Power spot transaction privacy decision method and system based on cross-domain graph representation

Abstract

The disclosure provides a power spot transaction privacy decision method and system based on cross-domain graph representation, relating to the technical field of power spot transaction decision, comprising: constructing a power spot transaction graph model, and uniformly representing and learning the graph structure; using a cross-domain consistency alignment mechanism to align the distribution of node embedding, using a node semantic difference stripping mechanism to decouple shared features and domain private features, constructing a unified semantic space shared by two domains, and realizing cross-domain alignment; after cross-domain alignment, adaptive cross-domain migration is performed, the migration strength is dynamically adjusted, the migration proportion is controlled through node-level attention weight and graph-level gate control, dynamic graph embedding fusion is realized, and a target domain graph is obtained; a privacy enhancement strategy network model is constructed, the target domain graph is input into the privacy enhancement strategy network model, and through multi-objective collaborative optimization, a predicted privacy enhancement transaction decision is output. The disclosure improves the cross-domain generalization ability of the model, and effectively suppresses the privacy risk in the decision process.

Description

Technical Field

[0001] This disclosure relates to the field of electricity spot trading decision technology, specifically to a privacy decision-making method and system for electricity spot trading based on cross-domain graph representation. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] As a crucial component of electricity market reform, spot electricity trading is influenced by multiple factors, including load changes, generation costs, inter-regional dispatch strategies, market rules, and price signals. In recent years, the increasing number of participants and the growing complexity of trading structures have led to highly dynamic and strongly correlated market data. To enhance the analysis of trading behavior, existing research commonly employs graphical models, market characteristic modeling, and statistical analysis to identify relationships between nodes, capture inter-regional market coupling effects, and form a structured description of the trading status. However, these methods are often limited to a single market or region and are difficult to directly apply to data migration scenarios across regions and multiple markets.

[0004] With the increasing penetration of new energy sources and the expansion of inter-regional power transmission, significant differences exist in data distribution, pricing mechanisms, and trading strategies across different regional electricity spot markets. This leads to severe performance degradation of traditional models when migrated to new regions or different time periods, prompting existing solutions to address this issue using transfer learning. On the other hand, as the digitalization of the electricity market continues to increase, the privacy protection of transaction data becomes increasingly prominent. Existing privacy protection technologies mainly include differential privacy, encrypted computation, and federated learning, which are used to protect data privacy. However, the aforementioned existing solutions have the following limitations: (1) Although existing transfer learning methods can be used for cross-domain modeling, they generally rely on the assumption of feature space consistency, which makes it difficult to deal with problems unique to power trading scenarios such as graph structure heterogeneity, node semantic inconsistency, and dynamic changes in edge relationships.

[0005] (2) In addition, many models lack adaptive adjustment mechanisms. When cross-domain differences are significant, they often fail to maintain the stability and generalization ability of the model, and cannot meet the requirements of the electricity spot market for real-time performance and high reliability.

[0006] (3) Existing privacy protection methods have two prominent problems in real-world scenarios: First, overprotection leads to a decrease in model availability, such as excessive noise disturbance in differential privacy interfering with the optimization process; Second, privacy protection strategies are usually separated from specific tasks and cannot simultaneously take into account the security of market transactions, decision performance and availability of cross-domain transfer.

[0007] (4) In addition, traditional privacy protection mechanisms focus on data protection but lack effective constraints on the "privacy leakage risk in the decision-making process". When faced with the insufficient interpretability of deep model reasoning, it is easy to cause problems such as strategy back-reasoning and leakage of price-sensitive information.

