to dimensions that do not directly participate in the bidding process but have an indirect impact on electricity prices. Exemplarily, it can be economic development, fuel prices (such as coal prices, liquefied natural gas prices), climate change, or relevant policy updates, etc. Correspondingly, the second time-series data can refer to the time-series data corresponding to the second type of feature dimension extracted from the operation-related time-series data. Exemplarily, the second time-series data of economic development can be data such as GDP growth rate or industrial output value that changes over time; the second time-series data of fuel prices can be historical data of coal prices, liquefied natural gas prices, etc.

[0056] Furthermore, similarly, by focusing on key features, the feature dimensions that have no direct relation to the bidding space can be determined from a large amount of operation-related time-series data, so as to extract and screen the second time-series data on the second type of feature dimension.

[0057] Correspondingly, for medium- and long-term electricity price forecasting based on bidding-related time-series data and operation-related time-series data to obtain the forecast electricity price, it includes:

[0058] Based on the bidding-related time-series data, the first time-series data, and the second time-series data, conduct medium- and long-term electricity price forecasting to obtain the forecast electricity price.

[0059] Similarly, before conducting medium- and long-term electricity price forecasting, the bidding-related time-series data, the first time-series data, and the second time-series data can be first subjected to alignment and synchronization integration operations to form a comprehensive data set, and a forecasting model is used to comprehensively consider the influence of various factors and output the forecast value of the future electricity price, that is, the forecast electricity price is obtained.

[0060] In the above embodiment, by dividing the entire feature dimension of the operation-related time-series data into the second type of feature dimension and extracting the corresponding second time-series data from it, factors that have nothing to do with the bidding space but have an indirect impact on electricity prices can be captured more comprehensively, thereby providing richer and more accurate input data for medium- and long-term electricity price forecasting, helping to improve the ability of the forecasting model to capture the complex dynamics of the market, and enhancing the reliability and accuracy of the forecasting results.

[0061] In some embodiments, according to the first time-series data, conduct time-series forecasting on the feature performance on the first type of feature dimension to obtain the bidding-related time-series data. Please refer to Appendix Figure 2 , including:

[0062] S210. Input the first time-series data into the LSTM model.

[0063] Specifically, a part of the first time-series data can be pre-divided as the training set, and according to the specific task requirements, the parameters of the LSTM model are initialized, including the number of network layers, the number of neurons in each layer, and the activation function, etc. At the same time, the training parameters of the model are set, such as the learning rate and the optimizer, etc., to train an LSTM model that can achieve time-series prediction.

[0064] It should be understood that LSTM (Long Short-Term Memory), that is, the long short-term memory network, is a special recurrent neural network structure that can be used to process and predict long-term dependencies in time-series data.

[0065] Specifically, there are three gate mechanisms in the network structure of the LSTM (long short-term memory network) model, namely the forget gate, the input gate, and the output gate, which respectively determine which output information in the previous time step needs to be discarded, which useful input information at the current moment needs to be retained, and which needs to be output by combining the information.

[0066] Among them, the forget gate can be used to determine which information to retain from the hidden state of the previous time step. Exemplarily, it can be analogized to determining which first time-series data (such as historical power load information) is important and needs to be retained, and which information is outdated and can be discarded. Its output calculation formula is:

[0067] f t =σ(W f ·[h t-1 ,x t +b f )

[0068] In the formula, f t represents the output of the forget gate at the t-th time step; σ represents the non-linear activation function, W f represents the different weight matrices obtained based on multiple trainings; b f represents the bias term obtained based on multiple trainings; [h t-1 , xt represents the hidden state of the previous time step and the input feature (first time-series data) of the current time step.

[0069] The input gate consists of two parts. Among them, one part determines which new information needs to be stored in the memory cell, and the other part updates the memory cell state. Exemplarily, it can be analogized to determining which new information at the current time step (such as the latest power demand or supply data included in the first time-series data) is important and needs to be incorporated into the memory of the model. Its output calculation formula is:

[0070] i t =σ(Wi ·[h t-1 , x t + b i )

[0071]

[0072] where i t represents the output of the input gate; is the candidate memory cell state; W i represents different weight matrices obtained through multiple trainings; b i represents the bias term obtained through multiple trainings.

