Electric power spot transaction volume prediction method

Abstract

The invention relates to the technical field of electric power transaction, in particular to an electric power spot transaction volume prediction method. The method comprises the following steps: multi-source data acquisition: acquiring historical transaction data, real-time market data, meteorological data, user side response data and power grid real-time state data of an electric power market, the historical transaction data comprises historical transaction volume, historical electricity price and market participant quotation information, the real-time market data comprises market supply and demand states and power transmission line capacity in the current time period, and the meteorological data comprises temperature, humidity, wind speed and extreme weather early warning information; the user side response data comprises a user power consumption behavior mode and a demand elastic coefficient, and the power grid real-time state data comprises a node load rate and an equipment operation state; according to the invention, the problem of large prediction deviation of a traditional method can be effectively solved, and the accuracy and market adaptability of electric power spot transaction volume prediction are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power trading, and in particular to a method for predicting power spot trading volume. Background Art

[0002] In the spot trading sector of the electricity market, accurately forecasting trading volumes is crucial for ensuring stable market operation and improving resource allocation efficiency. Existing spot trading volume forecasting technologies suffer from significant flaws: most rely solely on historical trading data and simple meteorological parameters, using a single machine learning model for prediction, which fails to fully capture the complex dynamics of the electricity market. In actual trading, the electricity market is influenced by multiple factors, including the real-time operating status of the power grid, changes in user electricity consumption behavior, and intermittent output from renewable energy sources. For example, operational constraints such as grid transmission line congestion and node overloads directly limit the actual volume of electricity trading, but existing methods fail to account for these key constraints. Furthermore, the impact of dynamic factors such as user demand elasticity and market participant strategic competition on trading volume cannot be accurately quantified due to traditional fixed-weight allocation mechanisms. This results in significant deviations between forecasts and actual trading results, failing to provide reliable basis for market participants to make trading decisions, and severely hindering the scientific management and efficient operation of the electricity market.

[0003] Based on the above problems, there is an urgent need for a technical solution for predicting electricity spot trading volume that can integrate multi-source dynamic data, quantify grid operation constraints, and adaptively adjust data weights to improve prediction accuracy and market adaptability. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and propose a method for predicting the spot trading volume of electricity, comprising the following steps: Multi-source data collection: Acquire historical transaction data of the power market, real-time market data, meteorological data, user-side response data and real-time status data of the power grid, The historical transaction data includes historical transaction volume, historical electricity prices, and quotation information of market participants; the real-time market data includes the market supply and demand status and transmission line capacity of the current period; the meteorological data includes temperature, humidity, wind speed and extreme weather warning information; the user-side response data includes user electricity consumption behavior patterns and demand elasticity coefficients; the real-time grid status data includes node load rate and equipment operating status; Feature Engineering: Standardize the collected multi-source data and perform feature selection using a hybrid model of an improved discrete particle swarm optimization algorithm and support vector regression to extract a feature subset that is strongly correlated with electricity spot trading volume. This feature subset includes meteorological influencing factors, user elasticity coefficients, market price volatility indexes, and grid load correction terms. Hybrid model training: Build a hybrid forecasting model that includes a long short-term memory network and a graph neural network. The long short-term memory network is used to capture the temporal features in time series data, and the graph neural network is used to model the interactive relationships between market participants. The model parameters are optimized through a bidirectional propagation algorithm to form an initial forecasting model. Real-time correction and prediction: Current market data is collected in real time, grid load correction items are calculated and input into the initial prediction model, and the influence weights of each data source are adjusted in combination with a dynamic weight allocation mechanism to output the forecast results of electricity spot trading volume in the future T period. The dynamic weight allocation mechanism adaptively adjusts the weight coefficients of meteorological data, user-side response data and grid status data according to market status and time window.

[0005] Preferably, in the multi-source data collection step, the meteorological data is obtained through the fusion of meteorological satellites, ground monitoring stations and numerical weather forecast models, and the user-side response data is constructed through smart meters, user electricity consumption behavior questionnaires and demand response event historical records; the historical transaction data is obtained from the power trading center database, real-time market data is captured in real time through the power market information interaction platform, and real-time power grid status data is collected through the power dispatching automation system.

[0006] Further preferably, in the feature engineering construction step, the standardization processing includes Z-score standardization of numerical data and one-hot encoding of categorical data; the improved discrete particle swarm optimization algorithm introduces a differential evolution operator and a dynamic inertia weight adjustment strategy to enhance the global search capability and avoid falling into local optimality; when performing feature selection, the prediction error minimization is used as the objective function, and the prediction performance of different feature combinations is evaluated by cross-validation.

