Power spot market supply and demand forecasting method and system fusing multi-source features and dynamic correction

Abstract

The application provides a power spot market supply and demand prediction method and system fusing multi-source features and dynamic correction, and relates to the technical field of power market supply and demand prediction. Multi-source heterogeneous power data including demand side data and power supply side data are acquired and preprocessed; the preprocessed multi-source heterogeneous power data are input into a space-time feature extraction model, space-time correlation features of power supply and demand are extracted based on system topological structure and time series; a supply and demand coupling correlation model is used to perform coupling analysis on the space-time correlation features, quantize interactive influence of the power supply side and the demand side, and output supply and demand collaborative prediction results; a dynamic correction prediction strategy based on a rolling time domain is used to perform real-time rolling correction on the supply and demand collaborative prediction results, and finally obtain the supply and demand prediction results. More reliable and timely decision support is provided for power spot market transactions, which helps to improve market operation efficiency and system safety and stability.

Description

Technical Field

[0001] This invention relates to the field of electricity market supply and demand forecasting technology, and in particular to a method and system for forecasting electricity spot supply and demand that integrates multi-source characteristics and dynamic correction. Background Technology

[0002] Regional electricity spot markets operate on a real-time electricity supply and demand basis, and their stable operation relies on accurate predictions of supply and demand relationships. Supply and demand forecasting, as a crucial link in market regulation and resource optimization, directly impacts electricity price stability and power system security. With the gradual advancement of regional electricity spot market construction, accurately capturing supply and demand patterns and predicting balance trends in advance have become core supports for ensuring efficient market operation and mitigating the risks of supply and demand imbalances.

[0003] Currently, research on supply and demand forecasting in regional electricity spot markets has accumulated a certain practical foundation. However, facing the complex scenarios of multiple intertwined factors and increased dynamism in market operation, existing technologies still struggle to meet the demand for high-precision forecasting. Traditional forecasting schemes have shortcomings in key aspects such as data utilization, feature mining, dynamic adaptation, and mechanism modeling, resulting in significant deviations between forecast results and actual market operations, making it difficult to support refined market regulation.

[0004] The shortcomings of existing technologies are as follows: On the one hand, they rely heavily on single types of data, failing to fully integrate key information such as economic indicators and new energy output. Furthermore, they depend on linear or shallow nonlinear models, making it difficult to capture the spatiotemporal correlation and implicit effects of supply and demand. At the same time, the static model structure is poorly adaptable to scenarios such as extreme weather and sudden public events, resulting in insufficient accuracy in basic forecasts. On the other hand, the core deficiency lies in the lack of a supply-demand coupling mechanism. Existing solutions often conduct power supply and demand forecasts separately, failing to establish a quantitative model of the interaction between the two sides. This makes it impossible to accurately reflect the dynamic correlation between fluctuations on the power supply side and responses on the demand side, resulting in a serious disconnect between the supply-demand balance prediction results and the actual market operation patterns, making it difficult to support the needs of refined market regulation. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a method and system for forecasting electricity spot market supply and demand that integrates multi-source features and dynamic correction. This improves the accuracy and timeliness of regional electricity spot market supply and demand forecasting under complex scenarios such as extreme weather, policy adjustments, and sudden load changes, providing a reliable basis for market transaction decisions, grid dispatch optimization, and resource allocation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction, comprising: Acquire multi-source heterogeneous power data and perform preprocessing. The multi-source heterogeneous power data includes demand-side data and power supply-side data. The preprocessed multi-source heterogeneous power data is input into the spatiotemporal feature extraction model, and the spatiotemporal correlation features of power supply and demand are extracted based on the system topology and time series. A supply-demand coupling correlation model is adopted to perform coupling analysis on the spatiotemporal correlation characteristics, quantify the interaction between the power supply side and the demand side, and output the supply-demand collaborative prediction results. Based on a dynamic correction prediction strategy in the rolling time domain, the supply and demand collaborative prediction results are corrected in real time to obtain the final supply and demand prediction results.

