Power transaction trend prediction and decision support system and method

Abstract

The invention discloses a power transaction trend prediction and decision support system and method, and relates to the technical field of power transaction decision, and the method comprises the steps: obtaining a spatio-temporal data set in a power system through a data collection module; modeling the data through a causal reasoning algorithm by using a space-time adaptive causal network module, and dynamically adjusting an edge weight in a causal relationship graph; based on the constructed causal atlas, a power transaction trend prediction and decision support module combines a space-time convolutional neural network to perform multi-level prediction, provides a change trend of power load and price, and generates a corresponding power transaction strategy; a prediction result is transmitted back to the causal network through a feedback and optimization module, and the causal atlas and the space-time modeling process are further optimized; and the precision of prediction and decision support is gradually improved by utilizing a multi-feedback closed-loop optimization process through an iterative optimization module. According to the invention, the decision-making quality of electricity market transaction can be effectively improved, and a more accurate and timely decision-making basis is provided for an electric power system manager.

Description

Technical Field

[0001] The present invention relates to the technical field of power trading decision-making, and particularly to a power trading trend prediction and decision-making support system and method. Background Art

[0002] The dynamic complexity and variability of the power market make the prediction and decision-making of power trading a challenging task. Existing technologies usually rely on historical data and traditional modeling methods to predict power load, price changes and their trends. However, with the continuous development of the power system and market environment, these traditional methods often have difficulty in dealing with the complexity of spatio-temporal changes, resulting in low prediction accuracy and decision-making efficiency.

[0003] In traditional power trading prediction methods, most models only rely on static data or simple regression analysis, and fail to fully consider the dynamic changes of spatio-temporal data. In addition, the variability of the power market requires in-depth understanding and modeling of the causal relationships between different variables. Although some machine learning-based algorithms, such as deep neural networks and regression models, can make a certain degree of prediction for power trading, they often ignore the influence of spatio-temporal dimensions and the complex causal relationships between variables.

[0004] Therefore, the present application proposes a power trading trend prediction and decision-making support system and method, which constructs a spatio-temporal adaptive causal network, combines multi-stage feedback closed-loop optimization, and adjusts the causal relationship map in real time to improve the prediction accuracy and decision-making efficiency of the power market. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems that existing methods often rely on static data and simple linear regression models, cannot accurately capture the dynamic changes of power load and price, resulting in large deviations in prediction results, and existing systems cannot update model parameters and decision-making strategies in real time in the face of rapid fluctuations in the power market, resulting in lagging decision-making support and inability to respond to market changes in a timely manner. A power trading trend prediction and decision-making support system and method are proposed.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A power trading trend prediction and decision-making support system, comprising: A data acquisition module: used to obtain spatio-temporal data sets in the power system, and the spatio-temporal data sets include multiple power load data, power price data, equipment status data and sensor data; A spatio-temporal adaptive causal network module: used to model the variables in the spatio-temporal data set and their mutual influence relationships through a causal inference algorithm, and the causal inference algorithm constructs causal relationships based on historical data and dynamically adjusts the weights of the edges to reflect the changes in causal relationships.

[0007] Power trading trend prediction and decision support module: used to generate multi-level prediction and decision support for the power system according to the joint modeling results of the spatio-temporal adaptive causal network and the spatio-temporal convolutional neural network; Feedback and optimization module: used to take the power trading trend prediction and equipment scheduling decision as feedback and input it into the spatio-temporal adaptive causal network map, and dynamically adjust and optimize the spatio-temporal causal map; Iterative optimization module: used to optimize the power trading trend prediction and equipment scheduling decision through multiple iterations and feedback closed-loop.

[0008] A power trading trend prediction and decision support method, using the above-mentioned power trading trend prediction and decision support system, includes: S1: Obtain the spatio-temporal data set in the power system. The spatio-temporal data set includes multiple power load data, power price data, equipment status data and sensor data. Each data item in the spatio-temporal data set has a timestamp and a spatial identifier, which are used to characterize the state changes of the power system at multiple time points and geographical locations; S2: Based on the historical data in the spatio-temporal data set, use the causal inference algorithm to determine the causal relationship between various variables in the power system, combine the spatio-temporal convolutional neural network to model the dynamic changes in the spatial dimension and time dimension, and weight the spatio-temporal data through the attention mechanism, so as to dynamically adjust the weight of the edges in the causal map according to the feedback information, generate the causal relationship map, and form a spatio-temporal adaptive causal network; S3: Use the causal map generated by the spatio-temporal adaptive causal network and the extracted spatio-temporal features to conduct trend prediction through a neural network; according to the output of the spatio-temporal data model, predict the future change trends of power load and price, and provide support for the trading strategies in the power market; combine the predicted power trading trends, and use the decision-making module of the spatio-temporal adaptive causal network to generate corresponding power trading strategies; S4: Input the power trading trend prediction and decision support generated in step S3 as feedback information into the spatio-temporal adaptive causal network to dynamically optimize the causal map and the spatio-temporal modeling process, including: taking the power trading trend prediction result and the decision support strategy as new inputs, and transmitting them into the spatio-temporal adaptive causal network map, and analyzing its impact on other power system variables through the causal inference module; based on the feedback information, dynamically adjust the causal map in the spatio-temporal adaptive causal network, update the edge weights and structures of the causal relationships, and ensure that the causal relationships can be adaptively adjusted as time evolves and the system state changes; S5: Multi-stage Feedback Closed-loop Optimization Prediction and Decision-making: Based on multiple feedbacks and optimizations, improve the accuracy of power trading trend prediction and decision-making support, including: taking the causal graph described in step S2 as the first causal relationship graph, and constructing a second causal relationship graph based on the feedback information described in step S4. After each feedback, the spatio-temporal adaptive causal network updates the causal graph by combining a spatio-temporal convolutional neural network and an adaptive attention mechanism, and retrains the prediction model. Compare the first causal relationship graph with the second causal relationship graph to identify the differences in causal relationships between the two, and adjust the weights of the corresponding edges in the causal graph based on the items that meet the predetermined statistical criteria in the differences; the feedback information is used to drive the causal inference module to dynamically adjust the edge weights and structure of the causal graph in the spatio-temporal adaptive causal network to ensure that the causal relationship can be adaptively adjusted as the state of the power system changes, optimize the modeling process of spatio-temporal data through a spatio-temporal convolutional neural network and an adaptive attention mechanism, thereby improving the prediction accuracy; after each feedback, adjust the causal graph by analyzing the feedback data, and retrain the prediction model in the spatio-temporal adaptive causal network. Through multiple feedback iterations, optimize the power trading trend prediction and decision-making support strategy until the predetermined prediction accuracy and decision-making efficiency goals are achieved.