[0008] In summary, most existing technical solutions focus on learning representations of single-region trading networks using graph neural networks, or on localized techniques for cross-domain transfer or privacy protection, lacking the ability to integrate them into a unified technical framework. In cross-domain scenarios, effective solutions have yet to be developed for maintaining graph structure consistency, semantic alignment, adaptive transfer adjustment, and balancing privacy risks and decision quality. This makes it difficult for existing technologies to support the comprehensive needs of "accurate representation, transferable learning, and privacy-friendly decision-making" in cross-regional, cross-market, and multi-entity electricity spot trading. Summary of the Invention

[0009] To address the aforementioned issues, this disclosure proposes a privacy-focused decision-making method and system for electricity spot trading based on cross-domain graph representation. It constructs an integrated approach for achieving high-quality graph representation, adaptive migration, and privacy-friendly decision-making in cross-domain scenarios. By introducing cross-domain graph representation consistency modeling, adaptive migration adjustment mechanisms, and intelligent privacy decision-making strategies, it achieves a unified expression of the structural relationships and semantic features of multi-source electricity spot trading networks, enhances the model's cross-domain generalization ability, and effectively suppresses privacy risks during the decision-making process. This provides a secure, reliable, and transferable intelligent auxiliary decision-making capability for electricity spot trading.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions: Privacy-based decision-making methods for electricity spot trading based on cross-domain graph representation include: Key elements for accessing the electricity spot market; Based on the structural relationships between various key elements, a graph model for electricity spot trading is constructed, and the graph structure is subjected to unified representation learning to obtain node embeddings; By using a cross-domain consistency alignment mechanism to align the distribution of embedded nodes, calculating the minimum distribution difference, and employing a node semantic difference stripping mechanism to decouple shared features from domain-private features, a unified semantic space shared by the two domains is constructed to achieve cross-domain alignment. After cross-domain alignment, adaptive cross-domain migration is performed. The migration intensity is dynamically adjusted according to the cross-domain embedding differences. The migration ratio is controlled by node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. A privacy-enhancing strategy network model is constructed. The target domain graph is input into the privacy-enhancing strategy network model, and through multi-objective collaborative optimization, the predicted privacy-enhancing transaction decision is output.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions: A privacy-preserving decision-making system for electricity spot trading based on cross-domain graph representation includes: The data acquisition module is used to acquire key elements of the electricity spot market; The representation learning module is used to construct a graph model of electricity spot trading based on the structural relationships between various key elements, and to perform unified representation learning on the graph structure to obtain node embeddings; The cross-domain alignment module is used to embed nodes for distribution alignment using a cross-domain consistency alignment mechanism, calculate the minimum distribution difference, adopt a node semantic difference stripping mechanism, decouple shared features from domain-private features, construct a unified semantic space shared by the two domains, and achieve cross-domain alignment. The adaptive migration module is used to perform adaptive cross-domain migration after cross-domain alignment. It dynamically adjusts the migration intensity based on the differences in cross-domain embedding and controls the migration ratio through node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. The strategy decision module is used to construct a privacy-enhancing strategy network model. The target domain graph is input into the privacy-enhancing strategy network model, and through multi-objective collaborative optimization, the predicted privacy-enhancing transaction decision is output.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned privacy decision-making method for electricity spot trading based on cross-domain graph representation.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned privacy decision-making method for electricity spot trading based on cross-domain graph representation.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the privacy decision-making method for electricity spot trading based on cross-domain graph representation.

[0015] Compared with the prior art, the beneficial effects of this disclosure are as follows: This disclosure presents a privacy-focused decision-making method for electricity spot trading based on cross-domain graph representation. By constructing electricity spot trading graphs in both the source and target domains, it expresses multi-dimensional relationships such as generation units, load nodes, market participants, and transmission coupling in a unified graph structure, and employs graph neural networks to extract high-dimensional structured representations. Compared to traditional methods that process trading data using time-series or feature concatenation, this disclosure can capture the potential correlations and regional coupling mechanisms between nodes, significantly improving the model's ability to understand complex market structures. It achieves a unified graph-structured representation of cross-regional electricity spot trading relationships, significantly enhancing the model's ability to characterize complex trading networks.

[0016] This disclosure presents a privacy-focused decision-making method for electricity spot market transactions based on cross-domain graph representation. The electricity spot market exhibits differences in regional mechanisms, load patterns, and price structures, making traditional models difficult to transfer across regions. This disclosure constructs a shared semantic space between the two domains through mechanisms such as distribution alignment, graph structure mapping, and shared-private feature decomposition, effectively reducing cross-domain data distribution bias. This method maintains stable predictive ability even with limited labeled data in the target domain, significantly improving the model's cross-domain generalization. The introduction of cross-domain semantic alignment and structural consistency constraints effectively addresses the problem of insufficient model transferability caused by regional differences.