[0007] Further preferably, in the hybrid model training step, the graph neural network adopts an attention mechanism to capture the dynamic interactive relationship between market participants, and the weight calculation of the attention mechanism is based on the participants' historical trading behavior, market share and quotation strategy similarity; the long short-term memory network sets multiple hidden layers and uses Dropout technology to prevent overfitting; during the model training process, an adaptive learning rate adjustment algorithm is adopted to dynamically adjust the learning rate according to the training error.

[0008] Further preferably, in the real-time correction and prediction step, the grid load correction term is calculated using the following formula: ; in, is the grid load correction factor, is the current node load rate, is the node rated load rate, Real-time power flow for transmission lines, Due to the capacity limitation of transmission lines, and is the weight parameter obtained by training based on historical data.

[0009] Further preferably, the dynamic weight allocation mechanism is implemented by the following formula: ; in, For the Class data source in The weight of the moment, For the Pearson correlation coefficient between class data source and transaction volume, To control the adjustment parameter of weight sensitivity, is the total number of data source categories.

[0010] Further preferably, the output result of the hybrid prediction model is corrected by the following formula: ; in, For the The revised forecast trading volume at the time, is the number of features in the feature subset, For the The LSTM or GNN model output corresponding to the feature, is the grid load correction factor, For the Dynamic weights of features.

[0011] Further preferably, a model updating step is included: regularly collecting new market data, and using an online learning algorithm to perform incremental training on the hybrid prediction model, wherein the online learning algorithm includes stochastic gradient descent and adaptive moment estimation to update the model parameters in real time; during the model updating process, an update threshold is set, and when the difference between the new data and the historical data exceeds the threshold, the model update is triggered.

[0012] Further preferably, the method also includes a prediction result evaluation step: using root mean square error, mean absolute percentage error and Sharpe ratio to perform multi-dimensional evaluation of the prediction results, and the Sharpe ratio is used to measure the risk-adjusted return of the prediction results; based on the evaluation results, the prediction model is optimized and adjusted, and when the evaluation indicators do not meet the preset standards, feature selection and model training are re-performed.

[0013] Further preferably, the method is applied to an electricity market transaction decision support system, which includes a data acquisition module, a feature engineering module, a model training module, a real-time correction module and a result visualization module. The modules interact with each other through message queues to achieve low-latency prediction in high-concurrency scenarios.

[0014] Technical Effect: This invention addresses the problem that traditional technologies fail to comprehensively consider the operational constraints of the power grid and the dynamic changes in the market, and achieves a technological breakthrough through multi-source data fusion, dynamic weight allocation, and hybrid model construction. On the one hand, data such as the power grid load rate and transmission line flow are incorporated into the prediction system, and the impact of power grid constraints is quantified through a revised formula; on the other hand, weights are dynamically adjusted based on data relevance, and the advantages of LSTM and GNN are combined to mine the time series and market subject interaction characteristics. This effectively solves the problem of large prediction deviations in traditional methods, significantly improves the accuracy and market adaptability of electricity spot trading volume predictions, and provides a reliable basis for power market trading decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the method for predicting electricity spot trading volume in this application. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] See also Figure 1 Traditional electricity spot trading volume forecasting technology has the following technical problems: it relies too much on a single data source or limited types of data, making it difficult to fully cover the complex factors affecting trading volume; the feature selection lacks an efficient optimization mechanism, resulting in redundant model input or omission of key features; the use of a single forecasting model cannot take into account the time series characteristics and the interaction between market entities; the lack of a real-time correction mechanism makes it difficult to cope with dynamic market changes.

[0018] Based on this, this embodiment provides a method for predicting electricity spot trading volume, including the following steps: a multi-source data collection step obtains historical trading data, real-time market data, meteorological data, user-side response data and real-time power grid status data of the electricity market, and defines in detail the specific information covered by each type of data; a feature engineering construction step standardizes the collected multi-source data, uses a hybrid model of an improved discrete particle swarm optimization algorithm (MDPSO) and a support vector regression (SVR) to perform feature selection, and extracts a feature subset that is strongly correlated with electricity spot trading volume; a hybrid model training step constructs a hybrid prediction model including a long short-term memory network (LSTM) and a graph neural network (GNN), respectively leveraging the advantages of LSTM in capturing time series features and GNN in modeling the interactive relationship between market participants, and optimizing model parameters through a two-way propagation algorithm; a real-time correction and prediction step collects current market data in real time, calculates the power grid load correction term and inputs it into the initial prediction model, adjusts the influence weight of each data source in combination with a dynamic weight allocation mechanism, and outputs the electricity spot trading volume prediction result for the future T period.

