Power market information intelligent decision-making method and system based on hierarchical model fusion

Abstract

The invention discloses an intelligent decision-making method and system for electricity market information based on hierarchical model fusion. The method comprises the following steps: collecting multi-time scale operation data, executing multi-model grading prediction, and generating an electric power system operation key parameter prediction result containing a confidence interval; constructing a multi-target evaluation sample set based on the prediction result and the operation data, quantifying a conflict relationship among income, fluctuation, risk and clean energy consumption, and calculating a decision target weight by adopting an analytic hierarchy process; combining the weights to construct a comprehensive fitness function, and generating candidate decision-making schemes through an adaptive genetic algorithm; performing feasibility check on the candidate schemes in a multi-time scale scene to form a virtual running path, and fusing cross-scene information; and determining a final scheme according to a benefit balance requirement, and feeding back the operation deviation to update a prediction model and a weight system, thereby realizing a dynamically adaptive power market intelligent decision.

Description

Technical Field

[0001] This invention belongs to the field of intelligent services and optimization decision-making in the power market, and specifically relates to an intelligent decision-making method and system for power market information based on hierarchical model fusion. Background Technology

[0002] Against the backdrop of high-proportion renewable energy integration, enhanced load-side response, and the coexistence of multiple time-scale operating mechanisms, the electricity market exhibits characteristics such as drastic price fluctuations, increased system operational uncertainty, and intensified conflicts of interest among market participants. Existing electricity market decision-making methods typically rely on a single forecasting model or a static weighting system, making it difficult to simultaneously handle the complex changes in electricity prices, load, and renewable energy output across multiple time scales, and also difficult to quantify the conflicts between different decision objectives. Furthermore, existing optimization methods lack adaptability to real-time market fluctuations and lack automatic feedback mechanisms based on operational deviations, resulting in insufficient stability of decision schemes under uncertain environments and an inability to achieve a coordinated balance between revenue, risk, and clean energy consumption. Therefore, there is an urgent need for an intelligent processing method that can integrate multi-model forecasting results, quantify multi-objective conflicts, and adaptively adjust decision strategies under dynamic market conditions to improve the stability and coordination of electricity market dispatch and clearing. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent decision-making in the power market based on hierarchical model fusion, in order to solve the technical problems of intelligent decision-making in the power market.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution.

[0005] This invention first discloses an intelligent decision-making method for the power market based on hierarchical model fusion, which includes the following steps: Collect multi-timescale data of the electricity market, including short-term operating indicators, long-term planning parameters and external market environment data, and standardize the collected data. Using the multi-timescale data as input, hierarchical prediction is performed to generate prediction results for key parameters of power system operation with confidence intervals, wherein the key parameters of power system operation include electricity price, load and renewable energy output; The prediction results of the key parameters of power system operation are combined with the multi-timescale data to extract features and construct a multi-objective evaluation sample set to quantify the conflict intensity between decision objectives. The hierarchical structure of multiple decision objectives is established by using the analytic hierarchy process (AHP) and the weight coefficients of each decision objective are calculated. The decision objectives include market entity revenue objectives, price fluctuation stability objectives, operational risk control objectives, and clean energy consumption objectives. Based on the weight coefficients, a comprehensive fitness function is constructed by weighting multiple decision objectives. Market clearing decision variables or scheduling decision variables are used as optimization variables. A genetic algorithm is used to perform multi-objective optimization. When the fluctuation range of electricity price or the deviation of operating indicators exceeds the preset threshold, the weight coefficients are adaptively adjusted. Based on the adjusted weight coefficients, the optimization is re-executed to generate a set of candidate decision schemes. Under multi-timescale operation scenarios, the prediction results of the key parameters of the power system operation are used as input to simulate the candidate decision scheme set, generate virtual operation paths, perform information fusion on the virtual operation paths that meet the operation constraints and stability standards, determine the final decision scheme through feasibility verification, and use the actual operation deviation feedback to iteratively update the hierarchical prediction, weight coefficients and comprehensive fitness function.

[0006] The present invention further includes the following preferred embodiments: The short-term operating indicators include time-of-use electricity prices, load, and renewable energy output; the long-term planning parameters include medium- and long-term load forecast data and installed capacity planning data; and the external market environment data include after-hours order book quotes, transaction information, meteorological data, and fuel prices.

[0007] The hierarchical prediction is performed using the multi-timescale data as input, employing one or more of the following prediction paths: a. Using the feature vector composed of historical electricity price, load, renewable energy output and external market environment data as conditional input, the model generates path samples of electricity price, load and renewable energy output for multiple consecutive future periods through conditional generation. Based on the multidimensional scoring function, the model is trained or the samples are screened to output the prediction path of key parameters of power system operation for multiple future periods. b. Construct time series inputs from the buy and sell quotes and transaction information in the after-hours order book according to the buyer sequence and the seller sequence respectively, and use an end-to-end neural network to output the probability distribution or predicted values ​​of the key parameters of the power system operation in multiple future time periods; c. Represent the topology between power grid regions as a graph structure, fuse historical data from multiple regions based on the graph distance between regions, and output the joint prediction sequence of key power system operation parameters in each region for multiple future time periods through a graph neural network; d. The clearing price calculated by the physical constraint-based mechanism model is used as an intermediate variable, and combined with historical load, electricity price and external market environment data to form a joint input. This input is then fed into a deep neural network ensemble model, and the fused prediction result is output. By integrating the prediction results output from different modeling structures, and combining conformal prediction with confidence interval calibration based on historical residual distribution, confidence interval prediction results including upper and lower limits are generated for the key operating parameters of the power system.

[0008] In the hierarchical prediction, the conditional generation model uses historical feature vectors and random noise as inputs to conditionally model the generation network, and uses energy scores as a multidimensional loss function to train the model parameters, so that the generated key parameter paths of power system operation simultaneously satisfy distribution consistency and sample diversity. The end-to-end neural network uses quantile regression to output predicted values ​​of multiple quantiles for multiple future time periods, and decouples the cross risk between adjacent quantiles through non-negative residuals to achieve a stable probability characterization of key parameters for future power system operation. Graph neural networks employ graph attention mechanisms to weighted aggregate regional historical features, and integrate structural priors constructed from graph adjacency matrices and graph distance functions to guide joint modeling and cross-regional prediction of load, electricity price, and power output indicators between regions. The deep neural network ensemble model takes as input a feature vector composed of clearing price, historical load, electricity price and external market environment data output from the mechanistic model, performs nonlinear mapping through a multi-sub-network parallel structure, and outputs the fused predicted values ​​of key power system operation parameters.

[0009] The method of combining conformal prediction with confidence interval calibration based on historical residual distribution to generate confidence interval prediction results including upper and lower limits for the key operating parameters of the power system includes: For the key power system operation parameters output from multiple prediction paths, a conditional distribution set is constructed based on historical observation residuals, and the upper and lower limits of the predicted values ​​are corrected at the target confidence level to obtain the initial confidence interval. Using the conformal prediction method, without relying on specific distribution assumptions, the initial confidence interval is adjusted back on the validation sample set. If the default ratio corresponding to the target confidence level does not meet the statistical significance requirement, the interval width is adaptively increased or decreased to generate the final confidence interval that satisfies the distribution invariance assumption. The final confidence interval coverage is verified by multiple models, and abnormal prediction paths are eliminated or their weights are reallocated to form the integrated confidence interval prediction result.