[0017] This disclosed method for privacy-based decision-making in electricity spot trading, based on cross-domain graph representation, dynamically adjusts the migration intensity according to cross-domain embedding differences. It controls the migration ratio through node-level attention weights and graph-level gating, and enhances semantic consistency through knowledge distillation. This adaptive migration mechanism avoids performance degradation caused by over-migration, enabling the model to maintain high accuracy and stability even under significant differences in regional market mechanisms and data. The adaptive migration mechanism also avoids negative migration, improving the model's robustness and reliability under multi-regional market conditions.

[0018] This disclosed privacy-preserving decision-making method for electricity spot trading based on cross-domain graph representation constrains the model's dependence on sensitive features through gradient sensitivity constraints, incorporates local differential privacy noise into the embedding space, and applies policy smoothing constraints to ensure stable decision-making performance under privacy perturbations. This mechanism effectively prevents sensitive information such as generation costs, dispatch strategies, and marginal clearing information from being back-reasoned or leaked during inference, achieving a balance between privacy protection and decision accuracy. The privacy-enhancing decision-making mechanism significantly reduces the risk of leakage of sensitive transaction information during the inference stage.

[0019] This disclosure presents a privacy-focused decision-making method for electricity spot trading based on cross-domain graph representation. It constructs a unified multi-objective loss function, incorporating prediction error, cross-domain alignment, structural consistency, distillation constraints, and privacy regularization into a single optimization framework to achieve collaborative training across multiple objectives. This approach ensures prediction accuracy while providing cross-domain adaptability and privacy protection. The overall algorithm structure is modularly deployable and easily integrated with existing electricity market platforms, demonstrating high engineering value. Joint optimization achieves a global balance between prediction performance, transfer adaptability, and privacy protection, thereby enhancing the model's engineering usability and deployability. Attached Figure Description

[0020] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0021] Figure 1 This is a flowchart of a privacy decision-making method for electricity spot trading based on cross-domain graph representation, according to an embodiment of this disclosure. Figure 2 This is an experimental comparison of the prediction accuracy of the present disclosure and the comparative method under different degrees of cross-regional differences in a multi-regional electricity market environment. Figure 3 Stability curves of the cross-regional migration characteristics of this disclosure under different types of market differences (such as regional load differences, unit structure differences, and transaction rule differences) in the embodiments of this disclosure. Figure 4 A visual comparison of the privacy protection effects of this disclosure and traditional graph neural networks when making decision-making inferences on target domain graph data; Figure 5 This is a bar chart comparing the cross-domain forecasting performance metrics (MAE, RMSE, F1-score, etc.) after deployment in different electricity market mechanisms (such as spot, day-ahead, and intraday). Detailed Implementation

[0022] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] Example 1 One embodiment of this disclosure provides a privacy decision-making method for electricity spot trading based on cross-domain graph representation, the method steps of which include: Step 1: Obtain the key elements of the electricity spot market; Step 2: Based on the structural relationships between the key elements, construct a graph model for electricity spot trading, and perform unified representation learning on the graph structure to obtain node embeddings; Step 3: Use the cross-domain consistency alignment mechanism to embed nodes for distribution alignment, calculate the minimum distribution difference, adopt the node semantic difference stripping mechanism to decouple shared features and domain-private features, construct a unified semantic space shared by the two domains, and achieve cross-domain alignment; Step 4: After cross-domain alignment, adaptive cross-domain migration is performed. The migration intensity is dynamically adjusted according to the cross-domain embedding differences. The migration ratio is controlled by node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. Step 5: Construct a privacy-enhancing strategy network model. Input the target domain graph into the privacy-enhancing strategy network model, and through multi-objective collaborative optimization, output the predicted privacy-enhancing transaction decision.