[0019] This solution makes up for the single data dimension defect of traditional methods through comprehensive collection of multi-source data, ensuring the integrity of input information; uses hybrid algorithms for feature selection to effectively screen out key features and improve model training efficiency and accuracy; the combination of LSTM and GNN realizes dual modeling of time series and market entity relationships, breaking through the limitations of a single model; real-time correction and dynamic weight adjustment mechanism can quickly respond to market changes and ensure that the prediction results are in line with actual conditions.

[0020] The technical effects achieved by the above embodiments include: ensuring the input quality of the prediction model from the source of the data, laying the foundation for accurate prediction; the optimized feature selection process avoids redundant data interference, allowing the model to focus on core influencing factors; the hybrid model architecture deeply mines market data characteristics and accurately depicts the changes in trading volume; the real-time correction mechanism gives the model dynamic adaptability, significantly improving the timeliness and reliability of the prediction, and providing strong support for power market trading decisions.

[0021] Traditional power data collection has the following technical problems: data acquisition channels are scattered and lack systematic integration, resulting in insufficient data integrity; some key data come from a single source, making data accuracy and timeliness difficult to guarantee; and different types of data collection methods are not standardized, affecting subsequent data processing and analysis.

[0022] Based on this, in the multi-source data collection step, meteorological data is acquired through the fusion of meteorological satellites, ground monitoring stations and numerical weather forecast models, integrating space-based, ground-based and model prediction data to ensure the comprehensiveness and accuracy of meteorological information; user-side response data is constructed through smart meters, user electricity consumption behavior questionnaires and demand response event historical records, collecting user electricity consumption information in multiple dimensions, from real-time electricity consumption data, subjective behavior preferences to historical response events, to fully present user-side characteristics; historical transaction data is obtained from the power trading center database to ensure the authority and integrity of the data; real-time market data is captured in real time through the power market information interaction platform to achieve instant data updates; real-time power grid status data is collected through the power dispatching automation system (SCADA) to ensure the accuracy and real-time nature of power grid operation data.

[0023] This solution addresses the fragmented, single, and non-standard nature of traditional data collection by clarifying precise channels and methods for collecting various types of data. The integrated acquisition of meteorological data avoids information loss caused by blind spots in weather monitoring from a single data source, enabling comprehensive capture of the impact of meteorological changes on electricity demand. The multi-channel construction of user-side response data overcomes the limitations of relying solely on smart meter data and allows for in-depth exploration of user behavior. Standardized data collection sources ensure data quality and reliability, providing a solid foundation for subsequent feature engineering and model training.

[0024] The technical effects achieved by the above embodiments include: building a complete, accurate, and real-time multi-source data acquisition system to reduce data errors and omissions; providing high-quality input data for the electricity spot trading volume prediction model, improving the model's ability to characterize complex market factors; enabling data to truly reflect the operating status of the electricity market and changes in user demand, thereby enhancing the credibility and practicality of the prediction results.

[0025] Traditional feature engineering has the following technical problems: the data standardization processing method is single and cannot effectively adapt to the characteristics of different types of data; the feature selection algorithm easily falls into local optimality, making it difficult to find the globally optimal feature combination; there is a lack of scientific evaluation mechanism, and it is impossible to accurately judge the impact of feature combinations on prediction performance.

[0026] Based on this, during the feature engineering step, standardization is performed on numeric data using the Z-score function to eliminate the effects of data dimension and conform the data to a standard normal distribution, facilitating model processing. Categorical data is one-hot encoded to convert categorical information into a machine-readable vector form. An improved discrete particle swarm optimization (MDPSO) algorithm incorporates a differential evolution operator and a dynamic inertia weight adjustment strategy. The differential evolution operator enhances the algorithm's global search capabilities, while the dynamic inertia weight adjusts the search step size based on the iterative process, preventing the algorithm from falling into local optima. During feature selection, minimizing prediction error is used as the objective function, and cross-validation is used to evaluate the predictive performance of different feature combinations. Multiple training and test set partitioning is performed to comprehensively assess the generalization capabilities of the feature combinations.

[0027] Through diversified standardization processing, this solution adapts to the characteristics of different data types and improves the effectiveness of data processing; the improved MDPSO algorithm effectively overcomes the shortcomings of traditional algorithms and can screen out the most representative feature subsets from massive features; based on the cross-validation evaluation mechanism, it scientifically judges the pros and cons of feature combinations and avoids the degradation of model performance due to improper feature selection.