[0007] Secondly, the present invention provides a power spot supply and demand forecasting system that integrates multi-source features and dynamic correction, comprising: The data acquisition module is used to acquire multi-source heterogeneous power data and perform preprocessing. The multi-source heterogeneous power data includes demand-side data and power supply-side data. The feature extraction module is used to input the preprocessed multi-source heterogeneous power data into the spatiotemporal feature extraction model, and extract the spatiotemporal correlation features of power supply and demand based on the system topology and time series. The preliminary prediction module is used to perform coupling analysis on the spatiotemporal correlation characteristics using a supply and demand coupling correlation model, quantify the interaction between the power supply side and the demand side, and output the supply and demand collaborative prediction results. The result correction module is used to perform real-time rolling correction on the supply and demand collaborative prediction results based on a dynamic correction prediction strategy in the rolling time domain, so as to obtain the final supply and demand prediction results.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the electricity spot supply and demand forecasting method that integrates multi-source features and dynamic correction as described in the first aspect.

[0009] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the electricity spot supply and demand forecasting method that integrates multi-source features and dynamic correction as described in the first aspect.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention lays a comprehensive and high-quality data foundation for predictive analysis by comprehensively acquiring and preprocessing multi-source heterogeneous power data from both the demand and supply sides, effectively avoiding the problem of predictive bias caused by a single data source.

[0011] The spatiotemporal feature extraction model based on system topology and time series can accurately capture the spatiotemporal correlation features hidden in power supply and demand data, breaking through the limitations of traditional prediction methods in mining multidimensional features. By performing coupled analysis of spatiotemporal correlation features through the supply and demand coupling correlation model, the quantitative characterization of the interaction between the power supply side and the demand side is realized, making the supply and demand collaborative prediction results more consistent with the actual correlation logic of power system operation.

[0012] By combining a dynamic correction forecasting strategy with a rolling time domain, the initial collaborative forecasting results can be corrected in real time. This can promptly correct deviations in the forecasting process, significantly improve the timeliness and accuracy of the forecasting results, provide reliable support for trading decisions and dispatch optimization in the electricity spot market, help reduce the operating costs of the power system, and improve the efficiency of energy resource allocation and power supply reliability.

[0013] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0015] Figure 1 The main flowchart of a power spot supply and demand forecasting method that integrates multi-source features and dynamic correction is provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction, including the following steps: S1: Acquire multi-source heterogeneous power data and preprocess it, wherein the multi-source heterogeneous power data includes demand-side data and power supply-side data; S2: Input the preprocessed multi-source heterogeneous power data into the spatiotemporal feature extraction model, and extract the spatiotemporal correlation features of power supply and demand based on the system topology and time series; S3: Using a supply-demand coupling correlation model, the spatiotemporal correlation characteristics are coupled and analyzed to quantify the interaction between the power supply side and the demand side, and the supply-demand collaborative prediction results are output. S4: Based on the dynamic correction prediction strategy of rolling time domain, the supply and demand collaborative prediction results are rolled and corrected in real time to obtain the final supply and demand prediction results.

[0018] Next, combined Figure 1 This embodiment provides a detailed description of a method for predicting electricity spot supply and demand that integrates multi-source features and dynamic correction.

[0019] In S1, multi-source data affecting the supply and demand of the regional electricity spot market are obtained, specifically including: (1) Demand-side data: including historical electricity load data, user behavior data, economic data, meteorological data, and policy-related data.

[0020] The historical electricity load data includes the total regional load and the load by industry; user behavior data includes the distribution of electricity consumption time periods and demand response execution records; economic data includes the regional GDP growth rate, industrial output value, and the proportion of the tertiary industry; meteorological data includes temperature, humidity, wind speed, and light intensity; and policy-related data includes electricity price adjustment policies, environmental protection production restriction notices, and arrangements for large-scale events.

[0021] (2) Power supply side data: including conventional unit data, new energy data, cross-regional power transmission data, and energy storage data.