[0073] It can be understood that i t can be regarded as a decision factor that determines how much of the new information at the current time step (such as the current power demand, supply situation, or weather conditions, etc.) should be incorporated into the model's memory. This decision is calculated based on the current input features and the previous hidden state (i.e., the power load prediction results at the previous time step). then represents the new information that may be updated to the model's memory at the current time step, which may include the current power market supply and demand situation, changes in new energy generation capacity, or fluctuations in fuel costs, etc.

[0074] Furthermore, combining the concepts of the input gate and the forget gate, the update of the memory cell of the LSTM at each time step can be expressed as the following formula:

[0075]

[0076] where represents the memory cell state at the current time step; f t represents the output of the input gate; represents the memory cell state at the previous time step; i t represents the output of the input gate; represents the candidate memory cell state.

[0077] The network structure also includes an output gate, which is used to determine how the information in the memory cell contributes to the hidden state at the next time step. Exemplarily, it can be analogized to determining how to output the information in the memory cell (i.e., the combination of historical power load information and current power load information) to the next step of the model. Its calculation formula is:

[0078] o t = σ(W o ·[h t-1 , x t + b o )

[0079] h t= o t *tanh(C t )

[0080] where o t represents the output of the sigmoid function of the output gate; C t represents the updated memory cell state; h t represents the final hidden state output.

[0081] It can be understood that the sigmoid function output of the output gate can be regarded as a "valve", which determines how much information in the memory cell state (i.e., the memory containing historical power load information and current market conditions) at each time step should be output to the next step of the model. C t As the updated memory cell state, it is the core component for storing long-term information in the LSTM. It is updated at each time step according to the outputs of the input gate and the forget gate. Exemplarily, it can be regarded as a "knowledge base" maintained by the LSTM model, which contains a synthesis of useful historical information and current market conditions, reflecting the model's memory of past power load changes and the possible impact of the current market conditions on the power load. It is updated at each time step.

[0082] The final hidden state output is the output generated by the LSTM at the end of each time step. It contains not only the information of the current time step but also the information passed down through the memory cell from all previous time steps. It can be regarded as the "comprehensive understanding" of the power load prediction by the model at the current time step, and it will be used as one of the input features for the next time step to help the model capture the long-term dependencies in the time series data. That is, the hidden state output is the result of the model's comprehensive analysis of the current market conditions and historical power load information.

[0083] Furthermore, after the LSTM model is trained, during application, it is also necessary to preprocess the first time series data and organize it into a three-dimensional array format suitable for input to the LSTM model, including the number of samples, time step length, and feature dimension, and then input it into the LSTM model.

[0084] S220. Perform time series prediction on the feature representations in the first type of feature dimension through the LSTM model to obtain bid-related time series data.

[0085] Specifically, through this pre-trained LSTM model, perform time series prediction on the feature representations in the first type of feature dimension to obtain bid-related time series data that reflects the changing trends and patterns of each feature in the first type of feature dimension over time.

[0086] Optionally, the predicted time-series data related to bidding prices can also be analyzed to evaluate the prediction performance of the model. Based on the analysis results, the parameters and structure of the LSTM model can be further adjusted and optimized to improve the accuracy and reliability of predictions in future applications.

[0087] In the above embodiment, using LSTM as the time-series prediction model can effectively capture the time-series change patterns and long-term dependence relationships of feature manifestations in the first type of feature dimension, thereby accurately predicting the time-series data related to bidding prices. Through its unique gating mechanism and memory units, LSTM effectively reduces the problems of gradient vanishing and gradient explosion when processing long-sequence data, enabling the model to better utilize historical information and current input features for comprehensive analysis and generate prediction results with high accuracy and reliability, providing a solid foundation for subsequent medium- and long-term electricity price predictions.

[0088] In some embodiments, the second type of feature dimension includes any one of the economic development situation, coal price, liquefied natural gas price, and climate change situation.