[0009] Preferably, step S2 includes: using the Granger causality test method to determine the causal relationship between power load and power price: Use the AIC criterion to select the delay number to select the time delay p, and the selected delay is 1; The Granger test uses a regression model: test whether the past values of Y(t) can significantly predict the future values of X(t):

[0010]

[0011] where p and q are the lag orders, , , are the regression coefficients. If is significantly non-zero, it indicates that there is a Granger causal relationship between X(t) and Y(t); If the Granger test shows that the power load has a significant impact on the price, then in the causal graph, add a directed edge of "power load → power price"; Extract spatial features through 3×3 spatial convolution, and then extract features in the time dimension through 1D convolution. After each convolution operation, generate the corresponding spatio-temporal feature map; Apply the adaptive attention mechanism to calculate the attention weights based on spatial and temporal features:

[0012] Among them, is the weight score at time t and space s, is the normalized attention value; Combine the spatio-temporal convolution features with the attention weights to generate the final spatio-temporal feature map for subsequent decision support.

[0013] Preferably, the step S2 further includes: based on each timestamp and space identifier in the multi-source spatio-temporal dataset, use the inverse information entropy algorithm for causal inference to identify the potential causal relationships between variables in the power system; assign preliminary weights to each edge in the causal graph, and the weight represents the strength of the causal relationship between nodes; according to the changes in the spatio-temporal data, dynamically adjust the causal graph through a feedback mechanism so that the graph can adapt to the changing state in the power system and improve the prediction accuracy of the power trading trend.

[0014] Preferably, the spatio-temporal convolutional neural network includes: a spatial convolutional layer for extracting the spatial features of the equipment and sensor data in the power system and capturing the spatial dependence between the equipment; a temporal convolutional layer for extracting the time series features of the power load and power price variables and capturing the changing trends of various variables in the power system over time; a spatio-temporal fusion layer for fusing the features extracted by the spatial convolutional layer and the temporal convolutional layer to obtain a more comprehensive spatio-temporal feature representation for power trading trend prediction.

[0015] Preferably, the step S4 specifically includes: taking the power trading trend prediction result and the decision support strategy as new inputs and transmitting them into the spatio-temporal adaptive causal network graph, and analyzing their impacts on other power system variables through the causal inference module; based on the feedback information, dynamically adjust the causal graph in the spatio-temporal adaptive causal network and update the edge weights and structure of the causal relationship.

[0016] Preferably, the step S5 includes: in the first stage, generate a power trading trend prediction based on the initial spatio-temporal causal graph and the output of the spatio-temporal convolutional neural network, and generate decision support information according to the prediction result; in the second stage, take the power trading trend prediction result of the first stage as feedback and input it into the spatio-temporal causal graph to update the strength and structure of the causal relationship to enhance the prediction accuracy of the next step; in the third stage, optimize the prediction ability of the spatio-temporal convolutional neural network according to the feedback result and further adjust the power trading trend prediction result to form an adaptive optimization model to provide decision support.

[0017] Preferably, the dynamic adjustment process of the spatio-temporal adaptive causal network graph includes the following steps: According to the feedback information after the prediction period of the power system, the causal relationship and edge weights in the graph are adjusted in real time. The feedback information includes the actual transaction data of the power market, the load fluctuation situation, and other relevant events. The feedback information is used to optimize the edge weights and causal relationship strength in the causal graph. During the feedback adjustment process, the spatio-temporal adaptive causal network dynamically adjusts the structure of the causal graph by analyzing the evolution of spatio-temporal data. With the iteration of each prediction period, the spatio-temporal adaptive causal network optimizes the causal graph using the prediction results of the previous stage and the feedback information at each stage, so as to achieve higher prediction accuracy when the power system faces complex dynamic changes.