[0010] The construction of a multi-objective evaluation sample set to quantify the conflict intensity between decision objectives further includes: The available electricity price forecasts, load forecasts, or renewable energy output forecasts, along with the upper and lower limits of their confidence intervals, are synchronized with short-term operating indicators, long-term planning parameters, and external market environment data to form a sample data containing multi-dimensional features. Calculate the evaluation values ​​for the four decision objectives based on the sample data. , , , The evaluation value and the corresponding upper and lower limits of the confidence interval are then incorporated into the multi-objective evaluation sample set. The evaluation value sequence of each decision objective is extracted from the multi-objective evaluation sample set, and the correlation conflict strength is calculated. The formula is: ,in, Indicate decision objectives With decision-making objectives The Pearson correlation coefficient; Data containing the upper and lower limits of confidence intervals are extracted from the multi-objective evaluation sample set, and the interval overlap is calculated using the following formula: ,in Indicate decision objectives With decision-making objectives At any moment The overlap of confidence intervals, , Representing decision objectives With decision-making objectives At any moment The upper limit of the prediction confidence interval, , Representing decision objectives With decision-making objectives At any moment The lower bound of the prediction confidence interval; Based on confidence interval overlap Calculate the interval conflict intensity The formula is: ,in The length of the time window; The correlation conflict strength Intensity of Interval Conflict Weighted synthesis is performed to determine the conflict intensity among the decision objectives. : , This is a preset weighting factor.

[0011] The method of establishing a hierarchical structure of multiple decision objectives using the analytic hierarchy process (AHP) and calculating the weight coefficients of each decision objective specifically includes: Based on the conflict intensity among multiple decision objectives Construct a judgment matrix to compare the relative importance of four types of decision objectives, setting the elements of the judgment matrix for any two different decision objectives as follows: ,in Indicate decision objectives Relative to decision objectives The importance of This indicates determining the matrix scaling factor, and according to... Set the inverse elements of the judgment matrix; Perform eigenvector calculation on the matrix to obtain the eigenvector corresponding to the largest eigenvalue. The feature vector is then normalized. As the first The weight coefficients of each decision objective; when the judgment matrix does not meet the consistency requirement, Adjust the elements of the judgment matrix, where This represents the consistency adjustment factor, and the above eigenvector calculation steps are repeated to obtain the weight coefficients that meet the consistency requirements; The upper and lower limits of the confidence interval are calculated based on the records in the multi-objective evaluation sample set. The average forecast uncertainty of each decision objective The formula is: And calculate the weight coefficient of the decision objective after uncertainty bias, using the following formula: ,in These are the weighting coefficients after uncertainty bias. These are the weighting coefficients obtained after standardization.

[0012] The comprehensive fitness function is expressed as follows: ; Using market clearing decision variables or scheduling decision variables as optimization variables, a genetic algorithm is employed to perform multi-objective optimization, including: by As the basis for optimization in genetic algorithms, among which This represents a decision vector composed of market clearing decision variables or scheduling decision variables. This represents a feasibility penalty calculated based on power balance deviation, transmission line power flow exceeding limits, and insufficient reserve capacity. Indicates the penalty coefficient; Selection, crossover, and mutation operations are performed on the initial population. When electricity price fluctuations or operational indicator deviations exceed preset thresholds, [further actions are taken]. Adjust the mutation probability, where This represents the initial mutation probability. This represents the adjusted mutation probability. Indicates a variable-sensitive factor. A measure of uncertainty in market conditions; When an increase in the prediction uncertainty of key power system operating parameters is detected, based on Adjust the weighting coefficients of the decision-making objectives, where Indicate decision objectives The basic weighting coefficients, This represents the adjusted weighting coefficient. This represents the sensitivity coefficient for weight adjustment. Indicates the first The uncertainty of the predicted value corresponds to each decision objective; During the iterative process of the genetic algorithm, based on Generate a local neighborhood search solution, where This represents the current optimal decision vector. This represents the local disturbance factor. This represents a random vector sampled within the interval [-1, 1]. in accordance with Calculate the degree of difference between different candidate solutions and remove candidate solutions with a degree of difference less than a threshold to obtain a diverse set of candidate decision solutions.

[0013] The scenario simulation of the candidate decision set includes: Using the prediction results of key parameters of power system operation as exogenous input, power market operation scenarios are constructed at multiple short-term, medium-term and long-term time scales. Under each scenario, a physical model containing power flow constraints, unit start-up and shutdown constraints, reserve capacity constraints and transmission channel capacity constraints is invoked to generate virtual operation paths based on the candidate decision schemes. In the virtual operation path, the feasibility of power balance deviation, transmission line power flow change, unit status evolution and renewable energy utilization at each time scale is checked, and virtual operation paths that do not meet any operation constraints or stability check conditions are eliminated. The changes in operational indicators of the verified virtual operation paths are recorded in chronological order to form a set of filtered virtual operation paths for subsequent information fusion.

[0014] The determination of the final solution includes: Information fusion across time scales is performed on the selected set of virtual operation paths. Indicators reflecting changes in market participants' returns, price fluctuations, operational risks, and clean energy consumption at different time scales are processed in a unified manner to form a comprehensive path description that reflects the degree of coordination among multiple objectives. The benefit balance index is calculated based on the comprehensive path description. When the benefit balance index is lower than the preset threshold, the resource allocation parameters are readjusted and the candidate decision scheme set is updated to generate a scheme that meets the benefit balance requirements. After obtaining a final solution that meets the requirements of balancing interests, a risk level assessment is performed. The operational deviations, operational stability indicators, and renewable energy utilization corresponding to the final solution are used as feedback information to update the hierarchical prediction model, target weight coefficients, and comprehensive fitness construction strategy, so as to form a cyclical dynamic adaptation mechanism.

[0015] This invention also discloses a power market information intelligent decision-making system based on hierarchical model fusion, utilizing the aforementioned intelligent decision-making method for power market information based on hierarchical model fusion, comprising: The data acquisition module is used to collect multi-timescale data from the electricity market, including short-term operating indicators, long-term planning parameters, and external market environment data, and to standardize the collected data. The prediction module is used to perform hierarchical prediction with the multi-timescale data as input, and generate prediction results of key parameters of power system operation with confidence intervals, wherein the key parameters of power system operation include electricity price, load and renewable energy output; The feature extraction module is used to combine the prediction results of the key parameters of the power system operation with the multi-timescale data to extract features, construct a multi-objective evaluation sample set, quantify the conflict intensity between decision objectives, and use the analytic hierarchy process to establish a hierarchical structure of multiple decision objectives and calculate the weight coefficient of each decision objective. The decision objectives include market entity revenue objectives, price fluctuation stability objectives, operational risk control objectives, and clean energy consumption objectives. The optimization module is used to construct a comprehensive fitness function formed by a weighted combination of multiple decision objectives based on the weight coefficients. Market clearing decision variables or scheduling decision variables are used as optimization variables. A genetic algorithm is used to perform multi-objective optimization. When the fluctuation range of electricity price or the deviation of operating indicators exceeds a preset threshold, the weight coefficients are adaptively adjusted. The optimization is re-executed based on the adjusted weight coefficients to generate a set of candidate decision schemes. The decision-making module is used to simulate the candidate decision scheme set under multi-timescale operation scenarios, taking the prediction results of the key parameters of the power system operation as input, generate virtual operation paths, perform information fusion on the virtual operation paths that meet the operation constraints and stability standards, determine the final decision scheme through feasibility verification, and use the actual operation deviation feedback to iteratively update the hierarchical prediction, weight coefficients and comprehensive fitness function.

[0016] Accordingly, this application also discloses a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the aforementioned intelligent decision-making method for electricity market information based on hierarchical model fusion.

[0017] Accordingly, this application also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned intelligent decision-making method for electricity market information based on hierarchical model fusion.

