A power spot market collaborative decision-making method based on a trusted data space

Abstract

The application discloses a power spot market collaborative decision-making method based on a trusted data space, comprising the following steps: S1, connecting each market subject of the power spot market, realizing saving of private data of the market subject in a local place, and accessing a trusted data space service platform through a trusted data space connector; S2, constructing the trusted data space service platform, realizing data access and dynamic authorization, and trusted auditing based on a block chain, so that access and use of data meet the requirements; S3, after a global model initialized by a federal learning server is collaboratively trained to convergence through privacy protection of the corresponding client, only encrypted global model parameters are uploaded to obtain an updated global model through aggregation and are deployed on the federal learning server; and S4, the updated global model is deployed as a SaaS service to provide a bidding decision for authorized users. The application improves the prediction performance, global dependence and privacy security guarantee.

Description

Technical Field

[0001] This application relates to the field of power auxiliary decision-making technology, and in particular, to a collaborative auxiliary decision-making method for the power spot market based on a trusted data space. Background Technology

[0002] With the deepening of my country's power market reform, several provinces have launched trial operations of the electricity spot market, and market participants have an increasingly urgent need for accurate electricity price forecasts. However, existing forecasting methods still largely rely on limited data from a single entity, making it difficult to capture the overall dynamics of the market. Especially during the trial operation phase, immature market mechanisms and fragmented, heterogeneous data further constrain the performance of forecasting models. How to achieve collaborative utilization of cross-domain data while ensuring data privacy and the rights of stakeholders has become a key challenge for improving the operational efficiency and intelligent decision-making of the power market. The shortcomings of existing technologies are as follows: 1. Deep learning models, due to their powerful nonlinear fitting and feature learning capabilities, have gradually become the mainstream method in this field. Among various deep learning models, recurrent neural networks and their variants, such as long short-term memory networks and gated recurrent units, are widely used because they can effectively capture temporal dependencies. However, these models suffer from inherent bottlenecks such as gradient vanishing and difficulties in parallel computation when processing long sequences, which limit their performance ceiling. 2. Although convolutional neural networks and multilayer perceptrons can assist in feature extraction, they are still insufficient in modeling global dependencies. 3. In recent years, the Transformer model, with its self-attention mechanism, has demonstrated significant advantages in long-range dependency modeling. This mechanism can compute global correlations of sequences in parallel, greatly improving the processing capabilities of time-series data. However, the application of existing Transformer models is limited to mining publicly available historical data, based on the ideal assumption that data can be centrally obtained, failing to address the most severe challenge in the electricity market: the "data silo" problem. Key features such as generator operating status, grid congestion information, and flexible loads on the user side are scattered across different market participants. Due to data privacy and trade secrets, these features are difficult to utilize by centralized prediction models, which constitutes a fundamental obstacle to improving the performance of existing models. Summary of the Invention

[0003] This application provides a collaborative auxiliary decision-making method for the electricity spot market based on trusted data space, which solves the technical problems of limited performance, insufficient global dependence, and lack of privacy and security protection in existing technologies.

[0004] This application is achieved through the following solution: A collaborative decision-making support method for the electricity spot market based on trusted data space includes the following steps: S1. Connect with various market participants in the electricity spot market, enabling them to store their private data locally and access the Trusted Data Space service platform through the Trusted Data Space Connector; S2. Based on the national trusted data space technology architecture, build a trusted data space service platform to provide core services including identity authentication, smart contract management, metadata management, access connector management, and transaction clearing, realize data access and dynamic authorization, and trusted auditing based on blockchain, so that the access and use of all data meet the data owner's preset permission policy. S3. In a trusted data space environment, the initial global model is distributed by the federated learning server to the selected market entity's client for privacy-preserving federated learning collaborative training until convergence. Then, only the encrypted global model parameters are uploaded for aggregation to obtain the updated global model, which is then deployed to the federated learning server. The initial global model adopts an improved Transformer model, which uses non-autoregressive decoding and power time-series sensing location encoding. S4. Deploy the updated global model as a SaaS service to provide authorized users with multi-step time-series electricity price forecasting and risk warning services, assisting entities in making pricing decisions.