[0018] Preferably, step S3 includes the following steps: Input the cleaned power load data, price data, and equipment status information into the spatio-temporal convolutional neural network for feature extraction; input the extracted spatio-temporal features into the long short-term memory network, and predict the change trend of the load in the next 7 days by analyzing the power load data in the past 30 days: Among them, is the hidden state at the current moment, is the input spatio-temporal feature, and the LSTM will predict the future load change based on past data; according to the output of the LSTM, generate a prediction trend graph of future power load and price; Based on the prediction results, use the reinforcement learning strategy to generate a power trading strategy; when it is predicted that the future power price will rise significantly, the model will recommend buying power at the current lower price and selling it when the future price is high to maximize the profit.

[0019] Preferably, the spatio-temporal convolutional neural network further includes: a skip connection layer, which is used to avoid the problem of gradient disappearance during the learning process of spatio-temporal data, enabling the network to effectively learn long-term dependence relationships and helping to capture long-term market fluctuation patterns in power trading trend prediction; a pooling layer, which is used to reduce the dimension of spatio-temporal features, retain important information, and improve computational efficiency, thereby accelerating the processing speed of power trading trend prediction.

[0020] The present invention has the following beneficial effects: 1. In the present invention, by combining the spatio-temporal adaptive causal network with the spatio-temporal convolutional neural network and the causal inference algorithm, the accuracy of power trading trend prediction can be improved, the causal relationship graph can be adjusted in real time and adaptively, the causal relationship between variables in the power system can be deeply mined, and through the multi-stage feedback closed-loop optimization mechanism, continuous learning and self-optimization can be carried out, thereby enhancing the stability and reliability of the system.

[0021] 2. The present invention can optimize power trading decisions, improve the operating efficiency of the power system, enhance market response capabilities, support sustainable development, and help power market participants formulate more accurate trading strategies, improve economic efficiency, and reduce market risks through accurate prediction and real-time adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a system architecture and method flowchart of a power trading trend prediction and decision support system and method proposed by the present invention; Figure 2 It is a multi-stage feedback closed-loop optimization process diagram of the spatio-temporal adaptive causal network in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] As Figure 1 - Figure 2 shown, a power trading trend prediction and decision support system proposed by the present invention includes: Data acquisition module: used to obtain spatio-temporal data sets in the power system, and the spatio-temporal data sets include multiple power load data, power price data, equipment status data, and sensor data; Spatio-temporal adaptive causal network module: used to model the variables in the spatio-temporal data set and their mutual influence relationships through a causal inference algorithm. The causal inference algorithm constructs causal relationships based on historical data and dynamically adjusts the weights of the edges to reflect changes in causal relationships.

[0025] Power trading trend prediction and decision support module: used to generate multi-level prediction and decision support for the power system based on the joint modeling results of the spatio-temporal adaptive causal network and the spatio-temporal convolutional neural network; Feedback and optimization module: used to input the power trading trend prediction and equipment scheduling decision as feedback into the spatio-temporal adaptive causal network graph, and dynamically adjust and optimize the spatio-temporal causal graph; Iterative optimization module: used to optimize the power trading trend prediction and equipment scheduling decision through multiple iterations and feedback closed-loop.

[0026] A power trading trend prediction and decision support method using the above power trading trend prediction and decision support system includes: S1: Obtain the spatio-temporal dataset in the power system. The spatio-temporal dataset includes multiple power load data, power price data, equipment status data, and sensor data. Each data item in the spatio-temporal dataset has a timestamp and a spatial identifier, which are used to characterize the state changes of the power system at multiple time points and geographical locations; S2: Based on the historical data in the spatio-temporal dataset, use the causal inference algorithm to determine the causal relationships between various variables in the power system. Combine the spatio-temporal convolutional neural network to model the dynamic changes in the spatial and temporal dimensions, and weight the spatio-temporal data through the attention mechanism, so as to dynamically adjust the weights of the edges in the causal graph according to the feedback information, generate the causal relationship graph, and form the spatio-temporal adaptive causal network; S3: Use the causal graph generated by the spatio-temporal adaptive causal network and the extracted spatio-temporal features to perform trend prediction through the neural network; According to the output of the spatio-temporal data model, predict the future change trends of power load and price, and provide support for the trading strategies in the power market; Combine the predicted power trading trends, and use the decision-making module of the spatio-temporal adaptive causal network to generate corresponding power trading strategies; S4: Input the power trading trend prediction and decision support generated in step S3 as feedback information into the spatio-temporal adaptive causal network to dynamically optimize the causal graph and the spatio-temporal modeling process, including: taking the power trading trend prediction results and decision support strategies as new inputs and transmitting them into the spatio-temporal adaptive causal network graph, and analyzing their impacts on other power system variables through the causal inference module; Based on the feedback information, dynamically adjust the causal graph in the spatio-temporal adaptive causal network, update the edge weights and structures of the causal relationships, and ensure that the causal relationships can be adaptively adjusted as time evolves and the system state changes; S5: Multi-stage feedback closed-loop optimization prediction and decision-making: Based on multiple feedback and optimization, the accuracy of power trading trend prediction and decision support is improved, including: using the causal graph described in step S2 as the first causal relationship graph, and constructing a second causal relationship graph based on the feedback information described in step S4. After each feedback, the spatiotemporal adaptive causal network updates the causal graph by combining the spatiotemporal convolutional neural network with the adaptive attention mechanism, and retrains the prediction model, compares the first causal relationship graph with the second causal relationship graph, determines whether the causal relationship therein has changed significantly, and adjusts the edges in the causal graph based on the change. Weight; the feedback information is used to drive the causal reasoning module to dynamically adjust the edge weights and structure of the causal graph in the spatiotemporal adaptive causal network to ensure that the causal relationship can be adaptively adjusted as the state of the power system changes, and optimize the modeling process of spatiotemporal data through the spatiotemporal convolutional neural network and the adaptive attention mechanism, thereby improving the prediction accuracy; after each feedback, the causal graph is adjusted by analyzing the feedback data, and the prediction model in the spatiotemporal adaptive causal network is retrained. Through multiple feedback iterations, the optimization of power trading trend prediction and decision support strategy is achieved until the predetermined prediction accuracy and decision efficiency goals are achieved.