[0028] The technical effects achieved by the above embodiments include: optimized data preprocessing process, improving data quality and model training efficiency; accurate feature selection process, reducing redundant feature interference and reducing model complexity; the optimal feature combination found through scientific evaluation significantly improves the prediction accuracy and generalization ability of the model, so that the model can stably and accurately predict the spot trading volume of electricity in different scenarios.

[0029] Traditional model training has the following technical problems: a single neural network model is difficult to simultaneously process time series characteristics and complex market entity relationships; the model structure design is unreasonable and prone to overfitting; the learning rate is fixed during training and cannot be dynamically adjusted according to the training situation, affecting the model convergence speed and performance.

[0030] Based on this, a hybrid model training step employs a combined architecture of a graph neural network (GNN) and a long short-term memory network (LSTM). The GNN employs an attention mechanism to capture the dynamic interactions between market participants. By calculating attention weights based on participants' historical trading behavior, market share, and similarity in their pricing strategies, it accurately depicts the competitive and collaborative relationships between market players. The LSTM employs multiple hidden layers to enhance the model's ability to extract features from time series data, and employs a dropout technique to randomly discard neurons to prevent overfitting.

[0031] During the model training process, an adaptive learning rate adjustment algorithm is used to dynamically adjust the learning rate according to the training error. A larger learning rate is used in the early stage of training to speed up the convergence speed, and the learning rate is reduced when approaching the optimal solution to improve the convergence accuracy.

[0032] This solution overcomes the limitations of traditional single models by leveraging the complementary advantages of GNN and LSTM. LSTM focuses on mining long-term dependencies in time series, while GNN focuses on modeling the interaction between market players. The combination of the two comprehensively captures the characteristics of power market data. Reasonable model structure design and the application of Dropout technology effectively avoid overfitting and improve the generalization ability of the model. The adaptive learning rate adjustment algorithm makes model training more efficient and can quickly find the optimal parameters.

[0033] The technical effects achieved by the above embodiments include: the constructed hybrid model can deeply explore the time series patterns and subject interaction patterns in the power market data, and accurately predict the trend of changes in trading volume; the optimized model structure and training strategy improve the stability and reliability of the model, and reduce the prediction deviation caused by data fluctuations; the adaptive learning rate accelerates the model training speed, reduces the training cost, and enables the model to quickly adapt to changes in market data and update the prediction capability in a timely manner.

[0034] Traditional electricity spot trading volume forecasting has the following technical problems when considering the impact of grid status: 1. The impact of the actual grid operation constraints on trading volume is ignored, resulting in a disconnect between the forecast results and actual market trading conditions; 2. There is a lack of effective methods to quantify the impact of grid load on trading volume, making it impossible to accurately correct the forecast results.

[0035] Based on this, in the real-time correction and prediction steps, the formula is: ; Calculate the grid load correction term, where is the grid load correction factor, is the current node load rate, is the node rated load rate, Real-time power flow for transmission lines, Transmission line capacity limitations, and is the weight parameter obtained by training based on historical data.

[0036] In the prediction of electricity spot trading volume, the actual operating status of the power grid has an important impact on the transaction, but traditional methods often ignore this key factor. This formula is used to calculate the power grid load correction factor , aims to quantify the impact of power grid operation status on transaction volume forecast results, so that the forecast is more in line with the actual transaction situation. In the formula, Represents the current node load rate, which reflects the actual load level of each node in the power system at a certain moment and is a key indicator for measuring the local operating pressure of the power grid; is the node rated load rate, that is, the maximum load ratio that the node is designed to carry, This ratio reflects the relative pressure level of the node load. The larger the ratio, the closer the node is to full load operation, and the stronger the restrictive effect on electricity trading volume may be. Indicates the real-time power flow of the transmission line, which describes the actual power transmitted on the current transmission line and directly reflects the operating load of the line; is the capacity limit of the transmission line, that is, the maximum power limit that the line can safely transmit. It reflects the utilization rate of transmission lines. When the ratio is close to 1, it means that the line is close to the transmission limit, and the normal development of power trading may be affected by problems such as congestion. and The weight parameters are obtained through training based on historical data. Their function is to balance the influence of node load rate and transmission line flow on the grid load correction coefficient. Through training and optimization of a large amount of historical data, the optimal values of these two parameters are determined, so that the formula can more accurately reflect the actual impact of grid load on transaction volume. For example, in some areas, the problem of transmission line congestion is more prominent. At this time, through training, The value of may be relatively large to highlight the impact of transmission line flows on the correction factor. This formula comprehensively considers two key dimensions of grid operation—node load and line transmission. Through quantitative calculation, it converts grid status into coefficients that can be used to correct forecast results. This fills the gap in traditional forecasting that does not consider grid constraints and provides key support for improving forecast accuracy.