[0089] It can be understood that since the second type of feature dimension refers to those dimensions that do not directly participate in the bidding process but have an indirect impact on electricity prices, it can include the economic development situation, coal price, liquefied natural gas price, or climate change situation, etc. Similarly, the second type of feature dimension can also include technological progress factors such as power generation technology innovation or energy storage technology development, social factors such as population growth and population structure or changes in consumption habits, and international factors such as supply and demand changes or price fluctuations in the international energy market.

[0090] Taking the economic development situation as an example, illustratively, in terms of the change in industrial electricity demand, the rapid economic development will be accompanied by an increase in industrial production and commercial activities, resulting in an increase in industrial electricity demand. For example, during the economic growth period, the expansion of the manufacturing industry or the increase in the factory operating rate leads to a significant increase in industrial electricity consumption, further increasing the total demand in the electricity market and driving up electricity prices. On the contrary, during an economic recession, industrial production slows down, electricity demand decreases, and electricity prices may face a downward trend. Similarly, it is also reflected in the change in residents' consumption levels. With the development of the economy, residents' income levels increase, consumption capabilities enhance, and the demand for electricity also increases. For example, each household may purchase more electrical appliances, such as air conditioners or washing machines, etc., thereby increasing residents' electricity consumption and indirectly affecting electricity prices.

[0091] In the above embodiment, by incorporating the second type of feature dimension such as the economic development situation into the electricity price prediction model, factors that do not directly participate in the bidding process but have an important indirect impact on electricity prices can be more comprehensively captured. This helps to improve the accuracy and reliability of medium- and long-term electricity price predictions and enables the prediction results to better reflect the real dynamics and future trends of the market.

[0092] In some embodiments, medium- and long-term electricity price forecasting is performed based on auction-related time series data and operation-related time series data to obtain a forecast electricity price. Please refer to the appendix Figure 3 , including:

[0093] S310. Perform feature dynamic enhancement processing on the auction-related time series data and the operation-related time series data to obtain a predicted feature vector corresponding to the auction-related time series data and a historical feature vector corresponding to the operation-related time series data.

[0094] S320. Use the predicted feature vector, the historical feature vector, the auction-related time series data, and the operation-related time series data to perform medium- and long-term electricity price forecasting to obtain a forecast electricity price.

[0095] Among them, the feature dynamic enhancement processing may refer to a method for feature extraction and optimization processing of time series data. The feature dynamic enhancement processing can strengthen the prediction model's ability to capture key features by dynamically adjusting the expression ability and importance of features, so as to obtain feature vectors containing important information. The predicted feature vector may refer to a feature vector that contains key feature information in the auction-related time series data after being extracted and enhanced by the model. Similarly, the historical feature vector may refer to a feature vector that contains key feature information in the operation-related time series data after being extracted and enhanced by the model.

[0096] Specifically, the feature dynamic enhancement processing of the auction-related time series data and the operation-related time series data can be implemented through a gated residual network and Self-Attention (self-attention mechanism). Exemplarily, first, the obtained auction-related time series data and operation-related time series data can be preprocessed to ensure the integrity and consistency of the data. Then, the feature dynamic enhancement processing is performed through the gated residual network in the prediction model. The gating mechanism controls the degree of non-linear transformation by introducing a gating unit (such as a gated linear unit, GLU), and flexibly adjusts the expression of features. The residual connection allows the network to combine features at different levels, that is, instead of directly mapping the input features to the output, the input is directly added to the output across some network layers, thereby enhancing the prediction model's ability to process time series information and avoiding the problem of gradient disappearance in the training of deep networks. Further, based on the characteristics of the input auction-related time series data and operation-related time series data, the gated residual network can dynamically adjust the intensity of feature extraction, so as to better capture the dynamic changes in the time series data, and finally screen out the feature vectors that contribute the most to the forecast electricity price.

[0097] It can be understood that a Self-Attention (self-attention mechanism) is also connected after the gated residual network. Since the Self-Attention (self-attention mechanism) can capture the dependencies between any two positions in the input sequence rather than just local or adjacent positions, it can identify the features that have the greatest impact on predicting electricity prices throughout the time series. Then, based on the mutual relationships between the features and their contribution degrees to the prediction target, the attention scores between each feature and other features are calculated, so as to assign different weights to them and obtain the final attention weights. Finally, these weights are used to perform weighted summation on the value vectors to obtain the comprehensive representation of each feature.