[0026] As one embodiment, the privacy-focused decision-making method for electricity spot trading based on cross-domain graph representation disclosed herein achieves comprehensive modeling and optimization of trading structures, dynamic behaviors, and privacy-sensitive features in multi-regional electricity spot markets by constructing a unified cross-domain electricity trading graph model, designing an adaptive graph migration mechanism, and integrating a privacy-enhancing intelligent decision-making network. This disclosure constitutes a complete technical framework from four dimensions: graph representation learning, cross-domain feature alignment, privacy risk mitigation, and intelligent strategy decision-making, providing technical support for cross-domain collaborative analysis and secure trading decisions in complex market environments. The specific process is as follows: Step 1: Obtain the key elements of the electricity spot market, construct a graph model of electricity spot trading based on the structural relationships between the key elements, and perform unified representation learning on the graph structure to obtain node embeddings; Specifically, key elements for accessing the electricity spot market include generating units, load nodes, regional markets, transmission channels, and trading relationships, starting primarily in the source region. D s With the target domain Dt A graph model of electricity spot trading is constructed to depict the structural relationships among key elements such as generating units, load nodes, regional markets, transmission channels, and trading relationships. This structural relationship refers to the stable relational structure formed between generating units, load nodes, regional markets, and transmission channels, determined by the physical topology constraints of the power grid, the trading rules of the electricity spot market, and the dispatching and clearing mechanisms. This structure can be abstracted as the connection relationships between nodes and their weights, used to characterize the constraint transmission, price formation, and strategy coupling characteristics of electricity trading behavior.

[0027] Furthermore, the source domain and target domain graphs are defined as follows: in, For the source domain graph, For the target domain graph, V This refers to a set of nodes (e.g., generator sets, regional nodes, dispatch centers). E It is a set of edges (transmission lines, transaction edges, pricing relationships, etc.). X These are node or edge attributes (power generation, marginal price, load level, line impedance, etc.).

[0028] Furthermore, to extract multi-level information from the graph structure, this disclosure employs graph neural networks (GNNs) for unified feature representation learning, such as nonlinear aggregation models like graph convolutional networks (GCNs) and graph attention networks (GATs). The basic update form is as follows: Alternatively, an attention mechanism could be employed: in, For the normalized adjacency matrix, For learnable parameters, This represents the strength of node relationships; using the above structure, node embeddings of the source and target domains are extracted. This is used for subsequent cross-domain transfer and feature alignment.

[0029] Step 2: Use the cross-domain consistency alignment mechanism to align the node embeddings, calculate the minimum distribution difference, and use the node semantic difference stripping mechanism to decouple shared features from domain-private features, construct a unified semantic space shared by the two domains, and achieve cross-domain alignment. Specifically, due to the differences in pricing mechanisms, load timing, and market participants across different regional electricity markets, direct migration may lead to model degradation. Against this backdrop, this disclosure proposes a cross-domain consistency alignment mechanism to achieve the construction of a unified semantic space for graph representations by minimizing distributional differences.

[0030] First, align the overall feature distribution using the maximum mean difference (MMD) or Wasserstein distance: Simultaneously, a graph structure consistency constraint is proposed, which constructs a structure matching loss in the edge space: in, It is a learnable mapping function used to map the target domain graph structure to a semantic space similar to the source domain.

[0031] Furthermore, by utilizing a node semantic difference stripping mechanism, shared features are decoupled from domain-private features: And introduce orthogonal constraints: Ensure that cross-domain common information is preserved and domain-specific information is removed during the migration process.

[0032] Step 3: After cross-domain alignment, adaptive cross-domain migration is performed. The migration intensity is dynamically adjusted according to the cross-domain embedding differences. The migration ratio is controlled by node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. Specifically, to address the negative migration problem caused by large cross-domain distribution differences, an adaptive migration adjustment mechanism is designed to adjust the migration intensity through dynamic weights at the node and graph levels.

[0033] Dynamic weights at the node and graph levels are designed based on cross-domain embedding differences. When the differences are large, the migration intensity of the node-level migration weights is automatically reduced. in, It serves as a spatial distance or semantic difference metric, used to measure the similarity of node embeddings between the source and target domains. When the difference is too large, the adaptive weights automatically reduce the transfer intensity to avoid model degradation.

[0034] Graph-level transfer adjustment is achieved through learnable gating functions, which control the proportion of cross-domain information sharing, thereby enabling dynamic graph embedding fusion. in, g By controlling the proportion of cross-domain information sharing, dynamic graph embedding and fusion can be achieved.