[0027] In one embodiment, step S1: obtaining a spatiotemporal data set in the power system: first, it is necessary to obtain a spatiotemporal data set related to prediction and decision-making in the power system. The acquisition of the spatiotemporal data set mainly includes two sources: real-time collected data and historical data.

[0028] Real-time data collection: Use sensors, smart devices and communication networks in the power system to collect power load data, power price data, equipment status data and sensor monitoring data in real time. Each data item contains a timestamp and a spatial identifier to describe the time and location of its acquisition. Specifically, power load data is collected through power meters, power price data is obtained through the power market system, equipment status data is provided by the control center or the device itself, and sensor data (such as temperature, current, voltage, etc.) is collected by sensors at different locations.

[0029] Historical data: Obtain historical operation data of the power system, including information such as power load, price, and equipment status over a period of time. This data is usually provided by the power dispatch center or historical database and obtained through the query interface. Historical data provides the basis for training and model building of spatiotemporal datasets.

[0030] The acquired spatiotemporal dataset contains the data changes of multiple variables (power load, power price, equipment status, sensor data, etc.) at different time points and spatial locations. And data processing is performed. Specifically, the spatiotemporal dataset includes the following key parts: Power load data: Records the power load information of different regions (such as each load point or area), including real-time load and historical load data.

[0031] Power price data: Includes the power trading prices at multiple time points and different regions in the power market. These price data reflect the impact of factors such as supply and demand relationships and market regulation mechanisms on price fluctuations in the power market.

[0032] Equipment status data: Involves the operating status information of each device (such as transformers, switches, lines, etc.) in the power system. Equipment status data is mainly obtained through the SCADA (Supervisory Control and Data Acquisition) system or intelligent sensors, identifying the working status, fault information, maintenance status, etc. of power equipment.

[0033] Sensor data: The sensors installed in the power system (such as current sensors, voltage sensors, temperature sensors, etc.) are used to monitor the parameter changes of power equipment and the environment in real time. These data can reflect the working status of power equipment and environmental changes, assisting in diagnosing and predicting the operating status of the power system.

[0034] Each piece of data includes a timestamp (the time point when the data is recorded) and a spatial identifier (identifying the data collection location). For example, the power load data may include the timestamp "2024-01-01 10:00:00" and the spatial identifier "Area A", so that the specific data at each time point and each location can be clearly recorded.

[0035] In one embodiment, step S2: Construct a spatio-temporal adaptive causal network graph: Construct a preliminary causal relationship graph through a causal inference algorithm based on historical data. Identify the mutual influence relationships between variables in the power system through causal inference methods. The specific implementation steps are as follows: Use the spatio-temporal data set obtained from step S1 (including power load, power price, equipment status, and sensor data, etc.) as input data. This data set contains multi-dimensional data at multiple time periods and spatial locations.

[0036] Select an appropriate causal inference algorithm, such as the inverse information entropy algorithm and Granger causality test, to model the causal relationships between variables in the power system. By analyzing the historical relationships in spatio-temporal data, determine which variables have significant causal dependencies.

[0037] Inverse information entropy method: Determine the causal relationship between variables by calculating the change in information entropy. Information entropy reflects the degree of disorder of the system. If the change of one variable leads to the change of the uncertainty of another variable, it is considered that there is a causal relationship between the two.

[0038] Granger causality test: The causality between variables is tested through the lagged terms of time series data. If the past values of one variable can significantly predict the future values of another variable, causality is considered to exist: testing whether the past values of Y(t) can significantly predict the future values of X(t):

[0039]

[0040] where p and q are the lag orders, , , are the regression coefficients. If is significantly non-zero, it indicates that there is Granger causality from X(t) to Y(t).

[0041] Causality graph generation: Through the above causal reasoning method, a preliminary causality graph is generated, which shows the causal relationships between variables and their preliminary weight values. The edges of the graph represent the causal relationships between variables, and the nodes represent the variables themselves. The weight value of each edge reflects the strength and direction of the causal relationship.

[0042] For the generation of the preliminary causality graph, next, by combining the spatio-temporal convolutional neural network (ST-CNN) and the attention mechanism, the weights of the edges in the causality graph are adjusted according to the dynamic changes of spatio-temporal data, thus forming a spatio-temporal adaptive causal network (STACN). The specific implementation process is as follows: Use the spatio-temporal convolutional neural network (ST-CNN) to model the spatial and temporal dimensions. Through convolutional operations, ST-CNN can capture local and global dependencies from spatio-temporal data. ST-CNN extracts the temporal change patterns from time series data by applying convolutional kernels on the time series data. It learns the correlations between different time points, especially in the cases where the load, price, and device status change over time. Through spatial convolutional operations, ST-CNN can learn the dependencies between various spatial locations in the power system and identify the associations between different regions or devices.