[0018] The beneficial effects of this invention are that, compared with existing technologies, it provides an intelligent decision-making method and system for the power market based on hierarchical model fusion. By organically combining multi-timescale prediction models, conflict intensity quantification methods, hierarchical analysis weight calculation, adaptive genetic algorithms, multi-scenario operation simulation, and dynamic feedback update mechanisms, a continuously learning intelligent decision-making system for the power market is constructed. In a power market environment with significantly increased uncertainty, it achieves high-precision prediction of key power system operation parameters, quantitative trade-offs among multiple objectives, stability verification across time scales, and continuous correction of operational deviations, making the decision-making process data-driven, globally coordinated, and dynamically adaptable. It fully captures short-term operational changes and long-term planning trends, ensuring data consistency and comparability in subsequent prediction, evaluation, and optimization processes, and improving the accuracy and completeness of decision input data. By introducing multiple types of prediction models and generating confidence intervals including upper and lower limits, it enables the prediction of electricity prices, load, and renewable energy output to characterize uncertainty, providing a solid data foundation for multi-objective conflict quantification and risk control. By unifying forecast results with operational data to form a quantifiable dataset, the interplay between revenue, volatility, risk, and clean energy consumption can be comprehensively characterized, enabling precise data-driven characterization of multi-objective conflict intensity. Constructing a judgment matrix based on conflict intensity and operational preferences improves the objectivity and adaptability of weight determination, allowing for dynamic adjustment of the importance of each decision objective under changing market conditions. A unified optimization function is formed by weighted combination of multiple objectives, enabling coordinated trade-offs among market clearing or scheduling decision variables and timely adjustment of optimization strategies during market fluctuations through adaptive mechanisms. Constructing multi-level operational scenarios (short-term, medium-term, and long-term) allows for comprehensive evaluation of the feasibility of candidate schemes under different future conditions, ensuring the stability of the final decision scheme across time scales. Ensuring that power balance, line flow, unit status, and reserve capacity all meet physical operational requirements rigorously verifies the physical feasibility of the decision scheme. Unified processing of schemes validated through multiple scenarios enables coordinated optimization of multiple indicators such as revenue, risk, and clean energy consumption for various stakeholders, improving the overall fairness and rationality of the decision scheme. By continuously updating the prediction module, weighting system, and fitness function using actual operational data, the decision-making system possesses continuous learning capabilities, enabling it to maintain dynamic adaptability and stability in complex market environments over the long term. With the technical framework of this invention, the electricity market can more effectively cope with the complex changes brought about by the high proportion of renewable energy integration, improve market clearing efficiency, reduce operational risks, promote the consumption of clean energy, and lay the foundation for future large-scale flexible dispatch and smart electricity trading, demonstrating promising engineering application prospects and widespread value. Attached Figure Description

[0019] Figure 1This is a flowchart of the intelligent decision-making method for electricity market information based on hierarchical model fusion in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0021] The embodiments described in this application are merely some, not all, embodiments of the present invention. Based on the spirit of the present invention, other embodiments obtained by those skilled in the art without inventive effort are all within the protection scope of the present invention.

[0022] To address the shortcomings of existing technologies, this invention proposes an intelligent decision-making method and system for the power market based on hierarchical model fusion. The method collects operational data across multiple time scales, performs hierarchical predictions using multiple models, and generates prediction results for key power system operation parameters including confidence intervals. Based on the prediction results and operational data, a multi-objective evaluation sample set is constructed to quantify the conflict relationships between revenue, volatility, risk, and clean energy consumption. The analytic hierarchy process (AHP) is used to calculate the weights of the decision objectives. A comprehensive fitness function is constructed using these weights, and candidate decision schemes are generated through an adaptive genetic algorithm. The feasibility of these candidate schemes is verified under multi-time scale scenarios, forming virtual operation paths, and cross-scenario information is fused. The final scheme is determined based on the requirements of balancing interests, and operational deviations are fed back to update the prediction model and weight system, achieving dynamically adaptive intelligent decision-making in the power market.

[0023] See Figure 1 As shown, the intelligent decision-making method for electricity market information based on hierarchical model fusion disclosed in this invention uses multi-timescale electricity market data as basic input. Through steps such as hierarchical prediction, feature extraction, multi-objective conflict quantification, hierarchical decision modeling, multi-objective optimization, and scenario simulation, it generates a final decision scheme suitable for market clearing or scheduling processes. Specifically, it includes the following steps: Step S1: Collect multi-timescale data of the electricity market, including short-term operating indicators, long-term planning parameters and external market environment data, and standardize the collected data.

[0024] The operation of the electricity market involves input factors across multiple time scales. Therefore, the following data will be collected first: (1) Short-term operating indicators: These include information reflecting the daily operating status of the power system and changes over hourly or shorter periods, mainly including time-of-use pricing, system load, renewable energy output, and reserve capacity rate. Time-of-use pricing refers to the transaction prices published by the power trading platform at hourly or more detailed time intervals. It can be directly downloaded via the market operation platform's API or data interface. After collection, it can be aligned according to timestamps and standardized, for example, using conventional normalization (min–max) or mean-variance standardization methods. System load represents the real-time power consumption of the entire network or a specific region, which can be obtained from the dispatch center's load monitoring system. It is generally recorded at 5-minute, 15-minute, or 1-hour intervals. This data can be arranged chronologically, aligned with price data, and input into the prediction model. Renewable energy output includes wind power output, photovoltaic power output, etc., and is usually provided jointly by the power grid real-time dispatch system and the meteorological forecasting system.

[0025] (2) Long-term planning parameters: These reflect the development trends and resource capabilities of the power system at monthly, quarterly, or annual scales, and are important bases for medium- and long-term forecasting and planning. They include medium- and long-term load forecasts, installed capacity expansion plans, and unit technical parameters. Medium- and long-term load forecasts are provided by planning departments or dispatch centers based on historical loads, economic development levels, industrial structure, and meteorological trends. They are usually presented as "monthly forecasts" or "annual forecasts," and can be directly extracted from documents or databases published by planning departments, aligned to a time series according to the forecast period. Installed capacity planning data includes the capacity of units planned to be put into operation or decommissioned in the future, renewable energy capacity expansion plans, etc., and are generally released periodically by power grid companies or energy authorities.

[0026] (3) External market environment data: This describes the impact of non-power system internal factors on electricity prices and loads, including after-hours order book bid and ask prices, transaction information, meteorological data, fuel prices, etc. The order book records the bidding behavior of power generators and power sellers in the market, usually including buyer bid sequences, seller bid sequences, order volume for each bid, actual transaction status, and transaction time. This data is published by the power trading center after the end of each trading day. In subsequent forecasting models, the buyer and seller sequences can be input into the deep learning model in chronological order. Meteorological factors directly affect load and renewable energy output, including temperature, humidity, wind speed, wind direction, solar irradiance, etc. Data sources include national meteorological departments or commercial meteorological APIs, such as meteorological interfaces commonly used by wind farms and photovoltaic power plants. Gas prices, coal prices, carbon emission costs, etc., directly affect the marginal cost of power generation and are widely used exogenous variables of electricity prices. Fuel prices can be obtained from fuel trading platforms, energy industry databases, or price announcements from regulatory authorities. After data collection, it can be aligned by day or hour to form a unified time axis with other forecast inputs, improving forecast stability.

[0027] After collecting the above data, conventional normalization or standard deviation standardization methods are used to transform data of different dimensions into the same scale, making it directly processable by downstream prediction models. For example, to ensure the consistency of input data, mean-variance standardization can be performed along the variable dimension, which is suitable for continuous variables such as electricity price, load, and output; normalization by maximum and minimum values ​​is suitable for variables of different dimensions but with clear ranges, such as wind speed and radiation intensity; and logarithmic transformation is performed on data with large dimensions, such as order books, which helps to reduce the impact of extreme values.

