Electric power spot transaction decision support system based on multi-dimensional market analysis

Abstract

An electric power spot transaction decision support system based on multi-dimensional market analysis relates to the field of electric power spot transaction, and comprises a display layer which supports complex interaction through a Web end, large-screen cockpit command center monitoring and small program mobile end access, and realizes interaction result multi-terminal output. The application layer is communicated with the display layer, comprises a market analysis module and is used for processing multi-dimensional external market data to generate standardized background data; a user side load prediction module generates a user load day-ahead prediction result through an ARIMAX model based on historical power consumption and external factors, predicts monthly power by combining multiple linear regression and other models, and dynamically allocates weights by adopting an error reciprocal normalization method; and the day-ahead price prediction module cleans and preprocesses data such as historical electricity prices, and outputs a result through multi-model price comparison evaluation. The bottom layer is integrated with a scheduling management system through a data interaction interface, is in butt joint with a power grid operation environment and supports efficient decision-making of energy business.

Description

Technical Field

[0001] This invention relates to the field of electricity spot trading, and specifically to a decision support system for electricity spot trading based on multi-dimensional market analysis. Background Technology

[0002] Against the backdrop of deepening power system reform, the retail electricity market is gradually being liberalized, and retail electricity companies, as a new type of electricity sales entity, have become a crucial link in the entire electricity trading market. In this new retail market, retail electricity companies purchase electricity from power generation companies through competitive market transactions and then sell it to users, earning profits from the difference between the purchase and sale price. They have become an indispensable intermediary between the generation and consumption sides. In the process of electricity market transactions, retail electricity companies can independently determine the scale of their purchases. Similarly, users will adjust their purchases from retail electricity companies based on changes in market prices and their own production plans. This situation means that retail electricity companies must face both the randomness of market price fluctuations and the uncertainty of user load adjustments. These influencing factors bring significant risks to retail electricity companies participating in electricity market transactions.

[0003] Therefore, how electricity sales companies can make reasonable use of existing market rules, declare appropriate electricity consumption and set reasonable electricity sales prices, avoid market risks and maximize their own interests is an important issue they are currently facing. Summary of the Invention

[0004] The technical problem to be solved by this invention is how to address the issues of data dispersion and insufficient prediction accuracy in electricity spot trading.

[0005] This invention solves the above-mentioned technical problems through the following technical means: a decision support system for electricity spot trading based on multi-dimensional market analysis, comprising: The presentation layer, used to output interactive results to users in a visual format across multiple terminals, includes: Web-based, used to support complex interactive analysis; A large-screen cockpit is used for real-time monitoring by the command center; Mini programs are designed for flexible access on mobile devices. The application layer, which communicates with the presentation layer, is used to implement core business processing. The application layer includes: The market analysis module is used to acquire and process multi-dimensional external market environment data, form standardized market background data, and output it. The user-side load forecasting module is used to generate day-ahead forecasts of user load based on historical electricity consumption data and external influencing factors through the ARIMAX model. It uses a combination of multiple linear regression model, BP neural network model and support vector regression model for forecasting, and dynamically allocates the weights of each model using the normalization method of the inverse of the error to generate monthly electricity forecasts. The formula for dynamically allocating the weights of each model using the normalization method based on the inverse of the error is as follows:

[0006] in, , This represents the mean absolute percentage error of model k; Calculate the mean absolute percentage error of each model within the R-month validation window:

[0007] in, It is the first Actual electricity consumption over the month It is a model No. Forecast value for one month; The day-ahead price forecasting module is used to clean and preprocess historical electricity price data, grid boundary data and external related data, and to compare and evaluate prices based on multiple forecasting models, and output the day-ahead electricity price forecast results. At the bottom layer, data is exchanged with the dispatch and management system through a data interaction interface to achieve integration between the system and the actual operating environment of the power grid.