[0043] Use the attention mechanism to weight the spatio-temporal data. The attention mechanism can dynamically adjust the weights according to the importance of the data, highlighting the key spatio-temporal features and weakening the unimportant parts. The specific implementation includes: For each time step of the time series data, the attention mechanism dynamically weights the influences of different time points according to the historical performance of the data. By learning the weight coefficients, the model can pay more attention to those time periods with significant changes over time. The attention mechanism is used to weight the data at different spatial locations, so that the model pays more attention to those spatial regions that contribute more to the change of the system state.

[0044] Based on spatio-temporal convolution and attention mechanism, the model dynamically adjusts the weights of the edges in the causal graph. Specifically, through the real-time feedback and evolution of spatio-temporal data, the causal inference model can update the weights of each edge in the graph according to the current spatio-temporal state, forming an adaptive causal network. This dynamic adjustment can reflect the changes in causal relationships in the power system in real time.

[0045] By continuously obtaining new spatio-temporal data inputs, the model adjusts the causal graph according to the latest feedback information. The feedback information includes the change trend of power load, the state change of equipment, the change of external environment, etc. These information are input into the causal network to adjust the weights of each edge in the network to reflect the changes in causal relationships. The generated spatio-temporal adaptive causal network (STACN) contains the causal relationships between various variables in the power system and adaptively adjusts the weights of the edges with the dynamic changes of spatio-temporal data. This enables the STACN graph to accurately reflect the real-time state of the power system and provide support for subsequent prediction and decision-making:

[0046] Where: is the weight of the previous update.

[0047] is the learning rate, which controls the update speed.

[0048] is the change in causal impact calculated based on new data.

[0049] The generated spatio-temporal adaptive causal network (STACN) is used for the following aspects: Power load forecasting: Using the causal relationships in the STACN graph to predict the power loads of different regions and equipment in the future for a period of time.

[0050] Equipment fault diagnosis: Based on the causal relationships in the STACN graph, detecting abnormal changes in the operating state of equipment and timely discovering potential fault risks.

[0051] Power market decision support: By analyzing the causal relationship between price changes and load fluctuations in the STACN graph, providing support for price forecasting and decision-making in the power market.

[0052] In one embodiment, step S3: Generate power trading trend prediction and decision support based on the STACN output: According to the outputs of the spatio-temporal adaptive causal network (STACN) and the spatio-temporal convolutional neural network (ST-CNN), generate a power trading trend prediction, and generate an optimized decision support strategy according to the prediction result. The specific implementation is as follows: Obtain the output of the spatio-temporal data model processed by the spatio-temporal convolutional neural network (ST-CNN) and the adaptive attention mechanism from step S3. This data includes spatio-temporal features of variables such as power load, price, and equipment status; spatio-temporal features include, but are not limited to, power system operation data in different time dimensions (such as hours, days, months, etc.) and space dimensions (such as regions, subsystems, etc.).

[0053] Spatio-temporal data modeling and causal relationship reasoning: The causal reasoning module of STACN analyzes the causal relationships between key variables such as power load and price based on the spatio-temporal data model generated in step S3, and constructs a causal map of variables in the power system.

[0054] According to the causal map, analyze the mutual influence between different factors in the future period to obtain the evolution trend of the power system state; the spatio-temporal data at the current moment , and the output of STACN is the spatio-temporal prediction data at the future moment :

[0055] where G is the causal relationship graph.

[0056] Trend prediction model: Use a deep neural network (such as a long short-term memory network LSTM or other time series prediction models) to predict the spatio-temporal data of the power system. This model uses the causal reasoning results and spatio-temporal features to predict key variables such as power load and price in the future period.

[0057] The prediction results include the change trends of key variables such as power load and price in the future time period. Specifically, it includes: The change trend of power demand (such as peak load, valley load, etc.).

[0058] The predicted fluctuations of electricity prices (such as the upward or downward trend of market prices).

[0059] The evolution of the power equipment status (such as equipment overload, equipment response caused by load changes, etc.).

[0060] Input of the decision-making model: Through power trading trend prediction, obtain the trend data of the future power market, including load demand prediction, price fluctuation trend, etc. These prediction results are used as inputs to provide to the decision support module to support the optimization of power trading decisions based on the spatio-temporal causal network (STACN). In this module, based on the predicted trend data and the causal reasoning output of STACN, generate specific power trading strategies. These strategies include: Strategy for selecting power trading time periods: Determine the best time to conduct power trading at specific time periods based on load prediction and price prediction.

[0061] Power purchase and sales decision-making: Based on market price trends, optimize the decision to purchase or sell electricity to reduce costs or maximize revenue.

[0062] Demand response management: According to load forecasting, formulate demand response strategies for the power system to ensure that the system load is within a controllable range and avoid overload or power shortage situations.

[0063] Optimize the operation of the power system: The generated power trading strategies will optimize the operation status of the power system. For example, by optimizing power load distribution, dispatching transactions and other measures, ensure that the power system reaches the optimal operation status at different time periods.

[0064] Output optimized power trading decisions, specifically including: Trading suggestions for each time period: For example, when to buy electricity and when to sell electricity.

[0065] Specific strategies for load management: including power dispatching, demand response, etc.

[0066] Buy and sell decisions made based on power market price forecasts.

[0067] Decision support generation is based on the prediction results, combined with STACN causal reasoning analysis, to generate specific power trading decision strategies. By dynamically adjusting power trading timing, load distribution and market price management strategies, ensure that the power system can still achieve the optimal economic benefits and operation effects under various uncertainties.