[0028] S2: Using the multi-timescale data as input, perform hierarchical prediction to generate prediction results for key power system operation parameters with confidence intervals, wherein the key power system operation parameters include electricity price, load and renewable energy output.

[0029] After data standardization, these data are used as inputs to perform multi-path hierarchical forecasting of key parameters for power system operation (electricity price, load, and renewable energy output). Hierarchical forecasting can be understood as "the parallel or combined use of multiple forecasting models with different mechanisms and structures." The forecast results obtained by each forecasting model are key inputs for subsequent feature extraction and decision optimization. Therefore, a confidence interval needs to be output. The confidence interval reflects the forecast uncertainty and makes the decision more robust.

[0030] The overall approach to tiered forecasting is as follows: based on multi-timescale data, multiple forecast paths are constructed. Each path generates future forecasts for electricity prices, load, and renewable energy output based on different data structures and algorithms. These results are then integrated and calibrated with confidence intervals to obtain forecasts of key power system operation parameters with upper and lower limits. Four forecast paths exist concurrently, and the system can choose one or more combinations to execute based on scenario requirements. The output of tiered forecasting includes: forecast point values ​​for multiple future time periods, forecast paths for multiple future time periods, quantile forecasts for multiple future time periods, joint forecasts for multiple regions, and forecast results with upper and lower limits and confidence intervals.

[0031] The predicted paths include: Prediction path a. Conditional generation model: based on historical electricity prices Historical load Historical renewable energy output External market environment data The eigenvectors formed For conditional input, it is represented as:

[0032] Where t represents time, and random noise is sampled from a standard normal distribution to generate diverse future paths. Where I is the identity matrix, the feature vector and noise are simultaneously input into the conditional generative model (e.g., conditional GAN ​​or diffusion model) to generate predicted paths for the next H time periods, denoted as:

[0033] in For conditional generation models, This is a predicted path sequence for multiple future time periods; To ensure that the generated paths closely approximate the real data distribution while also possessing appropriate diversity, paths are trained or selected based on a multidimensional scoring function—the Energy Score. For the generated path The evaluation is as follows:

[0034] Where E represents the expected value. Indicates the actual path, This represents another generated path. The smaller the energy score, the higher the quality of the generated path. The filtered output is the prediction path for key parameters of power system operation in multiple future time periods.

[0035] Prediction path b. End-to-end probability prediction based on order book sequence: Construct a buyer sequence from the buyer quotes, buyer order volumes, and transaction information arranged in ascending order of time in the after-hours order book. The seller information is used to construct a seller sequence. ,in , Quotations for the buyer and seller are respectively. , These represent the order quantities for buyers and sellers, respectively. , This represents the relative time difference from the delivery time. The buyer and seller sequences are input into an end-to-end neural network (such as a Transformer or quantile regression network), which outputs a set of quantile predictions. For example, 5%, 50%, 95%. Or complete probability distribution parameters (such as the mean and variance of a Gaussian distribution). If quantile regression is used, then for the target quantile... Using the quantile loss function ,in For quantile loss function, These are the predicted quantile values; to ensure that the quantile outputs are monotonically increasing, a non-negative residual structure is used to decouple the crossover risk between adjacent quantiles: This ensures that the predicted values ​​of higher-order quantiles are not less than the predicted values ​​of lower-order quantiles.

[0036] Predicted path c. Multi-region joint prediction based on graph neural network: Each region of the power system is regarded as a node in a graph, and the transmission lines between regions form edges. Define the node set. Adjacency matrix The distance between regions can be calculated based on the transmission capacity or line reactance. The historical electricity price, load, processing capacity, and external environment data of each node constitute the node's feature vector. A graph attention network (GAT) is used to perform weighted aggregation of node features: ,in For nodes i Updated features For the features of node j, Let i be a neighboring node. This is the weight matrix. For activation function, Attention weights are determined by the correlation of features between nodes and graph distance, and the final output is... This represents the joint prediction sequence of regions 1 to n at future times; Prediction path d. Fusion prediction of mechanistic model and deep ensemble model: The mechanistic model is based on the supply and demand balance of the system. The clearing price is obtained by considering unit ramp-up constraints, line power flow constraints, and cost functions. As an intermediate variable, it forms a joint input with historical load, electricity price, and external market environment data. The input is a deep neural network ensemble model, which contains multiple sub-networks. Through a parallel structure, it captures nonlinear features at different scales and outputs a fused prediction result. , where F represents the integrated network.

[0037] By integrating the prediction results output from different modeling structures, and combining conformal prediction with confidence interval calibration based on historical residual distribution, confidence interval prediction results including upper and lower limits are generated for the key operating parameters of the power system.

[0038] Using the conformal prediction method, without relying on specific distribution assumptions, the initial confidence interval is adjusted back on the validation sample set. If the default ratio corresponding to the target confidence level does not meet the statistical significance requirement, the interval width is adaptively increased or decreased to generate the final confidence interval that satisfies the distribution invariance assumption. The final confidence interval coverage is verified by multiple models, and abnormal prediction paths are eliminated or their weights are reallocated to form the integrated confidence interval prediction result.

[0039] The predicted values ​​from different prediction paths are uniformly input into the integration module to calculate the residuals. Initial confidence intervals are constructed by taking the upper and lower quantiles from the residual distribution. Calculate the inconsistency score for each sample. Take the score that satisfies the confidence level. The final interval is obtained. Calculate the bias score for different prediction paths. If the threshold is exceeded, the element is removed or its weight is reduced to obtain a stable final confidence interval; where This represents the set of predicted values ​​output by different prediction paths (or sub-models within an ensemble module) under the same prediction time / same sample conditions. For a given path, the predicted value, For a moment The predicted value, This represents the median.

[0040] S3: Combine the prediction results of the key parameters of the power system operation with the multi-timescale data to extract features, construct a multi-objective evaluation sample set to quantify the conflict intensity between decision objectives, and use the analytic hierarchy process to establish a hierarchical structure of multiple decision objectives, calculate the weight coefficient of each decision objective, the decision objectives include market entity profit objectives, price fluctuation stability objectives, operation risk control objectives and clean energy consumption objectives.

[0041] The profit objectives of the market participants include cost minimization and profit maximization. The price volatility stability targets include electricity price variance and the probability of peak occurrence; The operational risk control objectives include reserve rate and load forecast deviation; The clean energy consumption targets include the curtailment rate of wind and solar power output.

[0042] Obtain available forecasts of key power system operating parameters. Due to the selectivity of forecast paths, only one or more forecasts may be obtained under different operating scenarios: for example, only electricity price forecasts may be obtained, or both electricity price and load forecasts may be obtained simultaneously. Regardless of which forecasts are obtained, the data should be organized as follows: At any time , obtain: Forecast values: such as electricity price forecasts If it exists, then obtain the upper and lower limits of the confidence interval. , ; Simultaneously record short-term operating indicators (load, output, etc.); Synchronously record long-term planning parameters; Simultaneously record external market environment data (order book, weather data, etc.); Concatenate the above content in chronological order to form a feature vector. ,in Indicates time For the first Predicted point values ​​for key parameters (e.g., point predictions of electricity price / load / renewable energy output, etc.); By combining input data from all time scales, a comprehensive data point is formed, including predicted values, prediction uncertainties, and market environment information, providing a unified input basis for target evaluation and conflict quantification.