[0008] This invention adapts to multiple terminals, including the web interface, large-screen dashboard, and mini-programs, to meet the diverse needs of scenarios such as complex analysis, real-time monitoring, and mobile access. In the application layer, the market analysis module standardizes multi-dimensional external data, providing a reliable data foundation for business operations. The user-side load forecasting module innovatively adopts a "combined model + dynamic weight" strategy—integrating the ARIMAX model to generate day-ahead load forecasts, and using multiple linear regression, BP neural network, and support vector regression models to generate monthly electricity forecasts. Weights are dynamically adjusted based on the inverse of the average absolute percentage error of each model, leveraging the complementary advantages of different models to reduce the bias of a single model, while adaptive weight allocation improves forecast timeliness and accuracy. The day-ahead price forecasting module enhances the credibility of electricity price forecasts through data cleaning and multi-model evaluation. At the bottom layer, deep integration with the dispatch management system via data interaction interfaces establishes a data link with the power grid. Overall, this achieves optimized user experience, improved business processing efficiency, and enhanced grid collaboration, effectively strengthening the reliability of electricity market analysis and forecasting, and providing strong support for accurate decision-making and intelligent operation.

[0009] Furthermore, the market analysis module calculates the market supply and demand adequacy index in real time based on the cleaned and standardized multi-dimensional external market environment data. The calculation formula is as follows:

[0010] in, express Total market supply capacity at any given time Let be the power output data of the i-th power source (i=1,...,n) at time t0, where n is the total number of power sources participating in the power supply on that day. This represents the total market demand at time t0. This represents the overall load sequence at time t0; The multi-dimensional external market environment data includes: the historical day-ahead electricity price series on the user side, the real-time provincial average node electricity price series, the unified dispatch load series, and the output data of various power sources.

[0011] Furthermore, the market analysis module allows users to select time periods through a graphical interface. The system automatically calculates and displays the price-load elasticity coefficient, and its calculation formula is as follows:

[0012] in, , , representing the changes in price and load respectively over the selected time period. and These represent the average price and load during that period, respectively.

[0013] Furthermore, the user-side load forecasting module employs the ARIMAX model when performing day-ahead forecasting of user load, specifically including: For the original non-stationary load sequence conduct Order difference to obtain stationary sequence :

[0014] in, It is a shift operator. ; Establish Holiday dummy variables ARIMAX model:

[0015] Where c is a constant term, It is an autoregressive term. It is a moving average term. This is the holiday effect coefficient. For holidays, dummy variables Noise term; The holiday dummy variable takes the value 1 when the target day t is a holiday, and 0 otherwise.

[0016] Furthermore, the holiday effect coefficient Historical day sets based on the same holiday type calculate:

[0017] in, As the base of the historical day set, It is a historical day The actual observed load value, It is a historical day The baseline load forecast value.

[0018] Furthermore, the specific form of the multiple linear regression model is as follows:

[0019] in, For the intercept term, ,..., For regression coefficients, ,..., For input features.

[0020] Furthermore, the specific form of the BP neural network model is as follows:

[0021] Suppose the network has a hidden layer, its mathematical expression can be refined as follows:

[0022] in, This is the weight matrix from the input layer to the hidden layer. This represents the bias vector of the hidden layer neurons. It is a non-linear activation function. The weight vector from the hidden layer to the output layer. This is the bias of the output layer neurons.

[0023] Furthermore, the specific form of the support vector regression machine model is as follows:

[0024] in, and For the j-th Lagrange multiplier, The number of support vectors, For kernel function, For bias terms; The input feature vector X for combined prediction includes: historical electricity consumption, contract information, temperature records, and macroeconomic data such as GDP and IPI.

[0025] Furthermore, the calculation formula for the monthly electricity consumption forecast result is as follows: .