[0005] Furthermore, in step S2, the data access and dynamic authorization specifically include the following steps: S201. Metadata Publishing and Registration: Data providers publish their shareable data resources in the form of metadata to the metadata management module of the Trusted Data Space Service Platform through their local Trusted Data Space Connector. The metadata only describes the basic information of the data and does not contain any specific sensitive data values. S202. Usage Policy Definition: When publishing metadata, provide: attach a machine-readable usage policy to the data, which defines the authorized usage conditions of the data in the form of rules; S203, Dynamic Authorization and Contract Agreement: When a data user discovers the required data through the directory service, it initiates a data access request to the provider's trusted data space connector. The provider's trusted data space connector verifies the user's digital identity certificate and evaluates whether its request complies with the preset usage policy. If the verification is successful, the trusted data space connectors of both parties will automatically negotiate and generate a legally binding smart contract, recording the details of this authorization. After the smart contract is uploaded to the blockchain, a connection is established between the connectors.

[0006] Furthermore, in step S2, the blockchain-based trusted audit specifically includes the following steps: S211. Key events are recorded on the blockchain, including defining key behaviors, identifying key behaviors during system operation, converting them into hash values, and recording them on the blockchain. S212. Construct a trust chain, including linking key on-chain behaviors in chronological order into a complete and tamper-proof trust chain through the inherent logic of the blockchain.

[0007] Further, in step S211, the key actions include data registration events, contract completion events, and model usage time. The data registration events include publishing metadata and usage policies; the contract completion events include granting data usage authorization and recording the hash value of the smart contract; and the model usage time includes recording the data credentials, model version number, and hash value of the prediction result called for this prediction when the finally trained model is used for a prediction task.

[0008] Furthermore, in step S212, after the trust chain is established, any participant is allowed to independently verify the following: whether a prediction result comes from a compliantly trained model, whether the model legally used authorized data during training, and whether the entire process complies with the usage policies set by each data provider.

[0009] Further, in step S3, the improved Transformer model includes improvements to the model structure and output layer, and optimization of the decoder. The improvements to the model structure and output layer include introducing a dedicated fully connected layer as a feature focusing layer after the high-dimensional sequence representation output by the decoder, which is used to map the high-dimensional features rich in temporal dependency information to a low-dimensional feature space focused on electricity price prediction, and using a linear projection layer as the final linear output layer. At the same time, the activation function LeakyReLU is restricted to after the feature focusing layer and before the linear output layer. The optimization of the decoder includes: using a learnable query matrix for input initialization to form a non-autoregressive decoder input mechanism, using fully parallel forward computation during computation, and mapping the entire prediction sequence at once through a shared linear projection layer during output.

[0010] Furthermore, the improvements to the Transformer model include input representation and preprocessing of multi-source heterogeneous data, adaptation of location encoding and power time series, and improvements to the model structure and output layer. The input representation and preprocessing of the multi-source heterogeneous data includes numerical continuous feature preprocessing, non-numerical state feature preprocessing, and sequence construction and alignment. Numerical continuous features, including historical electricity prices, load, wind speed, and temperature, undergo local normalization. Non-numerical state features, including unit start-up / shutdown status, date type, and time period labels, cannot be directly input into the model; therefore, an "embedding layer" is used to map them into dense vector representations. Sequence construction and alignment concatenate all processed features, including normalized numerical features and embedded categorical features, according to time points to form a unified feature vector, and construct the input sequence according to the set historical window length. Each of them x t They all contain all available multi-dimensional information at that moment; The adaptation of the location coding to the power timing sequence retains the original sine and cosine coding for the basic location coding to capture the relative location information within the sequence; for the time-aware location coding, explicit time features are embedded and used as a supplement to or partial replacement of the location coding. The improvements to the model structure and output layer include: introducing a dedicated fully connected layer as a feature focusing layer after the high-dimensional sequence representation output by the decoder, mapping the high-dimensional features rich in time-dependent information to a low-dimensional feature space focused on electricity price prediction, effectively filtering redundant information; using a linear projection layer as the final linear output layer to directly map the compressed feature vector to the continuous electricity price prediction value at a future set time point; limiting the activation function LeakyReLU to after the feature focusing layer and before the linear output layer; optimizing the decoder input by using a learnable query matrix for input initialization to form a non-autoregressive decoder input mechanism, employing fully parallel forward computation during computation, and mapping the decoder output to the entire prediction sequence at once through a shared linear projection layer.