[0098] Further, after obtaining the comprehensive representation of each feature, it can be used for medium- and long-term electricity price prediction to obtain the predicted electricity price. Specifically, first, the bid-related time series data and operation-related time series data processed by the gated residual network and Self-Attention are integrated. At this time, each feature has obtained its comprehensive representation, including the global importance in the time series and the weight value of its contribution degree to the prediction target. Further, using the fully connected layer of the prediction model, based on the mapping relationship between the input features and the predicted electricity price, the final prediction result is obtained, that is, using the model to predict the future electricity price, and the final predicted electricity price result can be output.

[0099] In the above embodiment, by combining the gated residual network and Self-Attention for feature dynamic enhancement processing and feature importance evaluation, the features that contribute the most to predicting electricity prices in the bid-related time series data and operation-related time series data are accurately identified and utilized, their weights are dynamically adjusted, the expression ability of key features is enhanced, and the dynamic changes and long-term dependencies in the time series data are effectively captured, so as to give full play to the role of these features in medium- and long-term electricity price prediction, and finally the accuracy and stability of the prediction are improved.

[0100] In some embodiments, the operation-related time series data further includes second time series data on a second feature dimension.

[0101] Among them, the second feature dimension is a feature dimension that has nothing to do with the bidding space.

[0102] Perform feature dynamic enhancement processing on the bid-related time series data and operation-related time series data to obtain the predicted feature vector corresponding to the bid-related time series data and the historical feature vector corresponding to the operation-related time series data. Please refer to Figure 4 , including:

[0103] S410. Perform feature dynamic enhancement processing on the auction-related time series data, the first time series data, and the second time series data to obtain a first feature vector corresponding to the first time series data, a second feature vector corresponding to the second time series data, and a prediction feature vector.

[0104] Among them, the first feature vector refers to the feature vector obtained after the feature dynamic enhancement processing of the first time series data extracted from the first type of feature dimension (the feature dimension related to the auction space), which reflects the importance of the features on the first type of feature dimension in the prediction process. The second feature vector refers to the feature vector obtained after the feature dynamic enhancement processing of the second time series data extracted from the second type of feature dimension (the feature dimension unrelated to the auction space). It reflects the importance of the features on the second type of feature dimension in the prediction process.

[0105] Similarly, the feature dynamic enhancement processing of the auction-related time series data, the first time series data, and the second time series data can also be implemented through a gated residual network and Self-Attention (self-attention mechanism). Use the gated residual network in the prediction model to automatically control the flow of information to flexibly adjust the expression of features, dynamically adjust the weights of the input data, and obtain the comprehensive representation of each feature. That is, obtain a first feature vector corresponding to the first time series data, a second feature vector corresponding to the second time series data, and a prediction feature vector.

[0106] Correspondingly, use the prediction feature vector, the historical feature vector, the auction-related time series data, and the operation-related time series data to perform medium- and long-term electricity price prediction to obtain the predicted electricity price, including:

[0107] S420. Perform medium- and long-term electricity price prediction according to the first feature vector, the second feature vector, the prediction feature vector, the first time series data, the second time series data, and the auction-related time series data to obtain the predicted electricity price.

[0108] Similarly, after obtaining the comprehensive representation of each feature, the fully connected layer of the prediction model can be used to predict the future electricity price based on the mapping relationship between the input features and the predicted electricity price, and based on the first feature vector, the second feature vector, the prediction feature vector, the first time series data, the second time series data, and the auction-related time series data, and then output the final predicted electricity price result.

[0109] In the above embodiments, by subdividing the operation-related time-series data into a first type of feature dimension related to the bidding space and a second type of feature dimension unrelated to the bidding space, and performing feature dynamic enhancement processing on these two types of feature data, it is possible to more comprehensively capture the change trends and laws of factors that are unrelated to the bidding space but have an indirect impact on electricity prices. With the help of the gated residual network and the self-attention mechanism, the model can dynamically adjust the feature weights and generate a first feature vector, a second feature vector, and a predicted feature vector reflecting the importance of each feature. These enhanced feature vectors, combined with the original time-series data, provide richer information for medium- and long-term electricity price forecasting, thus significantly improving the accuracy and reliability of the forecasting and better meeting the needs of medium- and long-term electricity price forecasting in the power market.

