A spot market price forecasting method and related apparatus

Abstract

This application belongs to the field of electricity spot market price forecasting technology. Addressing the technical problems of existing electricity spot market price forecasting methods, which struggle to effectively integrate multi-source data and accurately depict the coupled changes in electricity prices under renewable energy access conditions, resulting in insufficient forecasting accuracy and stability, this application proposes a spot market price forecasting method and related apparatus. This method constructs a multi-source input dataset by acquiring multi-source historical data from the power system and performs variational mode decomposition on historical clearing price data, decomposing the price sequence into multiple modal components to enhance feature dimensions. The modal component sequences are combined with the multi-source input data to form an enhanced feature dataset, which is then input into a prediction model containing convolutional neural networks and long short-term memory networks. The convolutional neural network is used to extract local features, and the long short-term memory network is used to learn temporal dependencies, thereby enabling the prediction of clearing prices at multiple future time points, improving forecasting accuracy and stability.

Description

Technical Field

[0001] This application pertains to a forecasting method, specifically a spot market price forecasting method and related apparatus. Background Technology

[0002] Electricity is a crucial basic energy source for modern society, and its resource allocation efficiency directly impacts energy security and economic development. As the power industry shifts from a planned economy to a market-based competitive model, the electricity spot market has gradually become an important mechanism for achieving optimal resource allocation, leading to increasingly complex spot price formation processes. In a market-driven environment, spot electricity prices are influenced by multiple factors, including changes in supply and demand, market dynamics, and fluctuations in renewable energy output, exhibiting significant uncertainty and volatility. Traditional methods relying on experience or simple statistical approaches are insufficient to meet the accuracy requirements of price forecasting in the spot market.

[0003] To address the issue of efficient power resource allocation, existing technologies primarily rely on power system reform and spot market mechanisms to achieve market-based pricing. Power system reform breaks down the existing monopoly, introducing a competitive bidding mechanism for power generation, allowing power generation companies and users to directly participate in market transactions. Market prices are formed through volume-based bidding and marginal clearing. Simultaneously, with the increasing proportion of renewable energy installed capacity, relevant policies are promoting the gradual marketization of renewable energy prices, enabling spot prices to more accurately reflect supply and demand. Based on this, existing technologies typically establish price forecasting models using historical transaction data, load data, or statistical analysis methods to assist in market transaction decisions and power system dispatch.

[0004] While existing technologies have improved the marketization of electricity pricing, they still have shortcomings in price forecasting. Because spot prices are influenced by a combination of factors, including load changes, renewable energy output, weather conditions, and market trading behavior, and because renewable energy is random and intermittent, traditional forecasting methods based on empirical rules or single-data modeling struggle to accurately reflect complex supply and demand relationships. Furthermore, existing forecasting technologies lack the ability to integrate multi-source data, resulting in limited stability and accuracy of forecasts, making it difficult to meet the needs of real-time dispatching, risk control, and refined decision-making by market participants in the electricity spot market. Summary of the Invention

[0005] This application addresses the technical problem that existing electricity spot market price forecasting methods are unable to effectively integrate multi-source data and accurately depict the multi-factor coupling change law of electricity prices under the condition of new energy access, resulting in insufficient forecasting accuracy and stability. It provides a spot market price forecasting method and related device.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application proposes a method for predicting spot market prices, including: Acquire multi-source historical data of the power system, and construct a multi-source input dataset for spot market price forecasting based on the multi-source historical data; Variational mode decomposition is performed on the historical cleared electricity price data to decompose the historical cleared electricity price data into multiple modal component sequences; The modal component sequence is combined with the multi-source input dataset to obtain an enhanced feature dataset; The enhanced feature dataset is input into the spot market price prediction model to obtain the spot market clearing electricity price prediction results at multiple future time points. The spot market price prediction model is equipped with a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local features from the multi-source input dataset and modal component sequences, and the long short-term memory network is used to learn the temporal dependency relationship of the feature vector output by the convolutional neural network.

[0007] Furthermore, the multi-source historical data includes provincial dispatch load data, photovoltaic power data, wind power power data, hydropower output data, inter-provincial interconnection line output data, and non-market-based unit output data.