[0035] Furthermore, this invention introduces a cross-domain knowledge distillation mechanism: Further improve the semantic consistency after the target domain is migrated.

[0036] Step 4: Construct a privacy-enhancing strategy network model. Input the target domain graph into the privacy-enhancing strategy network model, and output the predicted privacy-enhancing transaction decision through multi-objective collaborative optimization.

[0037] Specifically, an intelligent privacy decision-making module is constructed, which predicts privacy policy output through privacy cost constraints, noise injection, and policy smoothing. Multiple losses are introduced, including prediction task loss, cross-domain distribution alignment loss, privacy enhancement loss, and structural consistency loss. A multi-objective collaborative optimization objective function is constructed. By calculating the multi-objective collaborative optimization objective function, unified modeling, efficient migration, and privacy-friendly decision-making of cross-domain transaction behavior are achieved.

[0038] As one embodiment, the electricity spot trading disclosed herein involves sensitive information such as the marginal cost of generating units, dispatch strategies, and regional supply and demand relationships. Directly using this information in decision-making models could easily lead to privacy leaks. This disclosure constructs an intelligent privacy-preserving decision-making module, achieving privacy-friendly policy output through privacy cost constraints, noise injection, and policy smoothing.

[0039] Modeling privacy breach risk as policy gradient sensitivity: Local Differential Privacy (LDP) is used to perturb the node embedding: To avoid policy oscillations caused by noise, a policy smoothing regularization term is designed: Here, s' represents the state after a slight perturbation, used to enhance privacy, security, and stability.

[0040] The ultimate goal of privacy-enhancing decision optimization is: in, It can be defined as a strategy return, price prediction accuracy, or scheduling utility function.

[0041] As one embodiment, this disclosure constructs a multi-task collaborative optimization objective function, integrating multiple optimization objectives such as prediction tasks, cross-domain alignment, privacy enhancement, and structural consistency, to build a unified collaborative optimization framework: F= in, Loss forecasting for spot prices / strategy For cross-domain distribution alignment loss, For graph structure consistency loss, For cross-domain knowledge distillation feature alignment loss, To mitigate the risk of privacy breaches and minimize losses, To smooth out the regularized loss for the strategy.

[0042] As one embodiment, through this collaborative optimization objective function, this disclosure can achieve unified modeling, efficient migration, and privacy-friendly decision-making for cross-domain transaction behavior. The specific algorithm is as follows: Simulation Experiment This disclosure includes comparative experiments. Table 1 shows the comparison of prediction accuracy and robustness between this disclosure and existing technologies in typical cross-domain scenarios. Table 2 shows the feature migration stability index (mean ± standard deviation) under different cross-domain difference types (load difference, structural difference, rule difference). Table 3 shows the actual deployment effect (practicability evaluation) under different electricity market mechanisms.

[0043] Table 1 Comparison of prediction accuracy and robustness in typical cross-domain scenarios. Table 2. Characteristic migration stability index (mean ± standard deviation) under different cross-domain difference types (load difference, structural difference, rule difference). Table 3. Actual Deployment Effects under Different Electricity Market Mechanisms (Practicality Evaluation) Example 2 One embodiment of this disclosure provides a privacy-focused decision-making system for electricity spot trading based on cross-domain graph representation, comprising: The data acquisition module is used to acquire key elements of the electricity spot market; The representation learning module is used to construct a graph model of electricity spot trading based on the structural relationships between various key elements, and to perform unified representation learning on the graph structure to obtain node embeddings; The cross-domain alignment module is used to embed nodes for distribution alignment using a cross-domain consistency alignment mechanism, calculate the minimum distribution difference, adopt a node semantic difference stripping mechanism, decouple shared features from domain-private features, construct a unified semantic space shared by the two domains, and achieve cross-domain alignment. The adaptive migration module is used to perform adaptive cross-domain migration after cross-domain alignment. It dynamically adjusts the migration intensity based on the differences in cross-domain embedding and controls the migration ratio through node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. The strategy decision module is used to construct a privacy-enhancing strategy network model. The target domain graph is input into the privacy-enhancing strategy network model, and through multi-objective collaborative optimization, the predicted privacy-enhancing transaction decision is output.