[0026] Furthermore, the day-ahead price forecasting module includes: The data acquisition and preprocessing unit is used to collect historical electricity price data, grid boundary data, and external correlation data. The historical electricity price data includes electricity price information for different time periods, the grid boundary data includes system load, renewable energy output, and Class A unit information, and the external correlation data is obtained through web crawling and covers weather conditions, holiday information, and macroeconomic indicators. The collected data is cleaned and preprocessed, including missing value handling, outlier detection and removal, and data standardization. The factor mining and model building unit is used to screen factors with strong correlation to electricity price fluctuations based on the historical electricity price data and preprocessed data, and to build a predictive model. The predictive model building process includes candidate model screening, model evaluation and comparison, and model parameter optimization. The multi-model ensemble prediction unit is used to integrate multiple optimized prediction models through Bagging, Boosting or Stacking methods, and combine the output results of each model to form the final electricity price prediction value. The historical data interaction and review unit is used by the system to provide an interface for importing historical electricity spot price information. After the transaction is completed, the actual electricity price data is imported, allowing traders to compare the predicted electricity price with the actual electricity price, analyze the deviation of strategy returns, and optimize subsequent trading strategies.

[0027] The advantages of this invention are: 1. The presentation layer of this invention adopts a layered design of Web terminal (supporting complex interactive analysis), large screen cockpit (real-time monitoring in command center) and mini-program (flexible access on mobile terminal) to adapt to the usage scenarios of different user roles. It not only meets the needs of in-depth data analysis, but also ensures the flexibility of real-time monitoring and mobile office, significantly improving the user operation experience and decision-making efficiency.

[0028] 2. The application layer market analysis module of this invention integrates multi-dimensional external market environment data (such as power output, dispatch load, etc.) and outputs standardized results, which solves the problems of scattered and inconsistent formats of traditional power trading data and effectively reduces data preprocessing time and errors.

[0029] 3. The user-side load forecasting module of this invention introduces the ARIMAX model (integrating holiday dummy variables) for day-ahead forecasting to solve the problem that traditional models ignore electricity consumption fluctuations on special dates; the monthly forecasting adopts a combination of multiple linear regression model, BP neural network model and support vector regression model, and optimizes the model output through a dynamic weight self-learning mechanism, taking into account the advantages of different models, effectively reducing the error of a single model, and significantly improving the accuracy of load forecasting.

[0030] 4. The day-ahead price forecasting module of this invention achieves accurate judgment on electricity price fluctuations through data cleaning, factor mining (screening strongly correlated influencing factors), and multi-model integration (Bagging / Boosting / Stacking), combined with historical review function (comparing forecasts with actual electricity prices), helping traders to avoid price risks and capture profit opportunities in advance. Attached Figure Description

[0031] Figure 1 This is a system framework diagram of a power spot trading decision support system based on multi-dimensional market analysis, according to Embodiment 1 of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1 like Figure 1 As shown, a decision support system for electricity spot trading based on multi-dimensional market analysis includes: The presentation layer is used to output interactive results to users in a visual format across multiple terminals. The presentation layer includes: Web-based, used to support complex interactive analysis.

[0034] A large-screen cockpit is used for real-time monitoring by the command center.

[0035] Mini programs are designed for flexible access on mobile devices.

[0036] The application layer, which communicates with the presentation layer, is used to implement core business processing. The application layer includes: The market analysis module is used to acquire and process multi-dimensional external market environment data, form standardized market background data, and output it.

[0037] Specifically, this module summarizes and displays publicly available data on spot market transactions, including: historical day-ahead electricity prices for users, real-time provincial average node electricity prices, centrally dispatched loads, various power source data, market supply and demand information published by the trading center, presented in chart form for traders to analyze and process.

[0038] The data integration and processing unit is used to periodically retrieve and verify raw data from the external power trading center API via a preset high-frequency data interface. Multi-dimensional external market environment data includes at least: historical day-ahead electricity price sequences on the user side. Real-time provincial average nodal electricity price sequence , Unified load sequence And output data for various power sources (such as thermal power, wind power, and photovoltaic power). To ensure the accuracy and consistency of subsequent correlation analysis, the raw data was standardized using the following formula to eliminate the influence of units:

[0039] in, For any original data sequence, This is the sample mean of the sequence. This is the sample standard deviation of the sequence. This is the standardized data sequence.