[0008] Furthermore, the variational mode decomposition process includes: performing frequency domain decomposition on historical clearing electricity price data, and obtaining multiple independent mode component sequences through iterative optimization.

[0009] Furthermore, the convolutional neural network is a one-dimensional convolutional neural network, including two convolutional blocks. Each convolutional block includes a convolutional layer, a ReLU activation function, and a pooling layer. The convolutional layer is used to perform convolution operations on the enhanced feature dataset, the ReLU activation function is used to introduce non-linearity to learn complex features, and the pooling layer is used to downsample the convolution operation results.

[0010] Furthermore, when the Long Short-Term Memory network performs temporal dependency learning on the feature vector output by the convolutional neural network, it calculates the hidden state at the current time step based on the feature vector output by the current convolutional neural network and the hidden state at the previous time step.

[0011] Furthermore, the long short-term memory network includes a forget gate, an input gate, and an output gate; The forget gate is used to calculate the forgetting coefficient based on the feature vector output by the current convolutional neural network and the hidden state at the previous time step, and to update the cell state at the previous time step based on the forgetting coefficient. The input gate is used to calculate input coefficients based on the feature vector output by the current convolutional neural network and the hidden state at the previous time step, and to generate candidate cell states, which are then combined with the updated cell states to obtain the current cell state. The output gate is used to calculate the output coefficients based on the current cell state and the feature vector output by the current convolutional neural network, and to obtain the hidden state at the current time based on the output coefficients.

[0012] Furthermore, the spot market price prediction model also includes a fully connected layer; The fully connected layer is used to linearly map the feature vector output by the long short-term memory network to obtain the clearing price prediction values ​​at multiple time points.

[0013] Secondly, this application proposes a spot market price forecasting system, comprising: The data module is used to acquire multi-source historical data of the power system and construct a multi-source input dataset for spot market price prediction based on the multi-source historical data. The decomposition module is used to perform variational mode decomposition processing on historical cleared electricity price data, decomposing the historical cleared electricity price data into multiple modal component sequences; The combination module is used to combine the modal component sequence with the multi-source input dataset to obtain an enhanced feature dataset; The prediction module is used to input the enhanced feature dataset into the spot market price prediction model to obtain the spot market clearing electricity price prediction results at multiple future time points. The spot market price prediction model is equipped with a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local features from the multi-source input dataset and modal component sequences, and the long short-term memory network is used to learn the temporal dependency relationship of the feature vector output by the convolutional neural network.

[0014] Thirdly, this application provides a computer device, including: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the above-described spot market price forecasting method.

[0015] Fourthly, this application proposes a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed by the above-described spot market price forecasting method.

[0016] Compared with the prior art, this application has the following beneficial effects: This application proposes a spot market price forecasting method. It constructs a multi-source input dataset by acquiring multi-source historical data from the power system and performs variational mode decomposition (VMD) on historical clearing price data, decomposing the original price sequence into multiple modal components at different frequency scales. This enhances the richness of the input features and strengthens the ability to express the non-stationary characteristics of electricity prices. Furthermore, the decomposed modal component sequences are combined with the multi-source input data to form an enhanced feature dataset, which is then input into a spot market price forecasting model incorporating a convolutional neural network (CNN) and a long short-term memory (LSTM) network. The CNN is used to extract local feature information from the input data, while the LSTM network is used to learn long-term dependencies in the time series. This effectively characterizes the electricity price change patterns under the coupled conditions of multiple factors such as renewable energy output, load changes, and inter-regional power exchange, improving the accuracy and stability of spot market price forecasting. It also enables continuous forecasting of clearing prices at multiple future time points, enhancing the model's adaptability and generalization ability in complex electricity spot market environments.

[0017] This application also proposes a spot market price forecasting system, an electronic device, and a computer-readable storage medium, which possess all the advantages of the aforementioned spot market price forecasting methods. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating one of the spot market price forecasting methods described in this application; Figure 2 This is a schematic diagram of a spot market price forecasting method in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the convolutional neural network in the embodiments of this application; Figure 4 This is a schematic diagram of the LSTM recirculation unit (cell) structure in the embodiments of this application; Figures 5(a) to 5(j) are schematic diagrams of the day-ahead clearing electricity price prediction results for 10 test samples in the embodiments of this application; Figures 6(a) to 6(j) are schematic diagrams of the real-time clearing electricity price prediction results of 10 test samples in the embodiments of this application; Figure 7 This is a schematic diagram of the spot market price forecasting system of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In recent years, the large-scale grid connection of new energy sources such as photovoltaic and wind power has led to the randomness, volatility, and intermittency of new energy output, making the operation of the power system more complex and the spot market price more uncertain. Therefore, in actual operation, it is usually necessary to use multi-source information such as historical clearing price data, load data, new energy output data, meteorological data, and unit operation status data, and use price forecasting technology to estimate the spot price for future periods in advance, so as to assist market participants in decision-making and ensure the safe and stable operation of the power system.