[0043] Based on the predicted values ​​and synchronous data contained in the feature vectors, the evaluation values ​​of the four types of decision objectives are calculated one by one: (1) Profit objectives of market entities

[0044] Revenue is estimated based on the cost of purchasing and selling electricity: ,in For a moment Electricity price forecast, For a moment The load forecast value; if the load forecast is unavailable at this time, it is simplified to Because electricity prices already have a fundamental impact on the profits of market participants; (2) Price volatility stability target

[0045] Evaluating volatility based on confidence interval width: If only quantile prediction is available, then quantile difference should be used instead. ,in , These represent the electricity price at time [time]. The predicted values ​​of the 95th and 5th percentiles; (3) Operational risk control objectives

[0046] Prioritize load tracking deviation: If load forecasting is unavailable, then the reserve capacity rate deviation will be used. ,in This refers to the historical load baseline value (such as historical observations for the same period, rolling window average, or the value of the previous cycle). For a moment The reserve capacity ratio (usually a ratio of available reserve capacity to load or peak load). Target reserve level / target reserve ratio (and (same dimensions) (4) Clean energy consumption target

[0047] Calculated based on renewable energy utilization rate: If no renewable energy forecast is available, the default estimate from the dispatch system will be used; whereby... For a moment The projected output of renewable energy To maximize output; The feature vector at each time step , No. Evaluation value of each decision objective ( The sample items are composed of the corresponding confidence intervals. The T time points constitute the multi-objective evaluation sample set. This is the source of all the information needed to calculate the intensity of the conflict; For any two decision objectives i and j, extract T evaluation values ​​for each to form a sequence, and calculate the correlation coefficient according to the formula: ; Calculate the correlation conflict strength The formula is: ; Evaluation values ​​themselves cannot reflect uncertainty, while confidence intervals can reflect the robustness of predictions. Therefore, to further quantify interval conflict, data containing the upper and lower limits of confidence intervals are extracted from the multi-objective evaluation sample set, and the interval overlap is calculated using the following formula: ,in Indicate decision objectives With decision-making objectives At any moment The overlap of confidence intervals, , Representing decision objectives With decision-making objectives At any moment The upper limit of the prediction confidence interval, , Representing decision objectives With decision-making objectives At any moment The lower bound of the prediction confidence interval; Based on confidence interval overlap Calculate the interval conflict intensity The formula is: ,in The length of the time window; The correlation conflict strength Intensity of Interval Conflict Weighted synthesis is performed to determine the conflict intensity among the decision objectives. : , The preset weighting factor is used to determine the proportion of correlation and uncertainty in the degree of conflict.

[0048] Based on the conflict intensity among multiple decision objectives Construct a judgment matrix to compare the relative importance of four types of decision objectives, setting the elements of the judgment matrix for any two different decision objectives as follows: ,in Indicate decision objectives Relative to decision objectives The importance of This represents the scale factor of the judgment matrix, used to map conflict intensity to the 1–9 scale interval of the analytic hierarchy process, and according to... Set the inverse elements of the judgment matrix; Perform eigenvector calculations on the matrix to obtain the eigenvector corresponding to the largest eigenvalue. The feature vector is then normalized. That is, the first The weighting coefficients of each decision objective; To ensure the validity of the judgment matrix, a consistency check of AHP needs to be performed, including calculating the consistency index. and consistency ratio ,when The judgment matrix is ​​considered to satisfy the consistency requirement; if Then, adaptive adjustment will be performed. Adjust the elements of the judgment matrix, where This represents the consistency adjustment factor, which can range from 0.01 to 0.05. The adjusted judgment matrix is ​​then recalculated for eigenvectors and subjected to consistency checks until the consistency is satisfied. .

[0049] To ensure that the weights reflect the uncertainty of prediction, the weights are further biased based on the confidence interval width; for the decision objective Extract the upper and lower limits of the prediction confidence interval from the multi-objective evaluation sample set, and calculate the average prediction uncertainty. : This value is then used to adjust the bias of the weighting coefficients: ,in These are the weighting coefficients after uncertainty bias. These are the standardized weight coefficients. This adjustment method automatically reduces the weight of objectives with higher uncertainty during comprehensive optimization, making the final weights more robust.

[0050] S4: Construct a comprehensive fitness function based on the weighted combination of multiple decision objectives according to the weight coefficients. Use market clearing decision variables or scheduling decision variables as optimization variables. Use a genetic algorithm to perform multi-objective optimization. When the fluctuation range of electricity price or the deviation of operating indicators exceeds the preset threshold, adaptively adjust the weight coefficients. Re-execute optimization based on the adjusted weight coefficients to generate a set of candidate decision schemes.

[0051] Comprehensive fitness function Represented as: ; Using market clearing decision variables or scheduling decision variables as optimization variables, a genetic algorithm is employed to perform multi-objective optimization, including: by As the basis for optimization in genetic algorithms, among which This represents a decision vector composed of market clearing decision variables or scheduling decision variables. This represents a feasibility penalty calculated based on power balance deviation, transmission line power flow exceeding limits, and insufficient reserve capacity. Indicates the penalty coefficient; Selection, crossover, and mutation operations are performed on the initial population. When electricity price fluctuations or operational indicator deviations exceed preset thresholds, [further actions are taken]. Adjust the mutation probability, where This represents the initial mutation probability. This represents the adjusted mutation probability. Indicates a variable-sensitive factor. A measure of uncertainty in market conditions; When an increase in the prediction uncertainty of key power system operating parameters is detected, based on Adjust the weighting coefficients of the decision-making objectives, where Indicate decision objectives The basic weighting coefficients, This represents the adjusted weighting coefficient. This represents the sensitivity coefficient for weight adjustment. Indicates the first The uncertainty of the predicted value corresponds to each decision objective; During the iterative process of the genetic algorithm, based on Generate a local neighborhood search solution, where This represents the current optimal decision vector. This represents the local disturbance factor. This represents a random vector sampled within the interval [-1, 1]. in accordance with Calculate the degree of difference between different candidate solutions and remove candidate solutions with a degree of difference less than a threshold to obtain a diverse set of candidate decision solutions.

[0052] S5: Under multi-timescale operation scenarios, using the prediction results of the key parameters of the power system operation as input, the candidate decision scheme set is simulated to generate virtual operation paths. Information fusion is performed on the virtual operation paths that meet the operation constraints and stability standards. The final decision scheme is determined through feasibility verification, and the actual operation deviation is fed back to iteratively update the hierarchical prediction, weight coefficients, and comprehensive fitness function.

[0053] To evaluate the performance of each solution under different future market conditions, multi-timescale operating scenarios are constructed, including: Short-term scenario: Inputs include hourly forecasts of electricity prices, load, and renewable energy output; Medium-term scenario: Inputs are average load and renewable energy output as predicted on a daily or weekly basis; Long-term scenario: Inputs are monthly or quarterly forecasts of load change trends and changes in installed capacity planning.

[0054] Each scenario uses the prediction results of key parameters obtained in the aforementioned prediction process as exogenous input conditions.

[0055] In each constructed scenario, the scheduling instructions or market clearing decisions corresponding to the candidate decision schemes are input into the physical constraint model. The physical constraint model is a well-known model in power system operation, used to perform feasibility verification or constraint optimization calculations on candidate decision schemes at various times to determine whether they meet the physical feasibility requirements for power system operation, including: Power balance model; Unit start-up and shutdown model; Unit ramp-up capability model; Power flow model of transmission lines; Backup capacity configuration model; Renewable energy adjustable output model; Each model is used to constrain the power balance relationship, unit operating state transition, output change rate, line transmission capacity, system reserve level, and the adjustment and reduction behavior of renewable energy within the maximum available output range at each time point. It also couples the operating states of adjacent time points in the time dimension to form a virtual operating path that reflects the dynamic changes of the system.

[0056] For each virtual running path, perform the following checks sequentially in chronological order: Power balance check: Check whether the difference between the total power generation output and the load demand is within the set allowable deviation range.