[0040] The core market analysis unit, through its built-in analysis engine, performs multi-dimensional market insights based on cleaned and standardized multi-dimensional external market environment data. One of its core functions is to calculate and visualize the market supply and demand margin index. This index dynamically reflects the degree of easing or tightening in market supply and demand through the following quantitative model:

[0041] in, express Total market supply capacity at any given time For the i-th power source (i=1,...,n) in The power output data at any given moment, where n is the total number of power sources participating in the market's power supply on that day. express Total market demand at any given moment express The load sequence of the unified dispatch at any given time; The visualization and interaction unit is used to render analysis results in a high-performance manner through web-based platforms and large-screen dashboards, presenting interconnected chart groups. Traders can select specific time periods via a graphical interface (such as mouse selection). The system will automatically trigger the calculation of the price-load elasticity coefficient for that period. The value is then highlighted in the chart. The formula for calculating this coefficient is:

[0042] in, , , representing the changes in price and load respectively over the selected time period. and These are the average prices and loads for that period, respectively. This coefficient... It is used to quantitatively reveal the sensitivity of local electricity prices to load changes, helping traders assess market risks and potential trading opportunities.

[0043] The user-side load forecasting module is used to generate day-ahead load forecasts based on historical electricity consumption data and external influencing factors using the ARIMAX model. It uses a combination of multiple linear regression, BP neural network, and support vector regression models to forecast loads, and dynamically allocates the weights of each model using the normalization method of the inverse of the error to generate monthly electricity forecasts.

[0044] Specifically, the optimization of electricity sales companies' trading strategies is based on user load forecast data, and the forecast accuracy of each variable has a significant impact on the effectiveness of adjusting the reporting strategy. Timely and accurate user load forecasting is a prerequisite for electricity sales companies to formulate spot trading strategies. Most of the customers of electricity sales companies are large industrial users whose loads are less affected by factors such as weather and holidays. However, changes in production plans, production processes, and the operating status of the equipment itself lead to drastic fluctuations in the electricity load of large users, with high randomness, making accurate load forecasting very difficult. The main idea of ​​large industrial user load forecasting is to combine historical load data with the electricity consumption characteristics of different users' production plans and external influencing factors such as economics to mine the electricity consumption patterns of users from the data, and obtain the energy consumption trend forecast of users at a certain point in time or within a certain period of the future through reasonable forecasting methods.

[0045] This module primarily enables the platform to forecast the day-ahead load of contracted aggregated users and the monthly electricity consumption, providing data support for day-ahead trading in the spot market.

[0046] The main implementation process is as follows: Based on the contract status of retail users, the day-ahead load forecast of contracted users is performed using the above-mentioned market analysis module or the historical power grid electricity consumption data manually imported by the users as the forecast data source.

[0047] The day-ahead user load forecasting unit is used to generate day-ahead user load forecasts using the ARIMAX model. Electricity load typically exhibits a sharp drop or unique pattern during holidays (such as National Day and weekends), differing significantly from weekdays. Pure time series models (such as ARIMAX) struggle to accurately capture these discontinuous pattern abrupt changes caused by calendar events.

[0048] Based on the ARIMA model, a "holiday dummy variable" is introduced as an external covariate to construct the ARIMAX model. The model, in which the external covariates are holiday dummy variables. .

[0049] For the original non-stationary load sequence conduct Order difference to obtain stationary sequence :

[0050] in, It is a shift operator. .

[0051] Establish Holiday dummy variables The model:

[0052] Where c is a constant term, It is an autoregressive term. It is a moving average term. The holiday effect coefficient quantifies the average impact of holidays on the load. For holidays, dummy variables This is the noise term.

[0053]

[0054] For a given target date Its corresponding holiday effect coefficient Defined and calculated in the following ways.