[0022] In practical applications, electricity spot market price forecasting faces significant challenges. Because spot electricity prices are determined by a combination of factors, including supply and demand, generator operation constraints, cybersecurity constraints, market participant dynamics, and fluctuations in renewable energy output, price series typically exhibit significant nonlinearity, non-stationarity, and strong stochasticity. This is especially true as the proportion of renewable energy sources continues to increase; wind and solar power output is heavily influenced by weather conditions, and their fluctuations directly alter marginal generator capacity, leading to drastic price changes. Furthermore, the spot market employs a short-cycle rolling clearing mechanism, resulting in frequent price changes and coupling relationships between different time scales, such as the superposition of intraday fluctuations with weekly and seasonal cycles, giving the price series multi-scale characteristics. Under these circumstances, relying solely on single historical electricity price data is insufficient to fully reflect the market's operational status. Moreover, multi-source data suffers from high dimensionality, complex correlations, and significant noise, leading to insufficient accuracy, poor stability, and weak generalization ability in price forecasting models in practical applications, making it difficult to meet the high-precision forecasting requirements of the electricity market.

[0023] To address the challenge of predicting electricity spot market prices, existing technologies typically employ statistical analysis methods or intelligent algorithms for modeling and forecasting. Early methods often used time series models, such as autoregressive models, autoregressive moving average models, and differential autoregressive moving average models, to achieve predictions by analyzing the temporal correlation of historical electricity price data. These methods are simple in structure and computationally inexpensive, but struggle to characterize complex nonlinear relationships. With the development of artificial intelligence, researchers have gradually adopted machine learning methods, such as support vector machines, random forests, and neural network models, to improve prediction accuracy. Furthermore, deep learning methods have been introduced into the field of electricity price prediction, such as long short-term memory networks, gated recurrent units, and convolutional neural networks, to improve prediction performance by extracting deep features from time series data. In addition, to reduce the complexity of price series data, some methods employ signal decomposition techniques before modeling, such as empirical mode decomposition, ensemble empirical mode decomposition, or wavelet transform. These techniques decompose the original electricity price series into multiple sub-sequences, perform predictions for each sub-sequence separately, and then reconstruct the data to improve model stability.

[0024] While existing technologies have improved the accuracy of electricity spot market price forecasting to some extent, shortcomings remain in practical applications. On one hand, traditional time series models have limited adaptability to nonlinear and non-stationary data, and while single deep learning models can fit complex relationships, they are easily affected by changes in the distribution of training data, resulting in insufficient stability of prediction results in situations with a high proportion of renewable energy and strong market volatility. On the other hand, existing signal decomposition methods are prone to mode aliasing or unstable decomposition results when processing electricity price series, making it difficult for subsequent prediction models to fully extract effective features. Furthermore, most existing methods only consider a single data source or simply superimpose multi-source data, failing to fully explore the coupling relationships between multiple sources of information such as load, renewable energy output, weather, and unit status. This leads to insufficient adaptability of the models to complex market environments, making it difficult to maintain high prediction accuracy and robustness under highly volatile and uncertain spot market conditions. Therefore, how to improve the stability of electricity price series decomposition under multi-source data conditions and enhance the ability of prediction models to characterize complex volatility characteristics remains a pressing issue in electricity spot market price forecasting technology.

[0025] Based on the above, this application proposes a spot market price forecasting method and related apparatus. The following detailed description of this application is provided in conjunction with embodiments and accompanying drawings.

[0026] like Figure 1 The diagram shown is a flowchart of one method for predicting spot market prices according to this application, which may include: S101, acquire multi-source historical data of the power system, and construct a multi-source input dataset for spot market price prediction based on the multi-source historical data.