[0110] In some embodiments, the first type of feature dimension includes any one of power load, wind power output, photovoltaic power output, and tie line.

[0111] It can be understood that since the first type of feature dimension is a dimension that characterizes the factors mainly involved in affecting the electricity market bidding process and electricity price formation, it can include power load, wind power output, photovoltaic power output, or tie line, etc. Further, the first type of feature dimension can also include a series of power generation side factors such as the fuel cost and operation and maintenance cost of thermal power generation, the reservoir water storage and river flow of hydropower generation, etc., can also include the grid transmission capacity factors such as the transmission line capacity or transformer capacity of the power grid, and can also include market participant factors such as the sensitivity of power users to electricity price changes and demand elasticity.

[0112] Taking the power load as an example for illustration, exemplarily, since the power load is a direct manifestation of demand in the electricity market, the increase or decrease of the power load directly affects the power supply-demand balance. When the power load increases, industrial production expansion or residential electricity consumption may both rise, thus increasing the market demand for electricity. At this time, if the supply cannot keep up in time, it will lead to a power shortage, thus driving up the electricity price. On the contrary, when the power load decreases, such as when economic activities slow down or energy-saving measures are implemented, the power supply is relatively excessive at this time, and the electricity price may decline. In addition, the volatility of the power load will also cause fluctuations in the electricity price. For example, during peak electricity consumption periods (such as high temperatures in summer or heating in winter), the power load may surge, resulting in a tight power supply, so the electricity price tends to rise during these periods; while during off-peak electricity consumption periods, the load decreases, resulting in a relatively low electricity price.

[0113] In the above embodiments, by refining the first type of feature dimension into specific factors such as power load, wind power output, photovoltaic power output, and tie lines, key factors affecting the electricity market bidding process and electricity price formation can be captured and analyzed more accurately, enabling the prediction model to specifically evaluate the specific impacts of each factor on power supply-demand balance, market competitiveness, and price fluctuations, thereby improving the accuracy and reliability of the prediction results.

[0114] The embodiments of this specification also provide a medium- and long-term electricity price prediction method suitable for the electricity market. The method includes the following steps:

[0115] S802. Obtain the time-series data related to the operation of the electricity market.

[0116] S804. Obtain the first time-series data on the first type of feature dimension from the time-series data related to the operation according to the first type of feature dimension.

[0117] Among them, the first type of feature dimension is the feature dimension related to the bidding space. The first type of feature dimension includes any one of power load, wind power output, photovoltaic power output, and tie lines.

[0118] S806. Obtain the second time-series data on the second type of feature dimension from the time-series data related to the operation according to the second type of feature dimension.

[0119] Among them, the second type of feature dimension is the feature dimension unrelated to the bidding space. The second type of feature dimension includes any one of the economic development situation, coal price, liquefied natural gas price, and climate change situation.

[0120] S808. Input the first time-series data into the LSTM model.

[0121] S810. Perform time-series prediction on the feature performance on the first type of feature dimension through the LSTM model to obtain the time-series data related to bidding.

[0122] S812. Perform feature dynamic enhancement processing on the time-series data related to bidding, the first time-series data, and the second time-series data to obtain the first feature vector corresponding to the first time-series data, the second feature vector corresponding to the second time-series data, and the prediction feature vector.

[0123] S814. Perform medium- and long-term electricity price prediction according to the first feature vector, the second feature vector, the prediction feature vector, the first time-series data, the second time-series data, and the time-series data related to bidding to obtain the predicted electricity price.

[0124] It should be understood that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this document, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above flowchart may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0125] The embodiments of this specification also provide a medium- and long-term electricity price prediction device 500 suitable for the electricity market, as Figure 5 shown, including: a time series data acquisition module 510, a dimension data acquisition module 520, a time series prediction module 530, and an electricity price prediction module 540, where:

[0126] The time series data acquisition module 510 is used to acquire the time series data related to the operation of the electricity market.

[0127] The dimension data acquisition module 520 is used to obtain the first time series data on the first type of feature dimension from the operation-related time series data; wherein, the first type of feature dimension is a feature dimension related to the bidding space.