[0022] The conventional unit data includes the installed capacity, maintenance plan, and marginal cost of thermal power units; the new energy data includes the installed capacity, predicted output power, and curtailment rate of wind power / photovoltaic power; the inter-regional power transmission data includes the transmission capacity of tie lines and the planned transmission power; and the energy storage data includes the installed capacity, charge and discharge plan, and state of charge (SOC).

[0023] Traditional models only collect single load or meteorological data. This embodiment expands the data dimensions to cover both direct and indirect influencing factors on both the supply and demand sides, laying a data foundation for a comprehensive depiction of the supply and demand situation.

[0024] Subsequently, for the collected multi-source data, the "3σ criterion" method was used to handle outlier data: (1) Identify regular outliers based on the 3σ criterion: For data that follows a normal distribution (such as historical load and temperature), calculate the mean μ and standard deviation σ of the data, and mark the values ​​that exceed the range of [μ-3σ, μ+3σ] as regular outliers.

[0025] (2) Outlier repair: An improved linear interpolation method is used to repair missing and outlier data. The formula is as follows: (1) in, The data value after repair at time t. , These are the normal data times adjacent to time t. , These are the normal data values ​​at the corresponding time points.

[0026] Traditional models only use simple linear interpolation without distinguishing the data distribution characteristics. This embodiment improves the accuracy of data repair by anomaly identification and improved interpolation methods, effectively reducing the interference of noise on subsequent predictions.

[0027] Furthermore, data normalization and fusion are carried out.

[0028] (1) Data normalization: The min-max normalization method is used to eliminate dimensional differences and map all data to the interval [0,1]. The formula is as follows: (2) in, Here, x represents the normalized data, and x represents the original data. , These are the minimum and maximum values ​​of the original data, respectively.

[0029] (2) Data fusion: Construct a multi-source data fusion module based on the attention mechanism, and assign weights according to the contribution of different data to supply and demand forecasting, as shown in the following formula: (3) (4) Where F is the fused data feature matrix, and K is the number of data types. , where is the attention weight for the k-th class of data (reflecting data importance, ranging from [0,1]). Let be the feature matrix of the k-th class of data. represents the weight parameters for the k-th class of data.

[0030] Traditional models use equal-weight fusion, which cannot distinguish the importance of data (e.g., temperature has a much greater impact on load than humidity). This embodiment uses an attention mechanism to dynamically allocate weights, improving the sensitivity of the fused data features to changes in supply and demand, and providing more effective data input for subsequent feature extraction.

[0031] This embodiment cleans, normalizes, and merges multi-source data on the supply and demand of the regional electricity spot market to eliminate data noise and dimensional differences, constructing a unified dataset that provides high-quality data support for subsequent spatiotemporal feature extraction.

[0032] S2, Construct a spatiotemporal feature extraction model based on Attention-STGCN.

[0033] Based on multi-source data preprocessing, deep learning models are used to mine the spatiotemporal correlation and latent features of supply and demand, capture the complex changing patterns of supply and demand, and solve the problem of insufficient feature extraction depth in traditional models.

[0034] First, we acquire spatiotemporal characteristics, dividing the supply and demand characteristics of the regional electricity spot market into temporal and spatial dimensions: 1) Time dimension characteristics: including intraday load time-series fluctuations (such as morning peak and evening peak), intraweekly periodic characteristics (such as the difference between weekday and weekend loads), seasonal characteristics (such as summer cooling load and winter heating load), and long-term trend characteristics (such as load increases brought about by economic growth).

[0035] 2) Spatial dimension characteristics: including the load distribution of nodes within the region (such as load differences between industrial parks and residential areas), cross-regional load transmission (such as the load spillover effect between neighboring cities), and the spatial distribution of new energy sources (such as the spatial correlation of output of wind power bases and photovoltaic power stations).

[0036] Next, the Attention-STGCN model structure is constructed.