[0027] In practical applications, historical operating data, load data, renewable energy output data, and historical clearing price data can be obtained through power dispatch automation systems, energy management systems, and power market trading systems. These data types are then uniformly read and stored via a data processing server. Since the sampling periods and timestamps of different data sources may vary, this embodiment also performs time alignment, outlier removal, and missing value completion on various historical data, and normalizes or standardizes the data to construct a unified format for a multi-source input dataset. Through these methods, the input data can simultaneously reflect the operating status of the power system and changes in market supply and demand, thus providing a complete data foundation for subsequent price forecasting. In some embodiments of this application, the multi-source historical data may also include meteorological data or unit maintenance status data to further improve the comprehensiveness of the input features.

[0028] S102, perform variational mode decomposition on the historical cleared electricity price data to decompose the historical cleared electricity price data into multiple modal component sequences.

[0029] Historically cleared electricity prices typically exhibit significant non-stationarity and multi-scale fluctuations, making them unsuitable for direct prediction as they can easily reduce model stability. Therefore, Variational Mode Decomposition (VMD) is used to decompose the electricity price series. As an example, the VMD algorithm can be invoked on a data processing server to construct a variational optimization model for historically cleared electricity prices. Through iterative solving, several modal component sequences with different center frequencies are obtained, allowing the trend, periodic, and high-frequency fluctuation components of the original electricity price series to be extracted. This decomposition process reduces the complexity of the electricity price series, enabling different frequency features to be learned separately by subsequent prediction models, thereby improving the stability of the prediction process.

[0030] S103, combine the modal component sequence with the multi-source input dataset to obtain the enhanced feature dataset.

[0031] In practical applications, to enable the prediction model to simultaneously utilize the frequency characteristics derived from electricity price decomposition and the operational characteristics of the power system, it is necessary to align the sequence of each modal component with load data, renewable energy output data, and other historical operational data in chronological order. Feature concatenation processing is then performed on the data processing server to construct an enhanced feature dataset in the form of multi-dimensional feature vectors. This approach allows the input data to include both decomposition features reflecting the internal patterns of electricity price changes and operational features reflecting changes in supply and demand, thereby enhancing the model's ability to express the mechanism of electricity price formation.

[0032] S104, the enhanced feature dataset is input into the spot market price prediction model to obtain the spot market clearing electricity price prediction results for multiple future time points; the spot market price prediction model is equipped with a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local features from the multi-source input dataset and modal component sequences, and the long short-term memory network is used to learn the temporal dependency relationship of the feature vector output by the convolutional neural network.

[0033] In practical applications, spot market price forecasting models incorporate both Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. The CNN extracts local features from the multi-dimensional input features in the enhanced feature dataset. By using convolutional kernels to perform sliding operations across the time and feature dimensions, it extracts relevant pattern information from the data. After obtaining the feature vector output by the CNN, this vector is input into the LSTM network for temporal dependency learning. The LSTM network uses a gating structure to remember and update information from historical moments, thus capturing the long-term dependencies of electricity prices over time. This combined structure enables the spot market price forecasting model to extract local correlation information from multi-source features and learn the temporal variation patterns of price series, ultimately outputting the predicted spot market clearing electricity prices for multiple future time points.

[0034] This application first collects and preprocesses multi-source historical data of the power system, then performs variational mode decomposition on historical clearing prices, and fuses the decomposed modal components with multi-source operational data to construct an enhanced feature dataset. Subsequently, a prediction model incorporating convolutional neural networks and long short-term memory networks is used to extract features and learn time series data from the enhanced features, thereby enabling the prediction of spot market clearing prices at multiple future time points. This improves the accuracy and stability of spot market price prediction under multi-source data conditions. Compared with prediction methods that only use historical prices, this application reduces the non-stationarity of price sequences by introducing variational mode decomposition, enhances the completeness of input features through multi-source data fusion, effectively extracts multi-dimensional features using convolutional neural networks, and learns the time dependence of prices through long short-term memory networks. This allows the prediction model to more accurately reflect the impact of supply and demand changes and renewable energy fluctuations on prices in the power market, improving prediction accuracy and robustness in complex market environments, and demonstrating good engineering applicability.