[0011] Furthermore, during local normalization, each client calculates the mean and standard deviation of its own features locally and performs Z-score normalization. ; in, and These are the mean and standard deviation of the client's local training data, respectively. The formula for adapting the location code to the power timing sequence is as follows: ; in, Represents the absolute position index in the sequence (e.g., time step 0, time step 1); This represents the enhanced position encoding vector; Represents a linear transformation; This represents a vector concatenation operation; The standard sine and cosine position encoding vector; Represents an embedding operation. Representing the The embedding vector corresponding to the hour; Representing the The embedding vector corresponding to the sky.

[0012] Furthermore, the input is initialized using a learnable query matrix: ; in, n steps Indicates the length of the predicted sequence; Indicates the dimension of the hidden layer of the model; This represents a trainable parameter matrix, which is randomly initialized at the start of training and optimized along with other parameters of the model during training. The fully parallel forward computation of the decoder is as follows: ; in, Decoder (•) represents the decoding module, which includes multi-head self-attention, encoder-decoder attention, and feedforward network; , represents the hidden state matrix output by the decoder, where each row contains prediction information for the corresponding future time step i; When the decoder outputs, the hidden state matrix is... The entire predicted sequence is mapped in one step using a shared linear projection layer, as shown in the formula: ; in, This represents the weight matrix of the output layer; This represents the entire prediction sequence generated at once.

[0013] Further, in step S3, the step of performing privacy-preserving federated learning collaborative training of the initial global model, which is distributed by the federated learning server to the selected market entity's client in a trusted data space environment, until convergence, and then only uploading the encrypted global model parameters for aggregation to obtain the updated global model and deploying it to the federated learning server, specifically includes the following steps: S31. Task Initialization and Model Distribution: The federated learning server defines the training task and initializes an improved Transformer model as the initial global model. At the same time, it distributes the structure and initial weights of the initial global model to all clients of the market entities authorized to participate in this training task through broadcasting. S32. Local Model Training: After receiving the initial global model, each client trains it on its local private data. The training data includes not only its own data, but also relevant authorized data obtained in real time and on demand from the Trusted Data Space Service Platform. After training is completed, each market entity's client will receive a locally updated global model. S33. Model Parameter Upload and Aggregation: Each market entity's client encrypts the parameter updates of the locally updated global model and sends them to the federated learning server. Without decrypting the updates from individual clients, the federated learning server uses a secure aggregation protocol to calculate the sum of all parameter update values. Then, it applies the aggregated parameter update values ​​to the initial global model on the federated learning server to generate the updated global model. S34. Global Model Evaluation and Distribution: The federated learning server evaluates the performance of the updated global model through the validation set. When the performance no longer improves significantly in multiple consecutive iterations, or reaches the preset maximum number of iterations, it is determined that the updated global model has converged and training is terminated. Otherwise, the federated learning server distributes the updated global model to all participants and starts the next round of training. This process is repeated until the updated global model converges. S35. Deploy the final trained and updated global model to the federated learning server and access the trusted data space as a data service to form a prediction API for external service.

[0014] This application also provides a collaborative auxiliary decision-making method for the electricity spot market based on trusted data space, including: The data docking and storage module is used to dock with various market participants in the electricity spot market, enabling them to store their private data locally and connect to the trusted data space service platform of the trusted data space through the trusted data space connector; The data security supervision module is used to build a trusted data space service platform based on the national trusted data space technology architecture. It provides core services including identity authentication, smart contract management, metadata management, access connector management, and transaction clearing, and realizes data access and dynamic authorization, and trusted auditing based on blockchain, so that the access and use of all data meet the data owner's preset permission policy. The global model federated learning module is used in a trusted data space environment to perform privacy-preserving federated learning collaborative training of the initial global model issued by the federated learning server to the selected market entity's client. After the initial global model converges, only the encrypted global model parameters are uploaded for aggregation to obtain an updated global model, which is then deployed to the federated learning server. The initial global model adopts an improved Transformer model, which uses non-autoregressive decoding and power time-series-aware location encoding. The global model deployment module is used to deploy the updated global model as a SaaS service, providing authorized users with multi-step time-series electricity price forecasting and risk warning services to assist entities in making pricing decisions.