[0128] The time series prediction module 530 is used to perform time series prediction on the feature performance on the first type of feature dimension according to the first time series data to obtain the time series data related to bidding.

[0129] The electricity price prediction module 540 is used to perform medium- and long-term electricity price prediction based on the time series data related to bidding and the operation-related time series data to obtain the predicted electricity price.

[0130] In some embodiments, the second time series data on the second type of feature dimension is obtained from the operation-related time series data according to the second type of feature dimension; wherein, the second type of feature dimension is a feature dimension unrelated to the bidding space. Correspondingly, the dimension data acquisition module 520 is also used to perform medium- and long-term electricity price prediction based on the time series data related to bidding and the operation-related time series data to obtain the predicted electricity price, including: performing medium- and long-term electricity price prediction based on the time series data related to bidding, the first time series data, and the second time series data to obtain the predicted electricity price.

[0131] In some embodiments, the time series prediction module 530 is further used to input the first time series data into the LSTM model; perform time series prediction on the feature performance on the first type of feature dimension through the LSTM model to obtain the time series data related to bidding.

[0132] In some embodiments, the dimension data acquisition module 520 is further configured to determine that the second type of feature dimension includes any one of economic development situation, coal price, liquefied natural gas price, and climate change situation.

[0133] In some embodiments, the electricity price prediction module 540 is further configured to perform feature dynamic enhancement processing on the bidding-related time series data and the operation-related time series data to obtain a predicted feature vector corresponding to the bidding-related time series data and a historical feature vector corresponding to the operation-related time series data; and use the predicted feature vector, the historical feature vector, the bidding-related time series data, and the operation-related time series data to perform medium- and long-term electricity price prediction to obtain a predicted electricity price.

[0134] In some embodiments, the operation-related time series data further includes second time series data on the second type of feature dimension; wherein, the second type of feature dimension is a feature dimension irrelevant to the bidding space; performing feature dynamic enhancement processing on the bidding-related time series data and the operation-related time series data to obtain a predicted feature vector corresponding to the bidding-related time series data and a historical feature vector corresponding to the operation-related time series data includes: performing feature dynamic enhancement processing on the bidding-related time series data, the first time series data, and the second time series data to obtain a first feature vector corresponding to the first time series data, a second feature vector corresponding to the second time series data, and the predicted feature vector. The dimension data acquisition module 520 is further configured to correspondingly use the predicted feature vector, the historical feature vector, the bidding-related time series data, and the operation-related time series data to perform medium- and long-term electricity price prediction to obtain a predicted electricity price, including: performing medium- and long-term electricity price prediction according to the first feature vector, the second feature vector, the predicted feature vector, the first time series data, the second time series data, and the bidding-related time series data to obtain a predicted electricity price.

[0135] In some embodiments, the dimension data acquisition module 520 is further configured to determine that the first type of feature dimension includes any one of power load, wind power output, photovoltaic power output, and tie line.

[0136] For the specific limitations on a medium- and long-term electricity price prediction device suitable for the electricity market, reference may be made to the limitations on a medium- and long-term electricity price prediction method suitable for the electricity market in the foregoing text, which will not be elaborated herein. Each module in the above-mentioned medium- and long-term electricity price prediction device suitable for the electricity market can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above-mentioned modules.

[0137] A medium- and long-term electricity price prediction device suitable for the electricity market in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0138] An embodiment of the present application also provides a computer device, which can be a terminal, and its internal structure diagram can be as Figure 6 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a medium- and long-term electricity price prediction method suitable for the electricity market. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0139] Those skilled in the art can understand that Figure 6 the structure shown in

[0140] Embodiments of the present application also provide a computer-readable storage medium. The methods according to the embodiments of the present application can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the methods described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0141] Embodiments of the present application provide a computer program product, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods of any embodiment of the present application.

[0142] A medium- and long-term electricity price prediction method, device, and computer device suitable for the electricity market clarified in the above embodiments can be specifically implemented by a computer chip or entity, or implemented by a product with a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0143] For the convenience of description, when describing the above device, various units are described separately according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0144] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0145] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0146] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0147] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0148] In the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0149] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0150] It should also be noted that the term "comprise", "include" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising said element.