[0035] The present application will be further described in detail below through other embodiments.

[0036] In this embodiment, the spot market price forecasting method is based on a deep learning model. It processes spot market operation data and combines variational mode decomposition, convolutional neural networks, and long short-term memory networks to predict spot market prices at multiple future points in time. For example... Figure 2 The diagram shown is a schematic representation of a spot market price forecasting method in this embodiment.

[0037] In this embodiment, the input to the spot market price prediction model consists of various power system operation data, including clearing price, provincial dispatch load, photovoltaic power, wind power, hydropower output, inter-provincial tie-line output, and non-market-based unit output. These multi-source input data characterize the power system's operating status and serve as input features for the spot market price prediction model. Clearing price data is first processed by the variational mode decomposition (VMD) module. The remaining input data is not decomposed but is used together with the decomposition results as input to the subsequent neural network. The VMD module, located at the top layer of the model structure, decomposes the clearing price sequence into multiple modal components at different frequency scales to increase the feature dimension of the input data. After VMD, the data enters the convolutional neural network (CNN) module, which extracts features from the input data. The CNN module includes convolutional layers, activation functions, and pooling layers to extract local feature information from the input sequence. Following the CNN, the feature data is input to the long short-term memory (LSTM) network module. Long Short-Term Memory (LSTM) networks are used to learn time series data to capture the dynamic characteristics of spot market price changes over time, thereby improving the predictive model's ability to characterize time correlations. The output of the LSTM network enters a fully connected layer, which maps the neural network's output to the prediction result. After the fully connected layer, the model outputs predictions for multiple time steps. Figure 2 The format is DAY×7, indicating that the prediction results cover multiple time periods. In the model's output section, Figure 2 The document provides forecast values ​​for multiple time points, denoted as Time1, Time2, Time3, ..., Time96, which are used to represent the predicted spot market prices at multiple future time points. Figure 2 The term "linear" in this context indicates that the model includes a linear mapping component. Therefore, this application achieves multi-step prediction of spot market prices by combining variational mode decomposition with convolutional neural networks and long short-term memory networks.

[0038] The following is a detailed description of each module.

[0039] (1) Variational mode decomposition.

[0040] In this embodiment, in order to improve the input feature dimension of the spot market price forecasting model and enhance the model's ability to characterize the multi-scale changes in the electricity price series, the clearing electricity price data is first subjected to variational mode decomposition.

[0041] Variational mode decomposition is an adaptive signal decomposition method that can decompose the original signal into several mode components with different center frequencies, thereby enabling multi-scale analysis of complex non-stationary signals.

[0042] Among them, the clearing price of the electricity spot market has obvious nonlinear and non-stationary characteristics. Directly predicting the original price sequence can easily lead to large prediction errors. Therefore, in this embodiment, the clearing price sequence is first decomposed, and then the multiple modal components after decomposition are used as the input of the prediction model.

[0043] Variational mode decomposition can represent the original electricity price signal as the sum of multiple modal components, as shown in the following equation:

[0044] in, This indicates the original clearing price signal. Indicates the first One modal component, This indicates the number of modal components obtained from the decomposition.

[0045] Through the above decomposition, the original signal can be decomposed into multiple sub-signals with different frequency scales, which makes it easier for the subsequent neural network to learn the variation patterns at different scales.

[0046] In the variational mode decomposition process, it is necessary to constrain each modal component to ensure that the modes do not overlap in the frequency domain. The constraint conditions are shown in the following equation:

[0047] in, The energy representing the modal components, This represents the modal components in the frequency domain. This represents the equilibrium parameters.

[0048] The above constraints are used to control the stability of the decomposition results.

[0049] To solve for each modal component, a variational optimization objective function needs to be constructed, which is expressed as follows in this embodiment:

[0050] in, The frequency domain representation of the original signal. Indicates the first One modal component, Represents the modal coefficients. Represents the penalty function. Indicates the number of modes. This represents the equilibrium parameters.

[0051] By iteratively solving the above optimization problem, multiple independent modal components can be obtained.

[0052] In this embodiment, the clearing electricity price sequence is decomposed into multiple modal components through variational mode decomposition, and each modal component, together with other input variables, is used as the input of a convolutional neural network to improve the prediction accuracy of the spot market price prediction model.