[0068] In one embodiment, step S4: Use the power trading trend prediction and decision support as feedback to input into STACN for optimization. Take the power trading trend prediction and decision support results generated from S3 as feedback inputs and transmit them into the spatio-temporal adaptive causal network (STACN). Then, through the dynamic adjustment of causal reasoning and spatio-temporal modeling, optimize the prediction results and decision support strategies. The specific implementation method is as follows: Take the power trading trend prediction data and decision support strategies generated in step S3 as new input data and transmit them to the causal reasoning module in the spatio-temporal adaptive causal network (STACN). These input data are fused with the previous spatio-temporal data as feedback information and transmitted into the network through the input layer of STACN. The feedback information is organized in a time series manner and contains historical and predicted data of multiple key factors such as power load, market price, equipment status, etc. It also includes the relevant decision results generated by the decision support strategies.

[0069] Causal reasoning analysis: The causal reasoning module in STACN determines the causal relationships between variables by analyzing the relationship between feedback information and the existing spatio-temporal data model. For example, the possible influence between the predicted power load and the electricity price, the potential impact of load fluctuations on the equipment status, etc. Technologies such as graph neural networks (GNNs) are used to infer causal relationships in different time periods through known causal graphs, and to identify the influence paths of various variables in the power system.

[0070] Based on the new prediction data and decision results in the feedback information, the causal reasoning module dynamically adjusts the causal graph in STACN. Specifically, it includes: Edge weight update: According to the feedback prediction results, update the edge weights of the causal relationships and adjust the relative strength between the influencing variables. For example, when the accuracy of load prediction improves, the system will automatically increase the weight of the causal relationships related to the load.

[0071] Self-adaptability of the causal graph: The adaptive adjustment process of the causal graph is based on the evolutionary mechanism of time and space, and can be adjusted in real time according to the changes in the power system and feedback information. For example, when there is a large error in power load prediction, STACN will adaptively adjust the causal graph structure to reduce errors in future predictions.

[0072] By inputting the power trading trend prediction and decision support as feedback information into STACN, the system can dynamically adjust the structure and edge weights of the causal graph according to the actual feedback, thereby adaptively optimizing the spatio-temporal data modeling process. This improves the operating efficiency and decision-making accuracy of the power system.

[0073] Step S4 realizes the in-depth mining and adjustment of the causal relationships between variables in the power system through the combination of causal reasoning and spatio-temporal adaptive mechanisms. By adaptively optimizing the causal graph structure, the system can continuously optimize the prediction results according to the feedback information and improve the accuracy of decision support. Each feedback will lead to the optimization of the causal graph and the prediction model.

[0074] In one embodiment, step S5: Multi-stage feedback closed-loop optimization of prediction and decision-making: Through a multi-stage feedback closed-loop mechanism, continuously optimize the accuracy and effectiveness of power trading trend prediction and decision support. Utilize the spatio-temporal convolutional neural network (ST-CNN) and adaptive attention mechanism in the spatio-temporal adaptive causal network (STACN), combine the prediction results and decision support strategies, and conduct repeated optimization and iteration. The specific implementation method is as follows: Obtain feedback information and construct the second causal relationship graph: After each prediction and decision, collect feedback information on actual operation data from relevant fields such as the electricity trading market and power load dispatching, including but not limited to: actual electricity trading results, power demand fluctuations, power equipment status, efficiency of system emergency response, etc.; Based on this feedback information, construct the second causal relationship graph. By analyzing new spatio-temporal data and feedback information, capture new causal relationships between variables in the power system and perform preliminary updates.

[0075] In step S2, the causal relationship graph generated by combining the causal inference algorithm and the spatio-temporal convolutional neural network (ST-CNN) with historical spatio-temporal data is used as the first causal relationship graph, which represents the preliminary causal relationships and dynamic changes between various variables in the power system. Based on the first causal relationship graph, the system will gradually optimize the structure and edge weights of the causal graph based on feedback information (from the actual operation data of the power market).

[0076] Compare the significance of the first causal relationship graph and the second causal relationship graph: Use statistical methods (such as t-tests, hypothesis tests, etc.) to compare the significance between the first causal relationship graph and the second causal relationship graph. Through significance analysis, identify which causal relationships become more important at different times or market states, so as to provide a basis for model optimization. For example, in a certain time period, the relationship between power load and electricity price is more significant, and the weight of this relationship is adjusted preferentially.

[0077] Input the generated second causal relationship graph as feedback information into the spatio-temporal adaptive causal network (STACN) and transfer it to the module as new input data. The causal inference module will update and infer causal relationships based on the latest feedback data and historical data. At this time, the causal inference module in STACN can automatically adjust the edge structure and edge weights of the graph according to the feedback information, further improving the causal relationship model of the system.

[0078] According to the new data in the feedback information, the causal inference module in STACN will adaptively adjust the structure and edge weights of the causal graph. Specific adjustments include: Dynamically adjust the relationship weights between variables in the prediction model according to the actual market situation and changes in power demand. For example, if the power load changes significantly in a certain time period, this module will enhance the causal relationship between the load and the equipment status, thereby optimizing the prediction accuracy. Automatically adjust the causal relationship graph structure based on the pattern changes in the feedback information, delete irrelevant relationships or add new associations, ensuring that the causal graph always remains consistent with the actual changes in the power market and load.