[0044] As one embodiment, the specific steps of a privacy decision-making system for electricity spot trading based on cross-domain graph representation disclosed herein are as follows: Step 1: Obtain the key elements of the electricity spot market, construct a graph model of electricity spot trading based on the structural relationships between the key elements, and perform unified representation learning on the graph structure to obtain node embeddings; Specifically, key elements for accessing the electricity spot market include generating units, load nodes, regional markets, transmission channels, and trading relationships, starting primarily in the source region. D s With the target domain D t A graph model of electricity spot trading is constructed to depict the structural relationships among key elements such as generating units, load nodes, regional markets, transmission channels, and trading relationships. This structural relationship refers to the stable relational structure formed between generating units, load nodes, regional markets, and transmission channels, determined by the physical topology constraints of the power grid, the trading rules of the electricity spot market, and the dispatching and clearing mechanisms. This structure can be abstracted as the connection relationships between nodes and their weights, used to characterize the constraint transmission, price formation, and strategy coupling characteristics of electricity trading behavior.

[0045] Furthermore, the source domain and target domain graphs are defined as follows: in, For the source domain graph, For the target domain graph, V This refers to a set of nodes (e.g., generator sets, regional nodes, dispatch centers). E It is a set of edges (transmission lines, transaction edges, pricing relationships, etc.). X These are node or edge attributes (power generation, marginal price, load level, line impedance, etc.).

[0046] Furthermore, to extract multi-level information from the graph structure, this disclosure employs graph neural networks (GNNs) for unified feature representation learning, such as nonlinear aggregation models like graph convolutional networks (GCNs) and graph attention networks (GATs). The basic update form is as follows: Alternatively, an attention mechanism could be employed: in, For the normalized adjacency matrix, For learnable parameters, This represents the strength of node relationships; using the above structure, node embeddings of the source and target domains are extracted. This is used for subsequent cross-domain transfer and feature alignment.

[0047] Step 2: Use the cross-domain consistency alignment mechanism to align the node embeddings, calculate the minimum distribution difference, and use the node semantic difference stripping mechanism to decouple shared features from domain-private features, construct a unified semantic space shared by the two domains, and achieve cross-domain alignment. Specifically, due to the differences in pricing mechanisms, load timing, and market participants across different regional electricity markets, direct migration may lead to model degradation. Against this backdrop, this disclosure proposes a cross-domain consistency alignment mechanism to achieve the construction of a unified semantic space for graph representations by minimizing distributional differences.

[0048] First, align the overall feature distribution using the maximum mean difference (MMD) or Wasserstein distance: Simultaneously, a graph structure consistency constraint is proposed, which constructs a structure matching loss in the edge space: in, It is a learnable mapping function used to map the target domain graph structure to a semantic space similar to the source domain.

[0049] Furthermore, by utilizing a node semantic difference stripping mechanism, shared features are decoupled from domain-private features: And introduce orthogonal constraints: Ensure that cross-domain common information is preserved and domain-specific information is removed during the migration process.

[0050] Step 3: After cross-domain alignment, adaptive cross-domain migration is performed. The migration intensity is dynamically adjusted according to the cross-domain embedding differences. The migration ratio is controlled by node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. Specifically, to address the negative migration problem caused by large cross-domain distribution differences, an adaptive migration adjustment mechanism is designed to adjust the migration intensity through dynamic weights at the node and graph levels.

[0051] Dynamic weights at the node and graph levels are designed based on cross-domain embedding differences. When the differences are large, the migration intensity of the node-level migration weights is automatically reduced. in, It serves as a spatial distance or semantic difference metric, used to measure the similarity of node embeddings between the source and target domains. When the difference is too large, the adaptive weights automatically reduce the transfer intensity to avoid model degradation.

[0052] Graph-level transfer adjustment is achieved through learnable gating functions, which control the proportion of cross-domain information sharing, thereby enabling dynamic graph embedding fusion. in, g By controlling the proportion of cross-domain information sharing, dynamic graph embedding and fusion can be achieved.