[0055] Extracting data related to the target day from historical data The set of all historical days belonging to the same holiday type (e.g., all being "the first day of National Day"). The cardinality of this set is ,Right now .

[0056] Calculate the target date Holiday effect coefficient The average load deviation for these historical days of the same period:

[0057] in, It is a historical day The actual observed load value, It is a historical day The baseline load forecast value.

[0058] The monthly electricity consumption forecasting unit is used to make predictions using a combination of multiple linear regression model, BP neural network model and support vector regression model. It allocates the weights of each model through a dynamic weight self-learning mechanism to generate monthly electricity consumption forecast results.

[0059] Factors influencing monthly electricity consumption forecasts include historical electricity consumption data, contract information, temperature records, and macroeconomic data (GDP, IPI, etc.). Three models are proposed for combined forecasting, and the weight parameters of these three models can be self-learned and self-optimized.

[0060] Since monthly electricity consumption forecasting is influenced by numerous factors, no single model is optimal in all scenarios. Therefore, this invention proposes a combined forecasting model with three complementary characteristics.

[0061] Multiple linear regression model: provides a robust linear benchmark based on historical patterns.

[0062] BP neural network model: flexibly captures complex nonlinear relationships in data.

[0063] Support vector regression models provide strong generalization in small sample or high-dimensional cases.

[0064] The specific form of the multiple linear regression model is as follows:

[0065] in, For the intercept term, ,..., For regression coefficients, ,..., For input features.

[0066] It provides an interpretable linear benchmark and fits well to stationary trends.

[0067] The specific form of the BP neural network model is as follows:

[0068] Suppose the network has a hidden layer, its mathematical expression can be refined as follows:

[0069] in, This is the weight matrix from the input layer to the hidden layer. This represents the bias vector of the hidden layer neurons. It is a non-linear activation function. The weight vector from the hidden layer to the output layer. This is the bias of the output layer neurons.

[0070] As a universal function approximator, it excels at characterizing the complex nonlinear mapping between electrical quantity and various factors.

[0071] The specific form of the support vector regression machine model is as follows:

[0072] in, and Let be the j-th Lagrange multiplier, which are key parameters obtained after model training. They are all non-negative, and most are zero. The sample points corresponding to the non-zero values ​​are the support vectors. The number of support vectors, For the kernel function, the radial basis function is preferred: It measures new samples by calculating distance. With support vectors The similarity is used to map the data to a high-dimensional space for linear partitioning. Parameters Control the width of the kernel function. This is the bias term, similar to the intercept in linear regression, representing the offset of the regression hyperplane. By minimizing structural risk, it effectively prevents overfitting and demonstrates robustness when dealing with data featuring complex interactions.

[0073] In the above model, the influencing factors for monthly electricity consumption forecasting are defined as follows:

[0074] The design philosophy of the weighting calculation scheme is: the more accurate a model's predictions are in the recent period, the more reliable its predictions for the future are.

[0075] A recent validation window sub-cell is selected to choose the R months closest to the predicted month as the validation period (e.g., the last 6 months). This ensures that the weights reflect the latest performance status of the model.

[0076] The recent prediction error calculation subunit is used to calculate the average absolute percentage error of the three models during this validation period, which serves as an evaluation metric for their recent performance.

[0077]

[0078] in, , It is the first Actual electricity consumption over the month It is a model No. The forecast for the next month.

[0079] The dynamically weighted sub-unit is used to dynamically allocate weights using a normalization method based on the inverse of the error. The weight allocation is inversely proportional to the model's prediction error.

[0080] The calculation is performed using the normalization method based on the reciprocal of the error:

[0081] in, This can be understood as the model's "accuracy score"; the smaller the error, the higher the score. The denominator is the sum of all model accuracy scores. This normalization operation ensures that the sum of the three weights is always 1. .

[0082] The formula for calculating the monthly electricity consumption forecast is as follows: .

[0083] The day-ahead price forecasting module is used to clean and preprocess historical electricity price data and grid boundary conditions, and to compare and evaluate prices based on multiple forecasting models, outputting day-ahead electricity price forecast results.