[0079] The updated causal relationship graph will be input into a spatio-temporal convolutional neural network (ST-CNN) to perform spatio-temporal data modeling in combination with an adaptive attention mechanism, further optimizing the prediction accuracy of variables such as power load and price. ST-CNN can handle the complexity of spatio-temporal data, extract features in the spatial and temporal dimensions through spatio-temporal convolution, and strengthen the model's learning of important spatio-temporal features.

[0080] After each feedback, STACN will retrain the prediction model based on the updated causal graph and the output of the spatio-temporal convolutional neural network, further improving the prediction accuracy. The results after each iteration will have an optimizing impact on the prediction results of power trading trends and decision support strategies. Specifically, it includes: regenerating the prediction of power trading trends according to the latest feedback data and the optimized model. Generating a more accurate power trading decision-making strategy based on the optimized prediction results, improving the timeliness and accuracy of decision-making.

[0081] Each round of feedback data and optimization results will form a new closed loop, continuously promoting the improvement of the prediction model and decision support strategy. Specifically, it is reflected in: The accuracy of power trading trend prediction is greatly improved: the prediction deviation is reduced, and the adaptability of the model to market changes is enhanced.

[0082] The power trading decision-making strategy is more accurate and reasonable: the timing of trading and the efficiency of resource allocation are improved, and the formulation of power trading strategies is optimized.

[0083] The closed-loop mechanism continues until the accuracy of power trading trend prediction and decision support reaches the set expected standard. With the changes in the market and system status, the system can adaptively adjust the model and strategy to adapt to the dynamic changes of the power market and ensure long-term stable and efficient operation.

[0084] Through this multi-stage feedback closed-loop optimization process, the overall performance of the system will be significantly improved, ensuring that the prediction of power trading trends and decision support can achieve the best results in different market environments.

[0085] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting and supporting power trading trends, characterized in that: include: S1: Acquire a spatiotemporal data set in a power system, wherein the spatiotemporal data set includes a plurality of power load data, power price data, equipment status data and sensor data, wherein each data item in the spatiotemporal data set has a timestamp and a spatial identifier; S2: Based on the historical data in the spatiotemporal dataset, a causal inference algorithm is used to determine the causal relationship between various variables in the power system. The spatiotemporal convolutional neural network is combined to model the dynamic changes in the spatial and temporal dimensions, and the spatiotemporal data is weighted through the attention mechanism. The weights of the edges in the causal graph are dynamically adjusted according to the feedback information, a causal relationship graph is generated, and a spatiotemporal adaptive causal network is formed. S3: Using the causal graph generated by the spatiotemporal adaptive causal network and the extracted spatiotemporal features, trend prediction is performed through neural networks; Based on the output of the spatiotemporal data model, the future trend of power load and price changes is predicted, and support is provided for the trading strategy of the power market; combined with the predicted power trading trend, the corresponding power trading strategy is generated using the spatiotemporal adaptive causal network; S4: input the power trading trend prediction and decision support generated in step S3 as feedback information into the spatiotemporal adaptive causal network to dynamically optimize the causal graph and spatiotemporal modeling process; S5: Multi-stage feedback closed-loop optimization prediction and decision-making: Based on multiple feedback and optimization, the accuracy of power trading trend prediction and decision support is improved, including: using the causal map described in step S2 as the first causal relationship map, and constructing a second causal relationship map based on the feedback information described in step S4. After each feedback, the spatiotemporal adaptive causal network updates the causal map by combining the spatiotemporal convolutional neural network with the adaptive attention mechanism, and retrains the prediction model to compare the first causal relationship map with the second causal relationship map to identify the difference in causal relationship between the two, and adjust the causal map based on the items in the difference that meet the predetermined statistical standards. The feedback information is used to drive the causal reasoning module to dynamically adjust the edge weights and structure of the causal graph in the spatiotemporal adaptive causal network to ensure that the causal relationship can be adaptively adjusted as the state of the power system changes, and optimize the modeling process of spatiotemporal data through the spatiotemporal convolutional neural network and the adaptive attention mechanism, so as to improve the prediction accuracy; after each feedback, the causal graph is adjusted by analyzing the feedback data, and the prediction model in the spatiotemporal adaptive causal network is retrained. Through multiple feedback iterations, the optimization of power trading trend prediction and decision support strategy is realized until the predetermined prediction accuracy and decision efficiency goals are achieved.

2. A method for predicting and supporting power trading trends according to claim 1, characterized in that: The step S2 includes: using the Granger causality test method to determine the causal relationship between power load and power price: Use the AIC criterion to select the number of delays. Select the time delay p, and the selected delay is 1; The Granger test uses a regression model to test whether the past values ​​of Y(t) can significantly predict the future values ​​of X(t): ; Where p and q are the lag orders, , , is the regression coefficient, if If it is significantly different from zero, it indicates that there is a Granger causal relationship between X(t) and Y(t); If the Granger test shows that power load has a significant impact on price, then add a directed edge of "power load → power price" to the causal graph; The spatial features are extracted through 3×3 spatial convolution, and then the features of the time dimension are extracted through 1D convolution. After each convolution operation, the corresponding spatiotemporal feature map is generated; Apply an adaptive attention mechanism to calculate attention weights based on spatial and temporal features: in, is the weight score of time t and space s, is the normalized attention value; The spatiotemporal convolutional features are combined with the attention weights to generate the final spatiotemporal feature map for subsequent decision support.