[0015] This application also provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the aforementioned collaborative auxiliary decision-making method for the electricity spot market based on trusted data space.

[0016] This application also provides a computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the aforementioned collaborative auxiliary decision-making method for the electricity spot market based on a trusted data space.

[0017] Compared with the prior art, this application has the following advantages: 1. This application adopts a fusion technology model of "improved Transformer + federated learning". The self-attention mechanism of Transformer is mathematically equivalent to a fully connected graph model, which can perfectly characterize the global dependency relationship between any two time points and any two features in the market, and can significantly improve the accuracy of the model. Federated learning, through distributed optimization, theoretically shares the same global optimal solution with the objective function of centralized learning, which can significantly improve the performance of the model. 2. This application introduces a trusted data space as the infrastructure for distributed data collaborative utilization in the electricity spot market, separating the right to use data from the right to control it. Without moving the physical ownership of the data, it realizes the secure and controllable transfer of the virtual right to use data, fundamentally solving the trust problem of electricity data sharing. This allows for the utilization of key characteristic data such as the operating status of generator units, grid congestion information, and flexible loads on the user side, which are scattered among different market participants, further improving the performance of the model. 3. This application designs an operating mechanism of "dynamic authorization-secure aggregation-trusted auditing". This mechanism integrates game theory and cryptography principles. Dynamic authorization and smart contracts ensure that the rational person assumption of the participants is satisfied, that is, the behavior is predictable and the rules are transparent. Secure aggregation based on differential privacy / homomorphic encryption provides privacy and security protection. Blockchain-based auditing utilizes its cryptographic immutability to build a trust environment that does not require third-party endorsement.

[0018] In addition to the purposes, features, and advantages described above, this application has other purposes, features, and advantages. A further detailed description of this application will be provided below with reference to the figures. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a flowchart illustrating a preferred embodiment of the collaborative auxiliary decision-making method for the electricity spot market based on a trusted data space. Figure 2 This is a schematic diagram of the privacy-preserving federated learning process according to a preferred embodiment of this application; Figure 3 This is a schematic diagram illustrating the composition principle of the collaborative auxiliary decision-making framework for the electricity spot market based on trusted data space, according to a preferred embodiment of this application. Figure 4 This is a schematic diagram of a module of a collaborative auxiliary decision-making device for the electricity spot market based on a trusted data space, according to a preferred embodiment of this application. Figure 5 This is a schematic block diagram of an electronic device according to a preferred embodiment of this application; Figure 6 This is an internal structural diagram of a computer device according to a preferred embodiment of this application. Detailed Implementation

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] Explanation of technical terms Trusted Data Space: A trusted data space is a data circulation and utilization infrastructure based on consensus rules, connecting multiple entities to achieve data resource sharing and utilization. In a trusted data space, data providers always retain control over their own data, while data users can only use the data within the authorized scope and according to agreed conditions.

[0023] Transformer Model: The Transformer is a neural network architecture built on a self-attention mechanism, capable of capturing global dependencies at different positions in a sequence. In the problem of predicting the clearing price in the electricity spot market, the Transformer can extract global features from the input electricity market-related data, thereby capturing the implicit relationships and complex patterns between different time points and various factors. This patent further improves upon existing Transformer models to enhance prediction accuracy.

[0024] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a collaborative auxiliary decision-making device for the electricity spot market based on a trusted data space capable of achieving the above functions. The following description uses a collaborative auxiliary decision-making device for the electricity spot market based on a trusted data space as an example to illustrate this embodiment and the subsequent embodiments.

[0025] like Figure 1 As shown, a preferred embodiment of this application provides a collaborative auxiliary decision-making method for the electricity spot market based on a trusted data space, including the following steps: S1. Connect with various market participants in the electricity spot market, enabling them to store their private data locally and access the Trusted Data Space service platform through the Trusted Data Space Connector; S2. Based on the national trusted data space technology architecture, build a trusted data space service platform to provide core services including identity authentication, smart contract management, metadata management, access connector management, and transaction clearing, realize data access and dynamic authorization, and trusted auditing based on blockchain, so that the access and use of all data meet the data owner's preset permission policy. S3. In a trusted data space environment, the initial global model is distributed by the federated learning server to the selected market entity's client for privacy-preserving federated learning collaborative training until convergence. Then, only the encrypted global model parameters are uploaded for aggregation to obtain the updated global model, which is then deployed to the federated learning server. The initial global model adopts an improved Transformer model, which uses non-autoregressive decoding and power time-series sensing location encoding. S4. Deploy the updated global model as a SaaS service to provide authorized users with multi-step time-series electricity price forecasting and risk warning services, assisting entities in making pricing decisions.