[0053] (2) Feature extraction from convolutional neural networks After completing the variational mode decomposition of the cleared electricity price data, the multiple modal components obtained from the decomposition are fed into the convolutional neural network module along with other input data to extract feature information from the input data.

[0054] Convolutional neural networks (CNNs) are neural network structures with local connectivity and weight sharing characteristics. They are capable of extracting local feature information from input data and are suitable for feature learning from multidimensional input sequences. For example... Figure 2 As shown, the convolutional neural network module follows the variational mode decomposition module. Its input consists of the clearing price mode component obtained from variational mode decomposition, as well as provincial dispatch load, photovoltaic power, wind power, hydropower output, inter-provincial interconnection line output, and non-market-based unit output. These input data collectively constitute the input features of the convolutional neural network. Figure 3 The diagram shows the structure of the convolutional neural network in this embodiment. The convolutional neural network includes convolutional layers, activation function layers, and pooling layers. The convolutional layers perform convolution operations on the input data, extracting local feature information from the input sequence through sliding calculations of the convolutional kernel across the input data. Convolution operations enhance the model's ability to perceive local changes in the input data, thereby improving the extraction effect of spot market price change patterns. After the convolutional layers, the ReLU activation function is used to perform a nonlinear transformation on the convolution result. The ReLU function introduces nonlinear characteristics, enabling the neural network to represent more complex input-output relationships, thereby improving the model's ability to fit nonlinear electricity price changes. After the activation function, a pooling layer is set to perform dimensionality reduction on the convolutional features. By performing convergence calculations on local regions, the feature dimension is reduced while retaining the main feature information. Pooling reduces the computational complexity of the model and improves the stability of the features. After processing by the convolutional layers, activation function layers, and pooling layers, the resulting feature vector is used as input to the subsequent Long Short-Term Memory (LSTM) network for further time-series feature learning.

[0055] (3) Long Short-Term Memory Network (LSTM) sequential learning.

[0056] To further capture long-term dependencies in spot market price series, this application employs a Long Short-Term Memory (LSTM) network, which excels at learning time-series information from data. Compared to recurrent neural networks, LSTM addresses the gradient explosion problem by adding gate structures and cell states. Gate structures control the inflow and outflow of information within the network, while cell states enable long-term storage of effective information. Figure 4 The diagram shown is a schematic of the LSTM's circulating unit (cell) structure in this embodiment.

[0057] As an example, the gate structure of LSTM includes forget gates, input gates, and output gates, specifically: 1) The Gate of Oblivion.

[0058] The forget gate determines the cell state at the previous moment. Which information is retained in the current cell state? .

[0059] The input to the forget gate includes the previous hidden state. and the input at the current moment The output is Defined as:

[0060] in, It is the sigmoid function. It is a weight matrix. It is a bias.

[0061] 2) Input gate.

[0062] The input gate determines the input at the current moment. What information is stored in the cell state? .

[0063] The input to the input gate includes the previous hidden state. and the input at the current moment The output includes the values ​​of the memory gates. and candidate memory states of cells The computational logic of the input gate is defined as follows:

[0064] in, and Both represent weight matrices. and Both indicate bias.

[0065] By using the input gate and the forget gate together, the cell state at the current moment can be calculated. The calculation process is as follows: in, This represents element-wise product.

[0066] 3) Output gate.

[0067] Output gate controls cell state Which information will be output as the current state?

[0068] The input to the output gate includes the previous hidden state. Current input and the current cell state The output includes the current output value. and hidden state The calculation logic of the output gate is defined as follows:

[0069] in, It is a weight matrix. It's a bias, and the function of tanh is to... Value scaled to Interval.

[0070] Through the combined action of the forget gate, input gate, and output gate, LSTM can retain long-term effective information in time series and suppress irrelevant information, thereby improving its ability to learn the long-term dependencies of spot market price series.

[0071] (4) Fully connected layer.