[0037] Load nodes, power supply nodes, and tie-line nodes are used as vertices in the system topology graph, and electrical connections between nodes (such as transmission lines) are used as edges. Spatial features are extracted using graph convolution (GCN) operations, as shown in the following formula: (5) in, Output matrix for spatial features. X is the normalized adjacency matrix (reflecting the connection strength between nodes), and X is the fused data input matrix. The spatial feature weight matrix is... For bias terms, This is the activation function.

[0038] Traditional models do not consider spatial topological relationships and cannot capture cross-regional supply and demand relationships. This embodiment effectively extracts spatial features through graph convolution, which effectively makes up for the limitations of traditional models. It can not only accurately capture cross-regional supply and demand relationships, but also more accurately depict the dynamic changes of the electricity spot market through deep fusion of spatiotemporal features, thereby improving the accuracy and timeliness of supply and demand forecasts.

[0039] Furthermore, an expanded convolutional structure, namely a time-series convolutional layer (TCN), is employed to broaden the receptive field in the temporal dimension and capture long-term temporal dependent features (such as multi-day load fluctuation correlations), as shown in the following formula: (6) in, Output matrix for time series features. For dilated convolution operations, dil is the dilation coefficient (controlling the size of the receptive field). This is the time-series feature weight matrix. This is a bias term.

[0040] Traditional time series models suffer from the problem of gradient vanishing over long sequences. This embodiment expands the temporal receptive field by more than 10 times through dilated convolution, effectively capturing long-term time series features and reducing the prediction error of intraday load peaks.

[0041] After spatiotemporal feature fusion, key features (such as load mutation features under extreme weather conditions) are enhanced through an attention mechanism, while irrelevant features are suppressed, as shown in the following formula: (7) in, The spatiotemporal feature matrix after attention weighting. , , These are query, key, and value matrices, respectively. is the dimension of the key matrix (to avoid gradient explosion), and softmax is the normalization function.

[0042] Traditional models treat all spatiotemporal features equally, failing to focus on key information. This embodiment enhances the weight of key features through an attention mechanism, thereby significantly increasing the influence of core features. It can more sensitively capture load changes in special scenarios, avoid interference from irrelevant information, make supply and demand forecasts more in line with actual working condition fluctuations, and further reduce prediction errors in key scenarios.

[0043] The Adam optimizer is used to minimize the prediction error, and the root mean square error (RMSE) is used as the loss function, as shown in the following formula: (8) in, Here, N represents the model loss value, and N is the number of samples. This refers to the actual supply and demand values ​​(such as actual load and actual power supply). These are the model's predicted values.

[0044] An early stopping strategy is employed to avoid model overfitting: training is terminated when the validation set loss shows no decreasing trend for 10 consecutive rounds. This approach not only ensures the model's generalization ability but also provides more accurate feature input support for subsequent supply-demand coupling correlation analysis.

[0045] S3, Establish a supply and demand coupling correlation analysis model.

[0046] Based on the spatiotemporal feature extraction results, the interaction between the power supply side and the demand side is quantified, and a supply-demand collaborative prediction model is constructed to solve the defects of the traditional model in independent supply and demand prediction, thereby quantifying the coupling effect between the supply and demand sides.

[0047] First, we define three types of core coupling factors to characterize the supply and demand interaction: (1) Output-load coupling factor ( : Reflects the impact of new energy output fluctuations on load demand (such as the matching degree between peak photovoltaic output and daytime industrial load), range [-1,1], positive value indicates that output and load are positively correlated (high matching degree), negative value indicates negative correlation (low matching degree).

[0048] (2) Price-demand coupling factor ( ): Reflects the regulating effect of spot electricity prices on electricity demand (such as electricity price increases suppressing high energy-consuming loads), range [0,1], the larger the value, the stronger the ability of electricity prices to regulate demand.

[0049] (3) Cross-regional-local coupling factor ( ): Reflects the impact of cross-regional power transmission on the local supply and demand balance (such as the increase in tie line power to alleviate the local supply and demand gap), with a range of [-1,1]. Positive values ​​indicate that the increase in power transmission improves the local supply and demand balance, while negative values ​​indicate that it worsens.