[0026] This embodiment relates to auxiliary decision-making in the electricity spot market. It can be used for electricity price forecasting in the electricity spot market, assisting power generation companies, electricity sales companies, and electricity users in quantity and price reporting, risk management, and business decision-making. This embodiment makes specific structural improvements to the existing Transformer model for the characteristics of electricity time-series data, and combines the trusted data space with the improved prediction model. It provides market participants with a feasible path to obtain richer information and make better decisions while protecting privacy, helping to improve the resource allocation efficiency of the entire electricity market and solving pain points such as "data not being shared, inaccurate models, and unintelligent decision-making" in actual business.

[0027] (1) To address the limitations on prediction accuracy caused by gradient vanishing and parallel computation difficulties in recurrent neural networks and their variant models, an improved Transformer model is designed to enhance prediction efficiency and accuracy.

[0028] (2) Design a “Federal Electricity Price Forecasting Framework Based on Trusted Data Space” to address the shortcomings of existing models in terms of global dependency.

[0029] (3) Design a dynamic, authorized high-value feature usage mechanism based on a trusted data space to solve the problems of existing models relying on publicly available historical data and insufficient utilization of scattered multi-source heterogeneous data, which makes it difficult to improve model performance. In addition, introduce a blockchain-based trusted audit mechanism to form an immutable audit trail and solve the "black box" trust problem of AI models.

[0030] Compared with the prior art, this embodiment has the following beneficial effects: 1. This embodiment adopts a fusion technology of "improved Transformer + federated learning". The self-attention mechanism of Transformer is mathematically equivalent to a fully connected graph model, which can perfectly characterize the global dependency relationship between any two time points and any two features in the market, and can significantly improve the accuracy of the model. Federated learning, through distributed optimization, theoretically shares the same global optimal solution with the objective function of centralized learning, which can significantly improve the performance of the model. 2. This embodiment introduces a trusted data space as the infrastructure for distributed data collaborative utilization in the electricity spot market, separating the right to use data from the right to control it. Without moving the physical ownership of the data, it realizes the secure and controllable transfer of the virtual right to use data, fundamentally solving the trust problem of electricity data sharing. This allows for the utilization of key characteristic data such as the operating status of generator units, grid congestion information, and flexible loads on the user side, which are scattered among different market entities, further improving the performance of the model. 3. This embodiment designs an operating mechanism of "dynamic authorization-secure aggregation-trusted auditing". This mechanism integrates game theory and cryptography principles. Dynamic authorization and smart contracts ensure that the rational person assumption of the participants is satisfied, that is, the behavior is predictable and the rules are transparent. Secure aggregation based on differential privacy / homomorphic encryption provides privacy and security protection. Blockchain-based auditing utilizes its cryptographic immutability to build a trust environment that does not require third-party endorsement.

[0031] Preferably, in step S2, the data access and dynamic authorization specifically include the following steps: S201. Metadata Publishing and Registration: Data providers publish their shareable data resources in the form of metadata to the metadata management module of the Trusted Data Space Service Platform through their local Trusted Data Space Connector. The metadata only describes the basic information of the data and does not contain any specific sensitive data values. S202. Usage Policy Definition: When publishing metadata, provide: attach a machine-readable usage policy to the data. This policy is a concrete manifestation of data sovereignty. The usage policy defines the conditions for authorized use of the data in the form of rules. S203, Dynamic Authorization and Contract Agreement: When a data user discovers the required data through the directory service, it initiates a data access request to the provider's trusted data space connector. The provider's trusted data space connector verifies the user's digital identity certificate and evaluates whether its request complies with the preset usage policy. If the verification is successful, the trusted data space connectors of both parties will automatically negotiate and generate a legally binding smart contract, recording the details of this authorization. After the smart contract is uploaded to the blockchain, a connection is established between the connectors.