[0053] Furthermore, this invention introduces a cross-domain knowledge distillation mechanism: Further improve the semantic consistency after the target domain is migrated.

[0054] Step 4: Construct a privacy-enhancing strategy network model. Input the target domain graph into the privacy-enhancing strategy network model, and output the predicted privacy-enhancing transaction decision through multi-objective collaborative optimization.

[0055] Specifically, an intelligent privacy decision-making module is constructed, which predicts privacy policy output through privacy cost constraints, noise injection, and policy smoothing. Multiple losses are introduced, including prediction task loss, cross-domain distribution alignment loss, privacy enhancement loss, and structural consistency loss. A multi-objective collaborative optimization objective function is constructed. By calculating the multi-objective collaborative optimization objective function, unified modeling, efficient migration, and privacy-friendly decision-making of cross-domain transaction behavior are achieved.

[0056] As one embodiment, the electricity spot trading disclosed herein involves sensitive information such as the marginal cost of generating units, dispatch strategies, and regional supply and demand relationships. Directly using this information in decision-making models could easily lead to privacy leaks. This disclosure constructs an intelligent privacy-preserving decision-making module, achieving privacy-friendly policy output through privacy cost constraints, noise injection, and policy smoothing.

[0057] Modeling privacy breach risk as policy gradient sensitivity: Local Differential Privacy (LDP) is used to perturb the node embedding: To avoid policy oscillations caused by noise, a policy smoothing regularization term is designed: Here, s' represents the state after a slight perturbation, used to enhance privacy, security, and stability.

[0058] The ultimate goal of privacy-enhancing decision optimization is: in, It can be defined as a strategy return, price prediction accuracy, or scheduling utility function.

[0059] As one embodiment, this disclosure constructs a multi-task collaborative optimization objective function, integrating multiple optimization objectives such as prediction tasks, cross-domain alignment, privacy enhancement, and structural consistency, to build a unified collaborative optimization framework: F= in, Loss forecasting for spot prices / strategy For cross-domain distribution alignment loss, For graph structure consistency loss, For cross-domain knowledge distillation feature alignment loss, To mitigate the risk of privacy breaches and minimize losses, To smooth out the regularized loss for the strategy.

[0060] As one embodiment, through this collaborative optimization objective function, this disclosure can achieve unified modeling, efficient migration, and privacy-friendly decision-making for cross-domain transaction behavior. The specific algorithm is as follows: Simulation Experiment This disclosure includes comparative experiments. Table 1 shows the comparison of prediction accuracy and robustness between this disclosure and existing technologies in typical cross-domain scenarios. Table 2 shows the feature migration stability index (mean ± standard deviation) under different cross-domain difference types (load difference, structural difference, rule difference). Table 3 shows the actual deployment effect (practicability evaluation) under different electricity market mechanisms.

[0061] Table 1 Comparison of prediction accuracy and robustness in typical cross-domain scenarios. Table 2. Characteristic migration stability index (mean ± standard deviation) under different cross-domain difference types (load difference, structural difference, rule difference). Table 3. Actual Deployment Effects under Different Electricity Market Mechanisms (Practicality Evaluation) Example 3 One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned privacy decision-making method for electricity spot trading based on cross-domain graph representation.

[0062] Example 4 One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the aforementioned privacy decision-making method for electricity spot trading based on cross-domain graph representation.

[0063] Example 5 One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the privacy decision-making method for electricity spot trading based on cross-domain graph representation.

[0064] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A privacy-focused decision-making method for electricity spot trading based on cross-domain graph representation, characterized in that, include: Key elements for accessing the electricity spot market; Based on the structural relationships between various key elements, a graph model for electricity spot trading is constructed, and the graph structure is subjected to unified representation learning to obtain node embeddings; By using a cross-domain consistency alignment mechanism to align the distribution of embedded nodes, calculating the minimum distribution difference, and employing a node semantic difference stripping mechanism to decouple shared features from domain-private features, a unified semantic space shared by the two domains is constructed to achieve cross-domain alignment. After cross-domain alignment, adaptive cross-domain migration is performed. The migration intensity is dynamically adjusted according to the cross-domain embedding differences. The migration ratio is controlled by node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. A privacy-enhancing strategy network model is constructed. The target domain graph is input into the privacy-enhancing strategy network model, and through multi-objective collaborative optimization, the predicted privacy-enhancing transaction decision is output.