[0037] This formula comprehensively considers two key grid operating indicators: node load and transmission line flow. Using historical data training to determine weighting parameters, it quantifies the impact of grid load on trading volume. By introducing a grid load correction term, this solution incorporates the real-time grid operating status into the prediction model, addressing the gaps in traditional methods in considering grid constraints.

[0038] By using quantitative formulas to accurately calculate the correction coefficient, the forecast results can be dynamically adjusted according to the actual load conditions of the power grid, making the forecast more consistent with actual market transactions. The technical effects achieved by the above embodiments include: enhancing the adaptability of the forecast model to the operating status of the power grid, avoiding forecast deviations caused by grid congestion, overload, etc.; improving the accuracy and reliability of the forecast results through quantitative correction, providing more valuable reference volume forecast information for both parties in the power market; enabling the forecast model to better reflect the relationship between the physical constraints of the power system and market transactions, and improving the scientific nature and stability of the power market operation.

[0039] Traditional data weight allocation has the following technical problems: the fixed weight allocation method cannot adapt to the dynamic changes of the power market and it is difficult to reflect the differences in the importance of different data sources under different market conditions; the lack of scientific weight calculation methods leads to unreasonable data weight settings, affecting the accuracy of prediction results.

[0040] Based on this, this embodiment provides a dynamic weight allocation mechanism through the formula: ; in For the Class data source in The weight of the moment, For the Pearson correlation coefficient between class data source and transaction volume, To control the adjustment parameter of weight sensitivity, The total number of data source categories. This formula calculates weights based on the correlation between data sources and trading volume, amplifies correlation differences through an exponential function, and uses normalization to ensure that the total weight is 1, achieving dynamic and adaptive weight allocation.

[0041] In the electricity market, the impact of different data sources on spot electricity trading volume varies with market conditions and time. Traditional fixed weight allocation methods cannot adapt to this dynamic nature. This formula implements a dynamic weight allocation mechanism that adjusts the weight of each data source in the forecast in real time based on its correlation with trading volume, thereby optimizing the forecast model's adaptability to market changes.

[0042] In the formula, Indicates the Class data source in The weight at a given moment is a quantitative reflection of the influence of the data source on the prediction result at the current moment. Representative Class data source in Time data, such as It can be the first Real-time meteorological data, user-side response data, etc. For the The actual electricity spot trading volume at the moment.

[0043] It is The Pearson correlation coefficient between a data source and trading volume measures the degree of linear correlation between the two. Its value ranges from -1 to 1, with the closer the absolute value is to 1, the stronger the correlation. This coefficient is calculated by calculating the covariance and standard deviation of the data source and trading volume data, and objectively reflects the impact of the data source on trading volume.

[0044] The adjustment parameter that controls the sensitivity of weights is used to amplify or reduce the impact of correlation differences on weights. When the value is large, the weight of data sources with high relevance will be significantly increased, while the weight of data sources with low relevance will be greatly reduced, making the model pay more attention to key data sources; on the contrary, when When the value is small, the weight adjustment is relatively smooth and the weight difference between each data source is reduced.

[0045] Denominator Normalize the molecules to ensure that all weights are The sum of the two is 1, ensuring the rationality and effectiveness of weight distribution. Through this dynamic weight calculation method based on correlation, the weight of data sources closely related to trading volume is automatically increased in different market scenarios, such as extreme weather and holidays, thereby improving the forecast model's responsiveness to market dynamics and forecast accuracy.

[0046] This solution breaks the limitations of traditional fixed weights through a dynamic weight allocation mechanism, and can adjust the weights of each data source in real time according to market conditions and time windows. In different market scenarios, such as extreme weather and major holidays, the weights of data sources with high correlation with transaction volume (such as meteorological data and user-side response data) are automatically increased to highlight their importance to the forecast. The technical effects achieved by the above embodiments include: enabling the forecast model to dynamically adapt to the complex and changing environment of the electricity market and improving the model's responsiveness to different market conditions; a reasonable weight calculation method ensures that the weight of each data source matches its actual influence, optimizes data utilization efficiency, and improves the accuracy of forecast results; and through dynamic weight adjustment, enhances the model's adaptability to abnormal market fluctuations and emergencies, making the forecast results more reliable and practical.