[0057] Power flow verification of transmission lines: Calculate the power flow for each transmission line and determine whether it exceeds the line's rated transmission capacity.

[0058] Unit status verification: Determine whether the unit's start-up and shutdown status violates requirements such as ramp-up limits, minimum start-up time, or minimum shutdown time.

[0059] Backup capacity verification: Confirm whether the system's backup capacity meets the proportional thresholds for each level of backup requirements.

[0060] Renewable energy utilization verification: Determine whether renewable energy can be reasonably absorbed in each period and whether there is any unnecessary curtailment of wind and solar power.

[0061] If any indicator fails to meet the above operational constraints, the corresponding virtual operational path is deemed infeasible and is directly eliminated.

[0062] The verified virtual operating path retains its operating indicators in time sequence, such as changes in power generation, changes in line power flow, renewable energy utilization rate, and risk-related indicators.

[0063] The final result is a dataset containing only feasible paths, which is used for subsequent information fusion and final solution determination.

[0064] The following steps are taken to merge the set of verified virtual running paths: The indicator sequences for each path at different time scales were extracted, including: market entity income change sequence, electricity price fluctuation change sequence, operational risk level sequence, and clean energy consumption ratio sequence.

[0065] Time alignment for short-term, medium-term, and long-term indicators: Use a unified time index to interpolate or aggregate data from different time scales to a unified reference time axis.

[0066] Weighted fusion of indicators across time scales: Based on the importance of different time scales to decision-making, data from multiple time scales are fused to form a unified comprehensive path description.

[0067] The comprehensive path description reflects the overall performance of the solution under different operating conditions in the future.

[0068] After integration is completed, a benefit balance index for the integration path is calculated. This index is used to determine whether the benefits, costs, and risks among the participating entities have reached the expected coordination range.

[0069] The calculation method includes the following: Entity revenue difference analysis: Based on the market clearing results or dispatch instructions of candidate schemes at each time, determine the revenue evaluation value of each market entity in the corresponding time period, and use the average revenue of all entities or the average revenue of similar entities (e.g., power generation entities / power consumption entities / regional entities) as the comparison benchmark to calculate the degree of deviation of each entity's revenue from the comparison benchmark, which is used to characterize the balance of revenue distribution. Risk-bearing difference analysis: The risk-bearing level of each entity is compared based on the operational risk-related indicators corresponding to the candidate schemes. The operational risk-related indicators include at least: output fluctuation risk indicators, line congestion risk indicators, or reserve shortage risk indicators. The comparison can be measured by the degree of deviation from the average risk level of all entities or the preset reference risk level to identify whether the risk-bearing is significantly unbalanced among the entities. Analysis of the distribution of renewable energy consumption responsibility: Based on the amount or proportion of renewable energy consumption undertaken by each entity at each time under the candidate scheme, the distribution of clean energy consumption responsibility among the entities is determined; when the consumption ratio of any entity exceeds the preset deviation threshold relative to the average consumption ratio of all entities, the average consumption ratio of similar entities, or the preset reference level, it is determined that there is a situation where the clean energy consumption responsibility is biased towards that entity.

[0070] Based on the results of the analysis of differences in main entity benefits, risk-bearing differences, and the distribution of responsibility for resource absorption, a benefit balance index is constructed to comprehensively reflect the overall equilibrium of candidate schemes in terms of benefit distribution, risk-bearing, and responsibility for resource absorption. When the benefit balance index falls below a preset threshold, the resource allocation parameters of the candidate schemes are reconfigured. The preset threshold can be determined based on historical operational data statistics, policy constraints, or system operation experience.

[0071] After reconfiguration is triggered, at least one of the following parameters shall be adjusted: renewable energy adjustment amount, clean price impact factor, standby configuration parameter or weight coefficient related to resource allocation; the candidate scheme set shall be updated accordingly, and the optimization steps related to the parameters shall be re-executed to obtain the updated candidate schemes and their corresponding target evaluation results.

[0072] The above process of "analysis-judgment-reconfiguration-update-local recalculation" can be repeated until the benefit balance index meets the preset requirements or reaches the preset iteration limit, and finally outputs the determined global solution.

[0073] After the final plan is determined, the operational deviations and changes in operational indicators after the decision is implemented will be used as feedback information for: Update the hierarchical prediction model: When the prediction error increases significantly in certain periods, update the model parameters for the corresponding periods based on actual operating data to improve the accuracy of subsequent predictions; Update weight coefficients: Based on the updated operational data and target achievement status, recalculate the relevant weights of each target or entity to reflect the changes in the importance of different targets under the current operational conditions; Update the overall fitness construction strategy: Adjust the sensitivity of the overall fitness function to changes in key indicators or the strength of constraint penalties based on feedback information, so that the overall fitness function is better matched to future changes in operating characteristics, thereby improving the effectiveness of subsequent decision-making and screening.

[0074] This cyclical update mechanism enables the model to remain adaptable to the ever-changing market environment, forming a dynamically evolving intelligent decision-making system.

[0075] The beneficial effects of this invention are that, compared with existing technologies, it provides an intelligent decision-making method and system for the power market based on hierarchical model fusion. By organically combining multi-timescale prediction models, conflict intensity quantification methods, hierarchical analysis weight calculation, adaptive genetic algorithms, multi-scenario operation simulation, and dynamic feedback update mechanisms, a continuously learning intelligent decision-making system for the power market is constructed. In a power market environment with significantly increased uncertainty, it achieves high-precision prediction of key power system operation parameters, quantitative trade-offs among multiple objectives, stability verification across time scales, and continuous correction of operational deviations, making the decision-making process data-driven, globally coordinated, and dynamically adaptable. It fully captures short-term operational changes and long-term planning trends, ensuring data consistency and comparability in subsequent prediction, evaluation, and optimization processes, and improving the accuracy and completeness of decision input data. By introducing multiple types of prediction models and generating confidence intervals including upper and lower limits, it enables the prediction of electricity prices, load, and renewable energy output to characterize uncertainty, providing a solid data foundation for multi-objective conflict quantification and risk control. By unifying forecast results with operational data to form a quantifiable dataset, the interplay between revenue, volatility, risk, and clean energy consumption can be comprehensively characterized, enabling precise data-driven characterization of multi-objective conflict intensity. Constructing a judgment matrix based on conflict intensity and operational preferences improves the objectivity and adaptability of weight determination, allowing for dynamic adjustment of the importance of each decision objective under changing market conditions. A unified optimization function is formed by weighted combination of multiple objectives, enabling coordinated trade-offs among market clearing or scheduling decision variables and timely adjustment of optimization strategies during market fluctuations through adaptive mechanisms. Constructing multi-level operational scenarios (short-term, medium-term, and long-term) allows for comprehensive evaluation of the feasibility of candidate schemes under different future conditions, ensuring the stability of the final decision scheme across time scales. Ensuring that power balance, line flow, unit status, and reserve capacity all meet physical operational requirements rigorously verifies the physical feasibility of the decision scheme. Unified processing of schemes validated through multiple scenarios enables coordinated optimization of multiple indicators such as revenue, risk, and clean energy consumption for various stakeholders, improving the overall fairness and rationality of the decision scheme. By continuously updating the prediction module, weighting system, and fitness function using actual operational data, the decision-making system possesses continuous learning capabilities, enabling it to maintain dynamic adaptability and stability in complex market environments over the long term. With the technical framework of this invention, the electricity market can more effectively cope with the complex changes brought about by the high proportion of renewable energy integration, improve market clearing efficiency, reduce operational risks, promote the consumption of clean energy, and lay the foundation for future large-scale flexible dispatch and smart electricity trading, demonstrating promising engineering application prospects and widespread value.