[0084] Specifically, the day-ahead electricity price forecasting module can predict day-ahead electricity prices based on historical power grid operation data and spot electricity price data. The day-ahead price forecasting algorithm adopts a multi-model integrated forecasting method.

[0085] The data acquisition and preprocessing unit is used to collect historical electricity price data, grid boundary data, and external related data. The historical electricity price data includes electricity price information for different time periods, the grid boundary data includes system load, renewable energy output, and Class A unit information, and the external related data is obtained through web crawling and covers weather conditions, holiday information, and macroeconomic indicators. The collected data is cleaned and preprocessed, including missing value handling, outlier detection and removal, and data standardization.

[0086] The factor mining and model building unit is used to screen factors with strong correlation to electricity price fluctuations based on the historical electricity price data and preprocessed data, and to build a prediction model. The prediction model building process includes candidate model screening, model evaluation and comparison, and model parameter optimization.

[0087] Based on the acquired data and historical electricity prices, correlation and causal analysis methods were used to mine the correlation between electricity prices and their influencing factors, identifying those with strong correlations. Next, a predictive model was constructed, which included: candidate model selection; model evaluation and comparison; and model parameter tuning and optimization. The task of candidate model selection is to compile a list of suitable machine learning model candidates based on the problem type and data characteristics. Factors such as model complexity, fitting ability, interpretability, and runtime efficiency also need to be considered.

[0088] Model evaluation and comparison involves using techniques such as cross-validation to evaluate the performance of candidate models on the training set. Evaluation metrics (such as accuracy, precision, recall, mean squared error, etc.) are used to compare the performance of different models.

[0089] Model parameter optimization is a step to further optimize the model. Its main goal is to tune the parameters of the selected candidate model in order to improve the model's performance.

[0090] The multi-model ensemble prediction unit is used to integrate multiple optimized prediction models through methods such as Bagging, Boosting, or Stacking, and combine the output results of each model to form the final electricity price prediction value.

[0091] The historical data interaction and review unit is used by the system to provide an interface for importing historical electricity spot price information. After the transaction is completed, the actual electricity price data is imported, allowing traders to compare the predicted electricity price with the actual electricity price, analyze the deviation of strategy returns, and optimize subsequent trading strategies.

[0092] Meanwhile, an interface for importing historical spot electricity prices is retained, allowing traders to independently call upon historical spot price information to plan trading strategies. The main purpose is to conduct strategy review after the spot trading process is completed. By comparing the returns of trading strategies based on algorithm-predicted prices with actual prices, the profit deviation caused by price prediction errors can be identified, and this deviation can be corrected when formulating subsequent trading strategies.

[0093] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A decision support system for electricity spot trading based on multi-dimensional market analysis, characterized in that, include: The presentation layer, used to output interactive results to users in a visual format across multiple terminals, includes: Web-based, used to support complex interactive analysis; A large-screen cockpit is used for real-time monitoring by the command center; Mini programs are designed for flexible access on mobile devices. The application layer, which communicates with the presentation layer, is used to implement core business processing. The application layer includes: The market analysis module is used to acquire and process multi-dimensional external market environment data, form standardized market background data, and output it. The user-side load forecasting module is used to generate day-ahead forecasts of user load based on historical electricity consumption data and external influencing factors through the ARIMAX model. It uses a combination of multiple linear regression model, BP neural network model and support vector regression model for forecasting, and dynamically allocates the weights of each model using the normalization method of the inverse of the error to generate monthly electricity forecasts. The formula for dynamically allocating the weights of each model using the normalization method based on the inverse of the error is as follows: in, , This represents the mean absolute percentage error of model k; Calculate the mean absolute percentage error of each model within the R-month validation window: in, It is the first Actual electricity consumption over the month It is a model No. Forecast value for one month; The day-ahead price forecasting module is used to clean and preprocess historical electricity price data, grid boundary data and external related data, and to compare and evaluate prices based on multiple forecasting models, and output the day-ahead electricity price forecast results. At the bottom layer, data is exchanged with the dispatch and management system through a data interaction interface to achieve integration between the system and the actual operating environment of the power grid.

2. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The market analysis module calculates the market supply and demand margin index in real time based on the cleaned and standardized multi-dimensional external market environment data. The calculation formula is as follows: in, express Total market supply capacity at any given time Let be the power output data of the i-th power source (i=1,...,n) at time t0, where n is the total number of power sources participating in the power supply on that day. This represents the total market demand at time t0. This represents the overall load sequence at time t0; The multi-dimensional external market environment data includes: the historical day-ahead electricity price series on the user side, the real-time provincial average node electricity price series, the unified dispatch load series, and the output data of various power sources.

3. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The market analysis module allows users to select time periods through a graphical interface. The system automatically calculates and displays the price-load elasticity coefficient, and its calculation formula is as follows: in, , , representing the changes in price and load respectively over the selected time period. and These represent the average price and load during that period, respectively.

4. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The user-side load forecasting module uses the ARIMAX model when performing day-ahead forecasting of user load, specifically including: For the original non-stationary load sequence conduct Order difference to obtain stationary sequence : in, It is a shift operator. ; Establish Holiday dummy variables ARIMAX model: Where c is a constant term, It is an autoregressive term. It is a moving average term. This is the holiday effect coefficient. For holidays, dummy variables Noise term; The holiday dummy variable takes the value 1 when the target day t is a holiday, and 0 otherwise.

5. The power spot trading decision support system based on multi-dimensional market analysis according to claim 4, characterized in that, The holiday effect coefficient Historical day sets based on the same holiday type calculate: in, As the base of the historical day set, It is a historical day The actual observed load value, It is a historical day The baseline load forecast value.

6. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The specific form of the multiple linear regression model is as follows: in, For the intercept term, ,..., For regression coefficients, ,..., For input features.

7. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The specific form of the BP neural network model is as follows: Suppose the network has a hidden layer, its mathematical expression can be refined as follows: in, This is the weight matrix from the input layer to the hidden layer. This represents the bias vector of the hidden layer neurons. It is a non-linear activation function. The weight vector from the hidden layer to the output layer. This is the bias of the output layer neurons.

8. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The specific form of the support vector regression machine model is as follows: in, and For the j-th Lagrange multiplier, The number of support vectors, For kernel function, For bias terms; The input feature vector X for combined prediction includes: historical electricity consumption, contract information, temperature records, and macroeconomic data such as GDP and IPI.

9. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The formula for calculating the monthly electricity consumption forecast is as follows: 。 10. The power spot trading decision support system based on multi-dimensional market analysis according to claim 1, characterized in that, The day-ahead price forecast module includes: The data acquisition and preprocessing unit is used to collect historical electricity price data, grid boundary data, and external correlation data. The historical electricity price data includes electricity price information for different time periods, the grid boundary data includes system load, renewable energy output, and Class A unit information, and the external correlation data is obtained through web crawling and covers weather conditions, holiday information, and macroeconomic indicators. The collected data is cleaned and preprocessed, including missing value handling, outlier detection and removal, and data standardization. The factor mining and model building unit is used to screen factors with strong correlation to electricity price fluctuations based on the historical electricity price data and preprocessed data, and to build a predictive model. The predictive model building process includes candidate model screening, model evaluation and comparison, and model parameter optimization. The multi-model ensemble prediction unit is used to integrate multiple optimized prediction models through Bagging, Boosting or Stacking methods, and combine the output results of each model to form the final electricity price prediction value. The historical data interaction and review unit is used by the system to provide an interface for importing historical electricity spot price information. After the transaction is completed, the actual electricity price data is imported, allowing traders to compare the predicted electricity price with the actual electricity price, analyze the deviation of strategy returns, and optimize subsequent trading strategies.

Images