[0032] This embodiment designs a data access and dynamic authorization mechanism to solve the technical problem of "whether the data can be used" and ensure the reliability of the collaborative auxiliary decision-making system based on a trusted data space.

[0033] Preferably, in step S2, the blockchain-based trusted audit specifically includes the following steps: S211. Key events are recorded on the blockchain, including defining key behaviors, identifying key behaviors during system operation, converting them into hash values, and recording them on the blockchain. S212. Construct a trust chain, including linking key on-chain behaviors in chronological order into a complete and tamper-proof trust chain through the inherent logic of the blockchain.

[0034] This embodiment uses a blockchain-based trusted auditing mechanism to ensure that all operations are traceable, verifiable, and tamper-proof by recording key events on the blockchain and building a trust chain.

[0035] Specifically, in step S211, the key actions include data registration events, contract completion events, and model usage time. The data registration events include publishing metadata and usage policies; the contract completion events include granting data usage authorization and recording the hash value of the smart contract; and the model usage time includes recording the data credentials (pointing to the previously completed contract), model version number, and hash value of the prediction result when the finally trained model is used for a prediction task.

[0036] Specifically, in step S212, after the trust chain is established, any participant is allowed to independently verify the following: whether a prediction result comes from a compliantly trained model, whether the model legally used authorized data during training, and whether the entire process complies with the usage policies set by each data provider.

[0037] Through the aforementioned blockchain-based trusted auditing mechanism, regulatory agencies can be provided with transparent regulatory tools to conduct audits; data ownership protection and data usage tracking can be provided to enhance the willingness of data providers to share data; at the same time, data users can be provided with data usage credentials to improve the credibility and value of decision-making results.

[0038] Preferably, in step S3, the improved Transformer model includes improvements to the model structure and output layer, and optimization of the decoder. The improvements to the model structure and output layer include introducing a dedicated fully connected layer as a feature focusing layer after the high-dimensional sequence representation output by the decoder, which is used to map the high-dimensional features rich in temporal dependency information to a low-dimensional feature space focused on electricity price prediction, and using a linear projection layer as the final linear output layer. At the same time, the activation function LeakyReLU is restricted to after the feature focusing layer and before the linear output layer. The optimization of the decoder includes: using a learnable query matrix for input initialization to form a non-autoregressive decoder input mechanism, using fully parallel forward computation during computation, and mapping the entire prediction sequence at once through a shared linear projection layer during output.

[0039] This embodiment addresses the limitations on prediction accuracy caused by recurrent neural networks and their variants, such as gradient vanishing and difficulties in parallel computing. It makes specific structural improvements to the existing Transformer model to suit the characteristics of power time-series data, and combines a reliable data space with the improved prediction model. This provides market participants with a feasible path to obtain richer information and make better decisions while protecting privacy, thereby helping to improve the resource allocation efficiency of the entire power market and solving pain points such as "data not being shared, inaccurate models, and unintelligent decision-making" in actual business operations.