3. A method for predicting and supporting power trading trends according to claim 1, characterized in that: The step S2 further includes: based on each timestamp and spatial identifier in the multi-source spatiotemporal data set, using an inverse information entropy algorithm to perform causal reasoning to identify potential causal relationships between variables in the power system; assigning a preliminary weight to each edge in the causal graph, wherein the weight represents the strength of the causal relationship between nodes; and dynamically adjusting the causal graph through a feedback mechanism according to changes in the spatiotemporal data, so that the graph can adapt to the ever-changing state in the power system and improve the prediction accuracy of power trading trends.

4. A method for predicting and supporting power trading trends according to claim 1, characterized in that: The spatiotemporal convolutional neural network includes: a spatial convolution layer, which is used to extract the spatial characteristics of equipment and sensor data in the power system and capture the spatial dependencies between devices; a temporal convolution layer, which is used to extract the time series characteristics of power load and power price variables and capture the changing trends of various variables in the power system over time; and a spatiotemporal fusion layer, which fuses the features extracted by the spatial convolution layer and the temporal convolution layer to obtain a more comprehensive spatiotemporal feature representation for use in power trading trend prediction.

5. The method for predicting and supporting power trading trends according to claim 1, characterized in that: The step S4 specifically includes: taking the power trading trend prediction results and decision support strategies as new inputs, passing them to the spatiotemporal adaptive causal network graph, and analyzing their impact on other power system variables through the causal reasoning module; based on feedback information, dynamically adjusting the causal graph in the spatiotemporal adaptive causal network, and updating the edge weights and structure of the causal relationship.

6. A method for predicting and supporting power trading trends according to claim 1, characterized in that: The step S5 includes: in the first stage, based on the initial spatiotemporal causal graph and the output of the spatiotemporal convolutional neural network, generating a power trading trend forecast, and generating decision support information based on the forecast results; in the second stage, inputting the power trading trend forecast results of the first stage into the spatiotemporal causal graph as feedback, updating the strength and structure of the causal relationship to enhance the prediction accuracy of the next step; in the third stage, optimizing the prediction ability of the spatiotemporal convolutional neural network based on the feedback results, and further adjusting the power trading trend forecast results to form an adaptive optimization model to provide decision support.

7. A method for predicting and supporting power trading trends according to claim 1, characterized in that: The dynamic adjustment process of the spatiotemporal adaptive causal network graph includes the following steps: according to the feedback information of the power system after the prediction cycle, the causal relationship and edge weight in the graph are adjusted in real time, the feedback information includes the actual transaction data of the power market, load fluctuation and other related events, and the feedback information is used to optimize the edge weight and causal relationship strength in the causal graph; in the feedback adjustment process, the spatiotemporal adaptive causal network dynamically adjusts the structure of the causal graph by analyzing the evolution of spatiotemporal data. With the iteration of each prediction cycle, the spatiotemporal adaptive causal network optimizes the causal graph at each stage using the prediction results and feedback information of the previous stage, thereby achieving higher prediction accuracy of the power system when facing complex dynamic changes.

8. The method for predicting and supporting power trading trends according to claim 1, characterized in that: Step S3 includes the following steps: The cleaned power load data, price data and equipment status information are input into the spatiotemporal convolutional neural network for feature extraction. The extracted spatiotemporal features are input into the long short-term memory network to predict the load change trend in the next 7 days by analyzing the power load data of the past 30 days: in, is the hidden state at the current moment, Based on the input spatiotemporal features, LSTM predicts future load changes based on past data; based on the output of LSTM, a forecast trend chart of future power load and price is generated; Based on the prediction results, a reinforcement learning strategy is used to generate an electricity trading strategy; when it is predicted that electricity prices will rise sharply in the future, the model will recommend buying electricity when the current price is low and selling it when the price is high in the future to maximize profits.

9. The method for predicting and supporting power trading trends according to claim 1, characterized in that: The spatiotemporal convolutional neural network further includes: a skip connection layer, which is used to avoid the gradient vanishing problem in the learning process of spatiotemporal data, so that the network can effectively learn long-term dependencies and help capture long-term market fluctuation patterns in power trading trend prediction; a pooling layer, which is used to reduce the dimension of spatiotemporal features, retain important information and improve computing efficiency, thereby speeding up the processing speed of power trading trend prediction.

10. A power trading trend prediction and decision support system, using the power trading trend prediction and decision support method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module: used to obtain spatiotemporal data sets in the power system, which include power load data, power price data, equipment status data and sensor data; Spatiotemporal adaptive causal network module: used to model the variables in the spatiotemporal data set and their mutual influence relationship through a causal reasoning algorithm, wherein the causal reasoning algorithm constructs causal relationships based on historical data and dynamically adjusts the weights of edges to reflect changes in causal relationships; Power trading trend forecasting and decision support module: used to generate multi-level forecasting and decision support for power systems based on the joint modeling results of spatiotemporal adaptive causal networks and spatiotemporal convolutional neural networks; Feedback and optimization module: used to input power trading trend prediction and equipment dispatch decision into the spatiotemporal adaptive causal network map as feedback, and dynamically adjust and optimize the spatiotemporal causal map; Iterative Optimization Module: It is used to optimize power trading trend prediction and equipment scheduling decisions through multiple iterations and feedback loops.

Images