[0047] Traditional electricity spot trading volume forecasting has the following technical problems: there is a lack of a comprehensive correction mechanism for the forecast results, which cannot effectively integrate the impact of grid load and changes in the weight of multi-source data on the forecast results; the forecast results do not fully consider the complex constraints and dynamic factors in actual market operations, resulting in large forecast deviations.

[0048] Based on this, this embodiment provides an output result of a hybrid prediction model through the formula: ; Make corrections, including For the The revised forecast trading volume at the time, is the number of features in the feature subset, For the The LSTM or GNN model output corresponding to the feature, is the grid load correction factor, For the Dynamic weights of features.

[0049] Traditional electricity spot trading volume forecasts lack comprehensive consideration of the impact of grid load and dynamic changes in data weights, resulting in deviations between the forecast results and actual conditions.

[0050] This formula is used to correct the output of the hybrid forecasting model, integrating the grid load correction term and the dynamic weight distribution results to make the forecast more consistent with the actual operation of the power market. For the The corrected predicted trading volume at the moment is the final output prediction result.

[0051] is the grid load correction coefficient mentioned above, which reflects the impact of the real-time operation status of the grid on the transaction volume. Used to adjust the basic prediction results. When the grid load is high, As the value increases, the predicted transaction volume is adjusted accordingly, reflecting the restrictive effect of power grid constraints on transactions.

[0052] It is the number of features in the feature subset, that is, the number of key features used to input the prediction model after feature engineering screening. It is The feature in The dynamic weight of a moment is calculated by the dynamic weight allocation mechanism formula, which determines the importance of the feature in the prediction at the current moment.

[0053] For the The LSTM or GNN model output corresponding to each feature. The LSTM model is good at processing time series data and capturing the temporal changes in trading volume; the GNN model focuses on modeling the interaction between market participants and exploring market dynamics. The outputs of the two are weighted and summed with dynamic weights to obtain the basic prediction value. .

[0054] This formula combines the grid load correction term with the dynamic weight allocation result, and comprehensively adjusts the prediction results from two key dimensions: grid operating status and data importance. It fully considers the complex factors affecting the spot trading volume of electricity, significantly improves the accuracy and reliability of the prediction, and provides more valuable reference volume forecast information for electricity market participants, helping them to formulate scientific and reasonable trading strategies.

[0055] This formula combines the grid load correction term with the dynamic weight allocation result, and comprehensively corrects the output of the basic forecast model, comprehensively considering the impact of the grid operation status and data weight changes on the transaction volume. Through the comprehensive correction formula, this solution constructs a complete forecast result adjustment system, which makes up for the defect of the traditional forecast lacking a correction mechanism. The dynamic adjustment of the grid load and data weight is incorporated into the forecast result calculation to make the forecast more consistent with the actual market operation. The technical effects achieved by the above embodiment include: significantly improving the accuracy of the electricity spot transaction volume forecast, reducing the forecast error caused by not considering the grid constraints and dynamic changes in data; through comprehensive correction, enhancing the forecast model's ability to handle complex market factors, so that the forecast results are closer to the actual transaction situation; providing more reliable transaction volume forecast information for electricity market participants, assisting them in formulating scientific and reasonable trading strategies, and improving the efficiency of electricity market operation and the rationality of resource allocation.

[0056] Traditional model updates have the following technical problems: the use of offline training methods cannot respond to market data changes in a timely manner, resulting in the model's predictive ability declining over time; the lack of an effective mechanism to trigger model updates makes it difficult to determine when to optimize the model, and the model is prone to being too outdated or over-updated.

[0057] Based on this, the method also includes a model update step: regularly collecting new market data and incrementally training the hybrid forecasting model using online learning algorithms, including stochastic gradient descent (SGD) and adaptive moment estimation (Adam), to update model parameters in real time. During the model update process, an update threshold is set, and when the difference between new data and historical data exceeds the threshold, the model update is triggered. This step continuously optimizes the model through online learning and utilizes efficient optimization algorithms to quickly update parameters. Setting the update threshold provides a scientific basis for update decisions, balancing the timeliness and stability of model updates.

[0058] This solution addresses the issues of offline training and blind updates in traditional models through online learning and an update threshold mechanism. Online learning enables the model to continuously absorb new data and information, maintaining its adaptability to market changes. The update threshold prevents model instability caused by frequent updates due to minor data fluctuations, while ensuring timely optimization when significant market changes occur.