2. The privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in claim 1, characterized in that, The key elements of the electricity spot market include generating units, load nodes, regional markets, transmission channels, and trading relationships.

3. The privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in claim 1, characterized in that, The process involves constructing a graph model for electricity spot trading based on the structural relationships between key elements, and performing unified representation learning on the graph structure to obtain node embeddings, including: Electricity spot trading graph models are constructed in the source and target domains respectively to depict the structural relationships between key elements such as generating units, load nodes, regional markets, transmission channels, and trading relationships. Define the source domain and target domain graphs as follows: G s =( V s , E s , X s )and G t =( V t , E t , X t ),in V For a set of nodes, E Let be the set of edges. X For node or edge attributes; A graph neural network is used for unified representation learning to obtain node embeddings in the source and target domains.

4. The privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in claim 1, characterized in that, The method involves using a cross-domain consistency alignment mechanism to align node embeddings, calculating and minimizing distribution differences, employing a node semantic difference stripping mechanism to decouple shared features from domain-private features, and constructing a unified semantic space shared by both domains to achieve cross-domain alignment. This includes: First, the overall feature distribution is aligned using the maximum mean difference or Wasserstein distance. By introducing graph structure consistency constraints, constructing structure matching loss in edge space, and using a learnable mapping function, the graph structure of the target domain is mapped to a semantic space similar to that of the source domain. A node semantic difference stripping mechanism is adopted. By decoupling shared features from domain-private features and introducing orthogonal constraints, it ensures that cross-domain common information is preserved and domain-specific features are stripped during the migration process.

5. The privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in claim 1, characterized in that, After cross-domain alignment, adaptive cross-domain migration is performed. The migration intensity is dynamically adjusted based on the cross-domain embedding differences. The migration ratio is controlled by node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion, resulting in a target domain graph, including: Based on the cross-domain embedding differences, dynamic weights at the node and graph levels are designed. When the differences are large, the migration weights at the node level automatically reduce the migration intensity. Graph-level migration adjustment is achieved through learnable gating functions, which control the proportion of cross-domain information sharing, thereby enabling dynamic graph embedding fusion.

6. The privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in claim 1, characterized in that, An intelligent privacy decision-making module is constructed, which predicts privacy policy output through privacy cost constraints, noise injection, and policy smoothing. It introduces various losses, including prediction task loss, cross-domain distribution alignment loss, privacy enhancement loss, and structural consistency loss, and constructs a multi-objective collaborative optimization objective function. By calculating the multi-objective collaborative optimization objective function, it achieves unified modeling, efficient transfer, and privacy-friendly decision-making for cross-domain transaction behavior.

7. A privacy-preserving decision-making system for electricity spot trading based on cross-domain graph representation, characterized in that, include: The data acquisition module is used to acquire key elements of the electricity spot market; The representation learning module is used to construct a graph model of electricity spot trading based on the structural relationships between various key elements, and to perform unified representation learning on the graph structure to obtain node embeddings; The cross-domain alignment module is used to embed nodes for distribution alignment using a cross-domain consistency alignment mechanism, calculate the minimum distribution difference, adopt a node semantic difference stripping mechanism, decouple shared features from domain-private features, construct a unified semantic space shared by the two domains, and achieve cross-domain alignment. The adaptive migration module is used to perform adaptive cross-domain migration after cross-domain alignment. It dynamically adjusts the migration intensity based on the differences in cross-domain embedding and controls the migration ratio through node-level attention weights and graph-level gating to achieve dynamic graph embedding fusion and obtain the target domain graph. The strategy decision module is used to construct a privacy-enhancing strategy network model. The target domain graph is input into the privacy-enhancing strategy network model, and through multi-objective collaborative optimization, the predicted privacy-enhancing transaction decision is output.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the privacy decision-making method for electricity spot trading based on cross-domain graph representation as described in any one of claims 1-6.

Images