[0151] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. Since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, reference can be made to the partial description of the method embodiment.

[0152] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

[0153] Although the embodiments of the present application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A medium- and long-term electricity price forecasting method suitable for the electricity market, characterized in that: The method comprises: Obtaining time series data related to the operation of the power market; Obtaining first time series data on the first type of feature dimension from the operation-related time series data according to the first type of feature dimension; wherein the first type of feature dimension is a feature dimension related to the bidding space; Performing time series prediction on the feature performance of the first type of feature dimension according to the first time series data to obtain bidding-related time series data; A medium- and long-term electricity price forecast is performed based on the bidding-related time series data and the operation-related time series data to obtain a forecasted electricity price.

2. The method according to claim 1, characterized in that The method further comprises: Obtaining second time series data on the second type of feature dimension from the operation-related time series data according to the second type of feature dimension; wherein the second type of feature dimension is a feature dimension that is unrelated to the bidding space; Accordingly, the mid- to long-term electricity price forecasting based on the bidding-related time series data and the operation-related time series data to obtain the forecasted electricity price includes: A medium- and long-term electricity price forecast is performed based on the bidding-related time series data, the first time series data, and the second time series data to obtain the forecasted electricity price.

3. The method according to claim 2, characterized in that The step of performing time series prediction on the feature performance of the first type of feature dimension according to the first time series data to obtain bidding-related time series data includes: Inputting the first time series data into the LSTM model; The LSTM model is used to perform time series prediction on the feature performance of the first type of feature dimension to obtain the bidding related time series data.

4. The method according to claim 2, characterized in that: The second type of characteristic dimensions includes any one of economic development, coal prices, liquefied natural gas prices, and climate change.

5. The method according to claim 1, characterized in that The performing of medium- and long-term electricity price forecasting based on the bidding-related time series data and the operation-related time series data to obtain the forecasted electricity price includes: Performing dynamic feature enhancement processing on the bidding-related time series data and the operation-related time series data to obtain a prediction feature vector corresponding to the bidding-related time series data and a historical feature vector corresponding to the operation-related time series data; The predicted electricity price is obtained by using the predicted characteristic vector, the historical characteristic vector, the bidding-related time series data and the operation-related time series data to perform medium- and long-term electricity price forecasting.

6. The method according to claim 5, characterized in that The operation-related time series data further includes second time series data on a second type of feature dimension; wherein the second type of feature dimension is a feature dimension that is unrelated to the bidding space; Performing dynamic feature enhancement processing on the bidding-related time series data and the operation-related time series data to obtain a prediction feature vector corresponding to the bidding-related time series data and a historical feature vector corresponding to the operation-related time series data, including: Performing dynamic feature enhancement processing on the bidding-related time series data, the first time series data, and the second time series data to obtain a first feature vector corresponding to the first time series data, a second feature vector corresponding to the second time series data, and the prediction feature vector; Accordingly, the use of the prediction feature vector, the historical feature vector, the bidding-related time series data, and the operation-related time series data to perform medium- and long-term electricity price forecasting to obtain the predicted electricity price includes: A medium- and long-term electricity price forecast is performed based on the first feature vector, the second feature vector, the predicted feature vector, the first time series data, the second time series data, and the bidding-related time series data to obtain the predicted electricity price.

7. The method according to any one of claims 1 to 6, characterized in that: The first type of feature dimensions includes any one of power load, wind power output, photovoltaic output, and interconnection lines.

8. A medium- and long-term electricity price forecasting device suitable for the electricity market, characterized in that: The device comprises: A time series data acquisition module is used to obtain time series data related to the operation of the power market; A dimension data acquisition module, configured to obtain first time series data on the first type of feature dimension from the operation-related time series data according to the first type of feature dimension; wherein the first type of feature dimension is a feature dimension related to the bidding space; A time series prediction module, configured to perform time series prediction on the feature performance of the first type of feature dimension according to the first time series data to obtain bidding-related time series data; The electricity price prediction module is used to perform medium- and long-term electricity price prediction based on the bidding-related time series data and the operation-related time series data to obtain a predicted electricity price.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images