[0050] Next, a supply and demand coupling model is constructed.

[0051] (1) Demand forecasting sub-model: Based on spatiotemporal characteristics and coupling factors, the regional electricity demand is predicted, and the formula is as follows: (9) in, To predict demand, , , Spatiotemporal characteristics, , The weight parameters, Here, f is the bias term, and f is the prediction function.

[0052] (2) Power Supply Prediction Sub-model: Combining spatiotemporal characteristics and coupling factors, the regional power supply capacity is predicted using the following formula: (10) in, To predict demand, , , Spatiotemporal characteristics, , The weight parameters, is the bias term, and g is the prediction function.

[0053] (3) Supply and demand balance forecast: Calculate the supply and demand gap / surplus, as follows: (11) in, To predict the supply-demand balance, a positive value indicates a power surplus, while a negative value indicates a power shortage.

[0054] Traditional models independently predict demand and supply without considering coupling effects. This embodiment uses coupling factors to synergistically link supply and demand forecasts, accurately quantifying the interactive impact of factors such as renewable energy output, electricity prices, and inter-regional power transmission on supply and demand. It also ensures that demand and supply forecasts are no longer isolated but form a linked feedback loop. The resulting supply-demand gap / surplus more accurately reflects the actual dynamic balance of supply and demand in the electricity spot market. This not only improves the accuracy and reliability of forecasts but also provides more realistic decision-making support for market trading strategy formulation, inter-regional power allocation, and early warning of supply-demand imbalance risks, thus contributing to the efficient and optimal allocation of power resources.

[0055] Furthermore, the coupling factor is dynamically updated based on real-time market data (such as real-time electricity price, real-time power output, and real-time load). The exponential smoothing method is used to dynamically update the coupling factor, as shown in the following formula: (12) in, Let be the updated value of the k-th type coupling factor at time t. This is a smoothing coefficient (with a value of 0.3-0.5, controlling the degree of influence on real-time data). Let t be the observed value of the k-th type coupling factor at time t (calculated based on real-time data). This represents the historical value of the k-th type of coupling factor at time t-1.

[0056] Traditional models use fixed coupling factors, which cannot adapt to dynamic market changes. This embodiment updates the coupling factors in real time, allowing them to dynamically adapt to market data such as real-time electricity prices and power output. This breaks the static limitations of fixed factors, enabling timely capture of instantaneous fluctuations in the electricity spot market. It ensures that supply and demand forecasts always align with the latest operating conditions, effectively improving the model's response speed to market dynamics. This further guarantees the timeliness and accuracy of supply and demand forecasts, better adapts to the real-time operational needs of the electricity market, and ensures the timeliness of supply and demand forecasts.

[0057] In S4, a dynamic correction prediction strategy based on the rolling time domain is constructed.

[0058] To address the dynamic adaptability issue in supply and demand forecasting, a rolling time-domain strategy is adopted based on supply and demand coupling correlation analysis to adjust forecast results in real time, track scenario changes, and solve the problem of decreased forecast accuracy in extreme scenarios.

[0059] First, a rolling time-domain window is designed. The forecast period is divided into multiple overlapping time-domain windows (e.g., the forecast period is 24 hours, the window length is 4 hours, and the rolling step is 1 hour). Within each window, the "data preprocessing - feature extraction - supply and demand coupling forecast" process is re-executed based on the latest real-time data (e.g., the actual load, output, and electricity price of the previous hour) to achieve rolling updates of the forecast results.

[0060] The window length and rolling step are set according to the market trading time scale (e.g., if the spot market trading interval is 15 minutes, the window length is set to 1 hour and the rolling step is set to 15 minutes) to ensure that the forecast results match the market rhythm.