[0059] The technical effects achieved by the above embodiments include: significantly improving the long-term prediction performance of the model, enabling it to continuously and accurately predict the spot trading volume of electricity; a scientific update mechanism reduces the cost of model maintenance and avoids invalid or over-training; enhancing the model's adaptability to the dynamic evolution of the electricity market, and maintaining a high level of prediction accuracy in changing scenarios such as market rule adjustments and new energy access, providing stable and reliable prediction support for electricity market transactions.

[0060] Traditional prediction result evaluation has the following technical problems: the evaluation indicators are single and only focus on prediction errors, which cannot comprehensively measure the quality of the prediction results; there is a lack of a model optimization closed loop based on the evaluation results, resulting in the evaluation results being unable to be effectively fed back into the model improvement, making it difficult to continuously improve the prediction performance.

[0061] Based on this, the method also includes a prediction result evaluation step: using the root mean square error (RMSE), mean absolute percentage error (MAPE) and Sharpe ratio (SharpeRatio) to conduct a multi-dimensional evaluation of the prediction results. The Sharpe ratio is used to measure the risk-adjusted return of the prediction results; according to the evaluation results, the prediction model is optimized and adjusted. When the evaluation indicators do not meet the preset standards, feature selection and model training are re-performed.

[0062] This step analyzes the prediction results from different perspectives using a variety of evaluation indicators. RMSE measures the overall deviation between the predicted value and the true value, MAPE reflects the relative error, and the Sharpe ratio assesses the risk-return situation. An optimization and adjustment mechanism based on the evaluation results forms a complete closed loop from evaluation to improvement.

[0063] This solution addresses the limitations of traditional evaluation through a multi-dimensional assessment and closed-loop optimization mechanism. Multiple evaluation metrics comprehensively cover the accuracy of forecast results, the relative magnitude of errors, and the risk-return characteristics, avoiding the one-sidedness of a single metric. Model optimization based on these evaluation results enables continuous improvement based on actual performance, continuously enhancing forecasting performance.

[0064] The technical effects achieved by the above embodiments include: achieving a comprehensive and scientific evaluation of the prediction results, providing an accurate basis for model optimization; continuously improving the model through a closed-loop optimization mechanism to improve prediction accuracy and stability; helping electricity market participants to more accurately understand the quality of prediction results and make reasonable trading decisions based on reliable evaluations, while promoting the continuous evolution and improvement of prediction models in practical applications.

[0065] Traditional power market forecasting applications have the following technical problems: lack of an integrated application system, data processing, model training and forecast result presentation are independent of each other, resulting in low work efficiency; the system architecture cannot meet the real-time requirements in high-concurrency scenarios, making it difficult to process large amounts of data and output forecast results in a short period of time; the forecast result presentation format is single, which is not conducive to users' intuitive understanding and analysis.

[0066] Based on this, the method was applied to a power market trading decision support system. The system consists of a data acquisition module, a feature engineering module, a model training module, a real-time correction module, and a result visualization module. Modules interact with each other via message queues to achieve low-latency predictions in high-concurrency scenarios. The result visualization module uses various charts, such as heat maps and line graphs, to intuitively display prediction results and their influencing factors. The system integrates the entire process of power spot trading volume forecasting through a modular design. Message queues enable efficient data transmission and processing. A variety of visualization formats provide an intuitive presentation of complex data.

[0067] This solution addresses the efficiency, real-time, and presentation challenges of traditional applications by building an integrated decision support system. Its modular architecture and message queuing technology ensure stable operation in high-concurrency scenarios, enabling rapid data processing and forecasting. Rich visualizations facilitate rapid access to key information and in-depth analysis of forecast results and influencing factors.

[0068] The technical effects achieved by the above embodiments include: significantly improving the application efficiency of power market transaction volume forecasting and shortening the time from data collection to result output; meeting the real-time transaction decision-making needs of the power market and providing users with timely and accurate forecast information; intuitive visual display helps users understand market trends and forecast results more clearly, assists them in formulating scientific and reasonable trading strategies, and improves the intelligence and scientific level of power market transactions.