[0076] This invention can be a system, method, and / or computer program product. This invention also discloses a power market information intelligent decision-making system based on hierarchical model fusion, which is based on the aforementioned hierarchical model fusion-based intelligent decision-making method for power market information, comprising: The data acquisition module is used to collect multi-timescale data from the electricity market, including short-term operating indicators, long-term planning parameters, and external market environment data, and to standardize the collected data. The prediction module is used to perform hierarchical prediction with the multi-timescale data as input, and generate prediction results of key parameters of power system operation with confidence intervals, wherein the key parameters of power system operation include electricity price, load and renewable energy output; The feature extraction module is used to combine the prediction results of the key parameters of the power system operation with the multi-timescale data to extract features, construct a multi-objective evaluation sample set, quantify the conflict intensity between decision objectives, and use the analytic hierarchy process to establish a hierarchical structure of multiple decision objectives and calculate the weight coefficient of each decision objective. The decision objectives include market entity revenue objectives, price fluctuation stability objectives, operational risk control objectives, and clean energy consumption objectives. The optimization module is used to construct a comprehensive fitness function formed by a weighted combination of multiple decision objectives based on the weight coefficients. Market clearing decision variables or scheduling decision variables are used as optimization variables. A genetic algorithm is used to perform multi-objective optimization. When the fluctuation range of electricity price or the deviation of operating indicators exceeds a preset threshold, the weight coefficients are adaptively adjusted. The optimization is re-executed based on the adjusted weight coefficients to generate a set of candidate decision schemes. The decision-making module is used to simulate the candidate decision scheme set under multi-timescale operation scenarios, taking the prediction results of the key parameters of the power system operation as input, generate virtual operation paths, perform information fusion on the virtual operation paths that meet the operation constraints and stability standards, determine the final decision scheme through feasibility verification, and use the actual operation deviation feedback to iteratively update the hierarchical prediction, weight coefficients and comprehensive fitness function.

[0077] Based on the spirit of this invention, those skilled in the art will readily conceive of a computer program product derived from the aforementioned intelligent decision-making method for electricity market information based on hierarchical model fusion. The computer program product may include a computer-readable storage medium on which computer-readable program instructions are loaded to enable a processor to implement various aspects of this disclosure. That is, this application also includes a terminal comprising a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the aforementioned intelligent decision-making method for electricity market information based on hierarchical model fusion.

[0078] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0079] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0080] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A smart decision-making method for electricity market information based on hierarchical model fusion, characterized in that, The process includes the following steps: Collect multi-timescale data of the electricity market, including short-term operating indicators, long-term planning parameters and external market environment data, and standardize the collected data. Using the multi-timescale data as input, hierarchical prediction is performed to generate prediction results for key parameters of power system operation with confidence intervals, wherein the key parameters of power system operation include electricity price, load and renewable energy output; The prediction results of the key parameters of power system operation are combined with the multi-timescale data to extract features and construct a multi-objective evaluation sample set to quantify the conflict intensity between decision objectives. The hierarchical structure of multiple decision objectives is established by using the analytic hierarchy process (AHP) and the weight coefficients of each decision objective are calculated. The decision objectives include market entity revenue objectives, price fluctuation stability objectives, operational risk control objectives, and clean energy consumption objectives. Based on the weight coefficients, a comprehensive fitness function is constructed by weighting multiple decision objectives. Market clearing decision variables or scheduling decision variables are used as optimization variables. A genetic algorithm is used to perform multi-objective optimization. When the fluctuation range of electricity price or the deviation of operating indicators exceeds the preset threshold, the weight coefficients are adaptively adjusted. Based on the adjusted weight coefficients, the optimization is re-executed to generate a set of candidate decision schemes. Under multi-timescale operation scenarios, the prediction results of the key parameters of the power system operation are used as input to simulate the candidate decision scheme set, generate virtual operation paths, perform information fusion on the virtual operation paths that meet the operation constraints and stability standards, determine the final decision scheme through feasibility verification, and use the actual operation deviation feedback to iteratively update the hierarchical prediction, weight coefficients and comprehensive fitness function.

2. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 1, characterized in that, The short-term operating indicators include time-of-use electricity prices, load, and renewable energy output; the long-term planning parameters include medium- and long-term load forecast data and installed capacity planning data; and the external market environment data include after-hours order book quotes, transaction information, meteorological data, and fuel prices.

3. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 2, characterized in that, The hierarchical prediction is performed using the multi-timescale data as input, employing one or more of the following prediction paths: a. Using the feature vector composed of historical electricity price, load, renewable energy output and external market environment data as conditional input, the model generates path samples of electricity price, load and renewable energy output for multiple consecutive future periods through conditional generation. Based on the multidimensional scoring function, the model is trained or the samples are screened to output the prediction path of key parameters of power system operation for multiple future periods. b. Construct time series inputs from the buy and sell quotes and transaction information in the after-hours order book according to the buyer sequence and the seller sequence respectively, and use an end-to-end neural network to output the probability distribution or predicted values ​​of the key parameters of the power system operation in multiple future time periods; c. Represent the topology between power grid regions as a graph structure, fuse historical data from multiple regions based on the graph distance between regions, and output the joint prediction sequence of key power system operation parameters in each region for multiple future time periods through a graph neural network; d. The clearing price calculated by the physical constraint-based mechanism model is used as an intermediate variable, and combined with historical load, electricity price and external market environment data to form a joint input. This input is then fed into a deep neural network ensemble model, and the fused prediction result is output. By integrating the prediction results output from different modeling structures, and combining conformal prediction with confidence interval calibration based on historical residual distribution, confidence interval prediction results including upper and lower limits are generated for the key operating parameters of the power system.

4. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 3, characterized in that, In the hierarchical prediction, the conditional generation model uses historical feature vectors and random noise as inputs to conditionally model the generation network, and uses energy scores as a multidimensional loss function to train the model parameters, so that the generated key parameter paths of power system operation simultaneously satisfy distribution consistency and sample diversity. The end-to-end neural network uses quantile regression to output multiple quantile prediction values ​​for multiple future time periods, and decouples the cross risk between adjacent quantiles through non-negative residuals to achieve a stable probability characterization of key parameters for future power system operation. Graph neural networks employ graph attention mechanisms to weighted aggregate regional historical features, and integrate structural priors constructed from graph adjacency matrices and graph distance functions to guide joint modeling and cross-regional prediction of load, electricity price, and power output indicators between regions. The deep neural network integrated model is input by a feature vector composed of clearing price, historical load, electricity price and external market environment data output by the mechanistic model. It is then nonlinearly mapped through a multi-sub-network parallel structure to output the fused predicted values ​​of key power system operation parameters.

5. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 4, characterized in that, The method of combining conformal prediction with confidence interval calibration based on historical residual distribution to generate confidence interval prediction results including upper and lower limits for the key operating parameters of the power system includes: For the key power system operation parameters output from multiple prediction paths, a conditional distribution set is constructed based on historical observation residuals, and the upper and lower limits of the predicted values ​​are corrected at the target confidence level to obtain the initial confidence interval. Using the conformal prediction method, without relying on specific distribution assumptions, the initial confidence interval is adjusted back on the validation sample set. If the default ratio corresponding to the target confidence level does not meet the statistical significance requirement, the interval width is adaptively increased or decreased to generate the final confidence interval that satisfies the distribution invariance assumption. The final confidence interval coverage is verified by multiple models, and abnormal prediction paths are eliminated or their weights are reallocated to form the integrated confidence interval prediction result.

6. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 5, characterized in that, The construction of a multi-objective evaluation sample set to quantify the conflict intensity between decision objectives further includes: The available electricity price forecasts, load forecasts, or renewable energy output forecasts, along with the upper and lower limits of their confidence intervals, are synchronized with short-term operating indicators, long-term planning parameters, and external market environment data to form a sample data containing multi-dimensional features. Calculate the evaluation values ​​for the four decision objectives based on the sample data. , , , The evaluation value and the corresponding upper and lower limits of the confidence interval are then incorporated into the multi-objective evaluation sample set. The evaluation value sequence of each decision objective is extracted from the multi-objective evaluation sample set, and the correlation conflict strength is calculated. The formula is: ,in, Indicate decision objectives With decision-making objectives The Pearson correlation coefficient; Data containing the upper and lower limits of confidence intervals are extracted from the multi-objective evaluation sample set, and the interval overlap is calculated using the following formula: ,in Indicate decision objectives With decision-making objectives At any moment The overlap of confidence intervals, , Representing decision objectives With decision-making objectives At any moment The upper limit of the prediction confidence interval, , Representing decision objectives With decision-making objectives At any moment The lower bound of the prediction confidence interval; Based on confidence interval overlap Calculate the interval conflict intensity The formula is: ,in The length of the time window; The correlation conflict strength Intensity of Interval Conflict Weighted synthesis is performed to determine the conflict intensity among the decision objectives. : , This is a preset weighting factor.

7. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 6, characterized in that, The method of establishing a hierarchical structure of multiple decision objectives using the analytic hierarchy process (AHP) and calculating the weight coefficients of each decision objective specifically includes: Based on the conflict intensity among multiple decision objectives Construct a judgment matrix to compare the relative importance of four types of decision objectives, setting the elements of the judgment matrix for any two different decision objectives as follows: ,in Indicate decision objectives Relative to decision objectives The importance of This indicates determining the matrix scaling factor, and according to... Set the inverse elements of the judgment matrix; Perform eigenvector calculation on the matrix to obtain the eigenvector corresponding to the largest eigenvalue. The feature vector is then normalized. As the first The weight coefficients of each decision objective; when the judgment matrix does not meet the consistency requirement, Adjust the elements of the judgment matrix, where This represents the consistency adjustment factor, and the above eigenvector calculation steps are repeated to obtain the weight coefficients that meet the consistency requirements; The upper and lower limits of the confidence interval are calculated based on the records in the multi-objective evaluation sample set. The average forecast uncertainty of each decision objective The formula is: And calculate the weight coefficient of the decision objective after uncertainty bias, using the following formula: ,in These are the weighting coefficients after uncertainty bias. These are the weighting coefficients obtained after standardization.

8. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 7, characterized in that, The comprehensive fitness function is expressed as follows: ; Using market clearing decision variables or scheduling decision variables as optimization variables, a genetic algorithm is employed to perform multi-objective optimization, including: by As the basis for optimization in genetic algorithms, among which This represents a decision vector composed of market clearing decision variables or scheduling decision variables. This represents a feasibility penalty calculated based on power balance deviation, transmission line power flow exceeding limits, and insufficient reserve capacity. Indicates the penalty coefficient; Selection, crossover, and mutation operations are performed on the initial population. When electricity price fluctuations or operational indicator deviations exceed preset thresholds, [further actions are taken]. Adjust the mutation probability, where This represents the initial mutation probability. This represents the adjusted mutation probability. Indicates a variable-sensitive factor. A measure of uncertainty in market conditions; When an increase in the prediction uncertainty of key power system operating parameters is detected, based on Adjust the weighting coefficients of the decision-making objectives, where Indicate decision objectives The basic weighting coefficients, This represents the adjusted weighting coefficient. This represents the sensitivity coefficient for weight adjustment. Indicates the first The uncertainty of the predicted value corresponds to each decision objective; During the iterative process of the genetic algorithm, based on Generate a local neighborhood search solution, where This represents the current optimal decision vector. This represents the local disturbance factor. This represents a random vector sampled within the interval [-1, 1]. in accordance with Calculate the degree of difference between different candidate solutions and remove candidate solutions with a degree of difference less than a threshold to obtain a diverse set of candidate decision solutions.

9. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 8, characterized in that, The scenario simulation of the candidate decision set includes: Using the prediction results of key parameters of power system operation as exogenous input, power market operation scenarios are constructed at multiple short-term, medium-term and long-term time scales. Under each scenario, a physical model containing power flow constraints, unit start-up and shutdown constraints, reserve capacity constraints and transmission channel capacity constraints is invoked to generate virtual operation paths based on the candidate decision schemes. In the virtual operation path, the feasibility of power balance deviation, transmission line power flow change, unit status evolution and renewable energy utilization at each time scale is checked, and virtual operation paths that do not meet any operation constraints or stability check conditions are eliminated. The changes in operational indicators of the verified virtual operation paths are recorded in chronological order to form a set of filtered virtual operation paths for subsequent information fusion.

10. The intelligent decision-making method for electricity market information based on hierarchical model fusion according to claim 9, characterized in that, The determination of the final solution includes: Information fusion across time scales is performed on the selected set of virtual operation paths. Indicators reflecting changes in market participants' returns, price fluctuations, operational risks, and clean energy consumption at different time scales are processed in a unified manner to form a comprehensive path description that reflects the degree of coordination among multiple objectives. The benefit balance index is calculated based on the comprehensive path description. When the benefit balance index is lower than the preset threshold, the resource allocation parameters are readjusted and the candidate decision scheme set is updated to generate a scheme that meets the benefit balance requirements. After obtaining a final solution that meets the requirements of balancing interests, a risk level assessment is performed. The operational deviations, operational stability indicators, and renewable energy utilization corresponding to the final solution are used as feedback information to update the hierarchical prediction model, target weight coefficients, and comprehensive fitness construction strategy, so as to form a cyclical dynamic adaptation mechanism.

11. A power market information intelligent decision-making system based on hierarchical model fusion, characterized in that, The term includes: The data acquisition module is used to collect multi-timescale data from the electricity market, including short-term operating indicators, long-term planning parameters, and external market environment data, and to standardize the collected data. The prediction module is used to perform hierarchical prediction with the multi-timescale data as input, and generate prediction results of key parameters of power system operation with confidence intervals, wherein the key parameters of power system operation include electricity price, load and renewable energy output; The feature extraction module is used to combine the prediction results of the key parameters of the power system operation with the multi-timescale data to extract features, construct a multi-objective evaluation sample set, quantify the conflict intensity between decision objectives, and use the analytic hierarchy process to establish a hierarchical structure of multiple decision objectives and calculate the weight coefficient of each decision objective. The decision objectives include market entity revenue objectives, price fluctuation stability objectives, operational risk control objectives, and clean energy consumption objectives. The optimization module is used to construct a comprehensive fitness function formed by a weighted combination of multiple decision objectives based on the weight coefficients. Market clearing decision variables or scheduling decision variables are used as optimization variables. A genetic algorithm is used to perform multi-objective optimization. When the fluctuation range of electricity price or the deviation of operating indicators exceeds a preset threshold, the weight coefficients are adaptively adjusted. The optimization is re-executed based on the adjusted weight coefficients to generate a set of candidate decision schemes. The decision-making module is used to simulate the candidate decision scheme set under multi-timescale operation scenarios, taking the prediction results of the key parameters of the power system operation as input, generate virtual operation paths, perform information fusion on the virtual operation paths that meet the operation constraints and stability standards, determine the final decision scheme through feasibility verification, and use the actual operation deviation feedback to iteratively update the hierarchical prediction, weight coefficients and comprehensive fitness function.

12. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the intelligent decision-making method for electricity market information based on hierarchical model fusion according to any one of claims 1-10.

Images