[0061] Next, a dynamic correction model is constructed. A bias feedback mechanism is employed to correct the current prediction result in real time based on historical prediction biases, as shown in the following formula: (13) in, This is the corrected forecast value (demand, power supply, or supply-demand balance value). This is the original predicted value for the current window. This is a correction factor (with a value of 0.2-0.4, representing the degree of influence of the control deviation). This represents the prediction error at time t-1.

[0062] (14) in, This is the actual value at time t-1. This is the original predicted value at time t-1.

[0063] Traditional static prediction models lack a real-time correction mechanism, and deviations accumulate over time. This embodiment dynamically adjusts the prediction results through deviation feedback, significantly improving the prediction accuracy in extreme scenarios (such as cold waves and typhoons) and effectively avoiding market risks caused by prediction deviations.

[0064] Furthermore, the prediction results will be evaluated and feedback will be provided.

[0065] (1) Evaluation indicators: The mean absolute percentage error (MAPE) and maximum absolute error (MAE) are used to evaluate the prediction accuracy, and the formulas are as follows: (15) (16) in, This represents the actual value (demand, power supply, or supply-demand balance value) at time i. Let be the corrected predicted value at time i, and N be the number of evaluation samples. MAPE reflects the relative error of the prediction (the smaller the value, the higher the accuracy), and MAE reflects the absolute error of the prediction (the smaller the value, the higher the accuracy).

[0066] (2) Feedback optimization: When MAPE exceeds a preset threshold (e.g., 5%), the model parameters are automatically adjusted (e.g., the attention weights of Attention-STGCN, the smoothing coefficient of the coupling factor). ), and retrain the model to ensure stable long-term prediction accuracy.

[0067] This embodiment enables the prediction model to have real-time adaptive capabilities through rolling time domain and dynamic correction, solving the problem of poor dynamic adaptability of traditional models and providing accurate and timely supply and demand situation support for real-time trading decisions in the regional electricity spot market.

[0068] This specific embodiment achieves more accurate and adaptive predictions of the supply and demand situation in the regional electricity spot market through multi-source heterogeneous data fusion and deep learning modeling. First, multi-dimensional data preprocessing covering both supply and demand sides, combined with attention-based fusion, significantly improves the comprehensiveness and representational ability of the input data. Second, the Attention-STGCN model, combining graph convolution and temporal convolution, effectively extracts the complex spatiotemporal correlation features of the power system, enhancing the model's ability to characterize cross-regional transmission and long-term temporal patterns. Third, by introducing and dynamically updating coupling factors, collaborative prediction on both the supply and demand sides is achieved, overcoming the limitations of traditional independent prediction models and better reflecting actual market interaction mechanisms. Finally, a dynamic correction strategy based on rolling time domain and deviation feedback enables the model to have real-time tracking and adaptive adjustment capabilities, significantly improving prediction accuracy and robustness in extreme scenarios.

[0069] Example 2 This embodiment provides a power spot supply and demand forecasting system that integrates multi-source features and dynamic correction, including: The data acquisition module is used to acquire multi-source heterogeneous power data and perform preprocessing. The multi-source heterogeneous power data includes demand-side data and power supply-side data. The feature extraction module is used to input the preprocessed multi-source heterogeneous power data into the spatiotemporal feature extraction model, and extract the spatiotemporal correlation features of power supply and demand based on the system topology and time series. The preliminary prediction module is used to perform coupling analysis on the spatiotemporal correlation characteristics using a supply and demand coupling correlation model, quantify the interaction between the power supply side and the demand side, and output the supply and demand collaborative prediction results. The result correction module is used to perform real-time rolling correction on the supply and demand collaborative prediction results based on a dynamic correction prediction strategy in the rolling time domain, so as to obtain the final supply and demand prediction results.

[0070] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the electricity spot supply and demand forecasting method that integrates multi-source features and dynamic correction as described in Embodiment 1 above.

[0071] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the electricity spot supply and demand forecasting method that integrates multi-source features and dynamic correction as described in Embodiment 1 above.