[0069] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for predicting electricity spot trading volume, characterized in that: The following steps are involved: Multi-source data collection: Acquire historical transaction data of the power market, real-time market data, meteorological data, user-side response data and real-time status data of the power grid, The historical transaction data includes historical transaction volume, historical electricity prices, and quotation information of market participants; the real-time market data includes the market supply and demand status and transmission line capacity of the current period; the meteorological data includes temperature, humidity, wind speed and extreme weather warning information; the user-side response data includes user electricity consumption behavior patterns and demand elasticity coefficients; the real-time grid status data includes node load rate and equipment operating status; Feature Engineering: Standardize the collected multi-source data and perform feature selection using a hybrid model of an improved discrete particle swarm optimization algorithm and support vector regression to extract a feature subset that is strongly correlated with electricity spot trading volume. This feature subset includes meteorological influencing factors, user elasticity coefficients, market price volatility indexes, and grid load correction terms. Hybrid model training: Build a hybrid forecasting model that includes a long short-term memory network and a graph neural network. The long short-term memory network is used to capture the temporal features in time series data, and the graph neural network is used to model the interactive relationships between market participants. The model parameters are optimized through a bidirectional propagation algorithm to form an initial forecasting model. Real-time correction and prediction: Current market data is collected in real time, grid load correction items are calculated and input into the initial prediction model, and the influence weights of each data source are adjusted in combination with a dynamic weight allocation mechanism to output the forecast results of electricity spot trading volume in the future T period. The dynamic weight allocation mechanism adaptively adjusts the weight coefficients of meteorological data, user-side response data and grid status data according to market status and time window.

2. A method for predicting electricity spot trading volume according to claim 1, characterized in that: In the multi-source data collection step, the meteorological data is obtained through the integration of meteorological satellites, ground monitoring stations and numerical weather forecast models, and the user-side response data is constructed through smart meters, user electricity consumption behavior questionnaires and demand response event historical records; the historical transaction data is obtained from the power trading center database, real-time market data is captured in real time through the power market information interaction platform, and real-time power grid status data is collected through the power dispatching automation system.

3. The method for predicting electricity spot trading volume according to claim 1, characterized in that: In the feature engineering construction step, the standardization processing includes Z-score standardization of numerical data and one-hot encoding of categorical data; the improved discrete particle swarm optimization algorithm introduces a differential evolution operator and a dynamic inertia weight adjustment strategy to enhance global search capabilities and avoid falling into local optimality; when performing feature selection, the objective function is to minimize the prediction error, and cross-validation is used to evaluate the prediction performance of different feature combinations.

4. The method for predicting electricity spot trading volume according to claim 1, characterized in that: In the hybrid model training step, the graph neural network adopts the attention mechanism to capture the dynamic interactive relationship between market participants. The weight calculation of the attention mechanism is based on the participants' historical trading behavior, market share and quotation strategy similarity. The long short-term memory network is equipped with multiple hidden layers and uses the Dropout technology to prevent overfitting. During the model training process, an adaptive learning rate adjustment algorithm is adopted to dynamically adjust the learning rate according to the training error.

5. The method for predicting electricity spot trading volume according to claim 1, characterized in that: In the real-time correction and prediction step, the grid load correction term is calculated using the following formula: ； in, is the grid load correction factor, is the current node load rate, is the node rated load rate, Real-time power flow for transmission lines, Due to the capacity limitation of transmission lines, and is the weight parameter obtained by training based on historical data.

6. The method for predicting electricity spot trading volume according to claim 1, characterized in that: The dynamic weight allocation mechanism is implemented by the following formula: ； in, For the Class data source in The weight of the moment, For the Pearson correlation coefficient between class data source and transaction volume, To control the adjustment parameter of weight sensitivity, is the total number of data source categories.

7. The method for predicting electricity spot trading volume according to claim 1, characterized in that: The output of the hybrid prediction model is modified by the following formula: ； in, For the The revised forecast trading volume at the time, is the number of features in the feature subset, For the The LSTM or GNN model output corresponding to the feature, is the grid load correction factor, For the Dynamic weights of features.

8. The method for predicting electricity spot trading volume according to claim 1, characterized in that: The method also includes a model updating step: regularly collecting new market data and incrementally training the hybrid forecasting model using an online learning algorithm, which includes stochastic gradient descent and adaptive moment estimation, to update model parameters in real time. During the model updating process, an update threshold is set, and when the difference between the new data and the historical data exceeds the threshold, the model update is triggered.

9. The method for predicting electricity spot trading volume according to claim 1, characterized in that: The method also includes a prediction result evaluation step: using root mean square error, mean absolute percentage error and Sharpe ratio to perform a multi-dimensional evaluation of the prediction results, and the Sharpe ratio is used to measure the risk-adjusted return of the prediction results; based on the evaluation results, the prediction model is optimized and adjusted, and when the evaluation indicators do not meet the preset standards, feature selection and model training are re-performed.

10. The method for predicting electricity spot trading volume according to claim 1, characterized in that: The method is applied to a power market transaction decision support system, which includes a data acquisition module, a feature engineering module, a model training module, a real-time correction module, and a result visualization module. The modules interact with each other through message queues to achieve low-latency prediction in high-concurrency scenarios.

Images