[0072] After the Long Short-Term Memory (LSTM) network completes temporal feature learning, the features output by the LSTM network are input into a fully connected layer to map from the high-dimensional feature space to the spot market price prediction result. The fully connected layer, through high-dimensional feature compression, nonlinear enhancement, overfitting suppression, and target mapping, transforms the local high-dimensional features extracted by the LSTM into the final prediction result. In this embodiment, the fully connected layer includes a first linear layer, an activation function layer, a Dropout layer, and a second linear layer. The first linear layer integrates the features output by the LSTM and adjusts the feature dimensions to ensure the feature vectors meet the requirements of subsequent mapping calculations. A ReLU activation function is introduced after the linear layer to increase the network's nonlinear expressive power, thereby improving the model's ability to fit complex spot market price changes. A Dropout layer is set after the activation function to randomly deactivate some neurons during training, reducing the neural network's over-reliance on training data, thus enhancing the model's robustness and suppressing overfitting. The features processed by Dropout are input into the second linear layer. The second linear layer performs the final target mapping, directly mapping the features extracted by the neural network to the predicted spot market price value. In this embodiment, the output is the predicted spot market price for the next 7 days, with one prediction point output every 15 minutes, for a total of 96 time points per day. Therefore, the output of the fully connected layer is the prediction result for multiple time steps, used to represent the spot market clearing price at multiple future moments. Through the above fully connected layer structure, while ensuring feature representation capabilities, the generalization ability of the model is improved, enabling the prediction model to be better applied to spot market price prediction scenarios.

[0073] The technical effects of this application are verified as follows: The spot market price forecasting model was built using the PyTorch framework. Two models with identical structures and parameters were trained for two forecasting objectives: day-ahead clearing electricity prices and real-time clearing electricity prices. The core difference between the two models lies only in the distinction between the multi-source datasets (corresponding to historical data from the day-ahead market and the real-time market, respectively). During the offline training phase, the model training parameters were set as follows: The number of iterations (epochs) is 1000, the learning rate is 0.0001, and the loss function is defined as:

[0074] To more accurately evaluate the generalization ability of the spot market price forecasting model and obtain reliable evaluation results, a 2-fold cross-validation method was used to divide the multi-source dataset into training and testing data. The testing data contained 10 randomly selected electricity price samples. The main technical indicator required an average accuracy of 85% or higher in predicting the daily 24-hour electricity price for day D+2. The formula for calculating the average accuracy is as follows:

[0075] in, For the number of samples, For predicted values, This is the actual value.

[0076] The prediction results of the spot market price forecasting model are shown in Table 1, mainly predicting the average accuracy of the daily 24-hour electricity price for D+2. Specifically, the average accuracy of the day-ahead clearing electricity price forecast is 88.24%, and the average accuracy of the real-time clearing electricity price forecast is 85.34%, both exceeding the target of 85.00%.

[0077] Table 1. Prediction Results of the Spot Market Price Forecasting Model

[0078] The visualizations of the day-ahead clearing price and real-time clearing price predictions are shown in Figures 5 and 6, respectively. Figures 5(a) to 5(j) illustrate the day-ahead clearing price prediction results for 10 test samples, while Figures 6(a) to 6(j) illustrate the real-time clearing price prediction results for 10 test samples. In Figures 5 and 6, the red curve represents the model prediction value, and the green curve represents the actual value.

[0079] like Figure 7 The diagram shown is a schematic representation of the spot market price forecasting system of this application, which may include: The data module is used to acquire multi-source historical data of the power system and construct a multi-source input dataset for spot market price prediction based on the multi-source historical data. The decomposition module is used to perform variational mode decomposition processing on historical cleared electricity price data, decomposing the historical cleared electricity price data into multiple modal component sequences; The combination module is used to combine the modal component sequence with the multi-source input dataset to obtain an enhanced feature dataset; The prediction module is used to input the enhanced feature dataset into the spot market price prediction model to obtain the spot market clearing electricity price prediction results at multiple future time points. The spot market price prediction model is equipped with a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local features from the multi-source input dataset and modal component sequences, and the long short-term memory network is used to learn the temporal dependency relationship of the feature vector output by the convolutional neural network.

[0080] Specific limitations regarding the spot market price forecasting system can be found in the limitations of the spot market price forecasting method described above; the corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned spot market price forecasting system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0081] This application also provides an electronic device, which may include one or more processors, memory and communication interfaces.

[0082] The memory, communication interface, and processor are coupled together. For example, the memory, communication interface, and processor can be coupled together via a bus.