[0072] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction, characterized in that, include: Acquire multi-source heterogeneous power data and perform preprocessing. The multi-source heterogeneous power data includes demand-side data and power supply-side data. The preprocessed multi-source heterogeneous power data is input into the spatiotemporal feature extraction model, and the spatiotemporal correlation features of power supply and demand are extracted based on the system topology and time series. A supply-demand coupling correlation model is adopted to perform coupling analysis on the spatiotemporal correlation characteristics, quantify the interaction between the power supply side and the demand side, and output the supply-demand collaborative prediction results. Based on a dynamic correction prediction strategy in the rolling time domain, the supply and demand collaborative prediction results are corrected in real time to obtain the final supply and demand prediction results.

2. The method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction as described in claim 1, characterized in that, The demand-side data includes historical electricity load data, user behavior data, economic data, meteorological data, and policy-related data; the supply-side data includes conventional generating unit data, new energy data, inter-regional power transmission data, and energy storage data.

3. The method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction as described in claim 1, characterized in that, The extraction of spatiotemporal correlation features of power supply and demand based on system topology and time series data specifically includes: The load nodes, power supply nodes, and tie-line nodes in the power system are constructed into a topology graph; Based on the topological graph, spatial dimension features are extracted using graph convolution operations to characterize the relationships between nodes; and long-term dependency features in the time dimension are extracted using temporal convolution structures. The extracted spatiotemporal features are weighted using an attention mechanism to enhance key features and suppress irrelevant features.

4. The method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction as described in claim 1, characterized in that, The training process of the spatiotemporal feature extraction model employs an adaptive optimization algorithm to minimize the prediction error and uses an early stopping strategy to prevent overfitting, thereby ensuring the model's generalization ability.

5. The method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction as described in claim 1, characterized in that, The method employs a supply-demand coupling correlation model to perform coupling analysis on the spatiotemporal correlation characteristics, quantifies the interactive influence between the power supply side and the demand side, and outputs supply-demand collaborative prediction results, specifically including: Three core coupling factors are defined: output-load, price-demand, and cross-regional-local to quantify the supply-demand interaction. By combining the spatiotemporal correlation features with the coupling factor, demand prediction sub-model and power supply prediction sub-model are constructed respectively for collaborative prediction. The supply and demand balance is predicted by calculating the difference between the two sub-models.

6. The method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction as described in claim 5, characterized in that, The core coupling factor is dynamically updated based on real-time market data using an exponential smoothing method, so that supply and demand co-forecasting adapts to dynamic changes in market conditions.

7. The method for predicting electricity spot supply and demand by integrating multi-source features and dynamic correction as described in claim 1, characterized in that, The dynamic correction forecasting strategy based on the rolling time domain performs real-time rolling correction on the supply and demand collaborative forecasting results to obtain the final supply and demand forecasting results, specifically including: The forecast period is divided into overlapping rolling time windows; Within each window, the prediction process is re-executed based on the latest real-time data, and the prediction deviation from the previous moment is used to correct the current prediction result. The corrected prediction results are evaluated, and the model parameters are automatically adjusted based on the evaluation error.

8. A power spot supply and demand forecasting system integrating multi-source features and dynamic correction, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous power data and perform preprocessing. The multi-source heterogeneous power data includes demand-side data and power supply-side data. The feature extraction module is used to input the preprocessed multi-source heterogeneous power data into the spatiotemporal feature extraction model, and extract the spatiotemporal correlation features of power supply and demand based on the system topology and time series. The preliminary prediction module is used to perform coupling analysis on the spatiotemporal correlation characteristics using a supply and demand coupling correlation model, quantify the interaction between the power supply side and the demand side, and output the supply and demand collaborative prediction results. The result correction module is used to perform real-time rolling correction on the supply and demand collaborative prediction results based on a dynamic correction prediction strategy in the rolling time domain, so as to obtain the final supply and demand prediction results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the electricity spot supply and demand forecasting method that integrates multi-source features and dynamic correction as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the electricity spot supply and demand forecasting method that integrates multi-source features and dynamic correction as described in any one of claims 1-7.

Images