[0083] The communication interface is used for data transmission with other devices. The memory stores computer program code. This computer program code includes computer instructions, which, when executed by the processor, cause the electronic device to perform the steps of the aforementioned spot market price prediction method.

[0084] The processor can be a processor or controller, such as a Central Processing Unit (CPU), a general-purpose processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The processor can be used to support an electronic device in performing the method steps provided in the above embodiments.

[0085] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. These buses can be categorized as address buses, data buses, control buses, etc.

[0086] This application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described spot market price prediction method.

[0087] The computer-readable storage media involved in this application include random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage media known in the art.

[0088] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0089] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for predicting spot market prices, characterized in that, include: Acquire multi-source historical data of the power system, and construct a multi-source input dataset for spot market price forecasting based on the multi-source historical data; Variational mode decomposition is performed on the historical cleared electricity price data to decompose the historical cleared electricity price data into multiple modal component sequences; The modal component sequence is combined with the multi-source input dataset to obtain an enhanced feature dataset; The enhanced feature dataset is input into the spot market price prediction model to obtain the spot market clearing electricity price prediction results at multiple future time points. The spot market price prediction model is equipped with a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local features from the multi-source input dataset and modal component sequences, and the long short-term memory network is used to learn the temporal dependency relationship of the feature vector output by the convolutional neural network.

2. The spot market price forecasting method according to claim 1, characterized in that, The multi-source historical data includes provincial dispatch load data, photovoltaic power data, wind power power data, hydropower output data, inter-provincial interconnection line output data, and non-market-based unit output data.

3. The spot market price forecasting method according to claim 1, characterized in that, The variational mode decomposition process includes: performing frequency domain decomposition on historical clearing electricity price data, and obtaining multiple independent mode component sequences through iterative optimization.

4. The spot market price forecasting method according to claim 1, characterized in that, The convolutional neural network is a one-dimensional convolutional neural network, including two convolutional blocks. Each convolutional block includes a convolutional layer, a ReLU activation function, and a pooling layer. The convolutional layer is used to perform convolution operations on the enhanced feature dataset. The ReLU activation function is used to introduce non-linearity to learn complex features. The pooling layer is used to downsample the convolution operation results.

5. The spot market price forecasting method according to claim 1, characterized in that, When the Long Short-Term Memory network learns the temporal dependencies of the feature vectors output by the convolutional neural network, it calculates the hidden state at the current time step based on the feature vectors output by the current convolutional neural network and the hidden state at the previous time step.

6. The spot market price forecasting method according to claim 5, characterized in that, The long short-term memory network includes a forget gate, an input gate, and an output gate; The forget gate is used to calculate the forgetting coefficient based on the feature vector output by the current convolutional neural network and the hidden state at the previous time step, and to update the cell state at the previous time step based on the forgetting coefficient. The input gate is used to calculate input coefficients based on the feature vector output by the current convolutional neural network and the hidden state at the previous time step, and to generate candidate cell states, which are then combined with the updated cell states to obtain the current cell state. The output gate is used to calculate the output coefficients based on the current cell state and the feature vector output by the current convolutional neural network, and to obtain the hidden state at the current time based on the output coefficients.

7. The spot market price forecasting method according to claim 1, characterized in that, The spot market price forecasting model also includes a fully connected layer; The fully connected layer is used to linearly map the feature vector output by the long short-term memory network to obtain the clearing price prediction values ​​at multiple time points.

8. A spot market price forecasting system, characterized in that, include: The data module is used to acquire multi-source historical data of the power system and construct a multi-source input dataset for spot market price prediction based on the multi-source historical data. The decomposition module is used to perform variational mode decomposition processing on historical cleared electricity price data, decomposing the historical cleared electricity price data into multiple modal component sequences; The combination module is used to combine the modal component sequence with the multi-source input dataset to obtain an enhanced feature dataset; The prediction module is used to input the enhanced feature dataset into the spot market price prediction model to obtain the spot market clearing electricity price prediction results at multiple future time points. The spot market price prediction model is equipped with a convolutional neural network and a long short-term memory network. The convolutional neural network is used to extract local features from the multi-source input dataset and modal component sequences, and the long short-term memory network is used to learn the temporal dependency relationship of the feature vector output by the convolutional neural network.

9. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by the processor, implements the spot market price forecasting method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 7.

Images