[0040] Preferably, the improved Transformer model includes input representation and preprocessing of multi-source heterogeneous data, location encoding and adaptation to power time series, and improvements to the model structure and output layer. Multi-source data from the power spot market have different dimensions, distributions, and physical meanings; directly inputting them into the model can lead to optimization difficulties and numerical instability. Therefore, refined preprocessing is required. Thus, the input representation and preprocessing of the multi-source heterogeneous data includes preprocessing of numerical continuous features, preprocessing of non-numerical state features, and sequence construction and alignment, wherein: Numerical continuous features include historical electricity prices, load, wind speed, and temperature. These are normalized locally. During local normalization, each client calculates the mean and standard deviation of its own features locally and then performs Z-score normalization. ; in, and These are the mean and standard deviation of the client's local training data, respectively. The non-numerical state features, including unit start-up / shutdown status, date type, and time period labels, cannot be directly input into the model. An "embedding layer" maps these features into dense vector representations. For example, the seven categories from "Monday" to "Sunday" can be converted into seven D-dimensional vectors using a trainable lookup table. This allows the model to learn the semantic relationships between categories. In the trusted energy data space, power generation companies can contribute their "unit status" embedding vectors, providing crucial supply-side information to the model and overcoming the bottleneck of relying solely on demand-side data for prediction. The sequence construction and alignment are used to concatenate all processed features, including normalized numerical features and embedded categorical features, according to time points to form a unified feature vector, and construct the input sequence according to the set historical window length (e.g., 96 time points). Each of them x t They all contain all available multi-dimensional information at that moment; The original Transformer uses sine and cosine positional encoding to inject sequential information into the sequence. Considering the strong periodicity and seasonality of power data, this embodiment improves and enhances it as follows: The adaptation of the positional encoding to the power time series retains the original sine and cosine encoding for the basic positional encoding to capture relative positional information within the sequence; for time-aware positional encoding, explicit time features (such as the hour of the day or the day of the week) are embedded and used as a supplement or partial replacement for the positional encoding. This allows the model to explicitly learn prior time patterns such as "morning peak," "evening peak," and "weekend off-peak," accelerating convergence and improving generalization ability. The calculation formula for the adaptation of the positional encoding to the power time series is as follows: ; in, Represents the absolute position index in the sequence (e.g., time step 0, time step 1); This represents the enhanced position encoding vector; Represents a linear transformation; This represents a vector concatenation operation; The standard sine and cosine position encoding vector; Represents an embedding operation. Representing the The embedding vector corresponding to the hour; Representing the The embedding vector corresponding to the day; The improvements to the model structure and output layer include: First, after the high-dimensional sequence representation output by the decoder, a dedicated fully connected layer is introduced as a feature focusing layer to map the high-dimensional features rich in temporal dependency information to a low-dimensional feature space focused on electricity price prediction, effectively filtering redundant information; then, a linear projection layer is used as the final linear output layer to directly map the compressed feature vector to continuous electricity price prediction values ​​for the next 48 time points, as detailed below: ; The above design ensures in principle that the model output is a continuous value without distortion. Then, unlike existing works that place the activation function LeakyReLU in the output layer, this embodiment strictly limits the activation function LeakyReLU after the feature focusing layer and before the linear output layer. This ensures that LeakyReLU is only used to introduce nonlinearity, enhance the model's expressive power, and prevent gradient vanishing. At the same time, it guarantees the continuity and unboundedness of the final output value, which meets the requirements of regression tasks, thereby improving the model's numerical stability and convergence speed. Most existing Transformer-based prediction models retain the autoregressive generation method of the decoder in the original architecture, i.e., generating target sequences one by one. This method has inherent drawbacks such as low computational efficiency and error accumulation in multi-step prediction scenarios in the power market. In this embodiment, the decoder input is optimized by using a learnable query matrix for input initialization to form a non-autoregressive decoder input mechanism. The learnable query matrix is ​​used for input initialization. ; in, n steps Indicates the length of the predicted sequence; Indicates the dimension of the hidden layer of the model; This represents a trainable parameter matrix. The query matrix is ​​randomly initialized at the start of training and optimized along with other parameters of the model during training. Optimizing the input to the decoder not only improves the training and inference efficiency of the model, making it more suitable for power trading decisions with extremely high real-time requirements, but also avoids the gradual accumulation of errors in the autoregressive process, and shows significant advantages in terms of stability and accuracy in long-term prediction. The computation employs fully parallel forward computation. The decoder output is mapped to the entire prediction sequence in one go through a shared linear projection layer. The decoder utilizes all historical context information provided by the encoder to generate the entire prediction sequence for the next 48 steps in one go and in parallel, eliminating the inherent sequence delay of traditional autoregressive models and improving inference speed by nearly [missing information]. times ( (For the predicted sequence length), the fully parallel forward computation of the decoder is as follows: ; in, Decoder (•) represents the decoding module, which includes multi-head self-attention, encoder-decoder attention, and feedforward network; , represents the hidden state matrix output by the decoder, where each row contains the prediction information for the corresponding future time step i. Since all positions in the query matrix are input simultaneously and there is no temporal dependency between them, the output can be computed in parallel. When the decoder outputs, the hidden state matrix is... The entire predicted sequence is mapped in one step using a shared linear projection layer, as shown in the formula: ; in, This represents the weight matrix of the output layer; Represents the entire prediction sequence generated at once. .

[0041] Preferably, in step S3, the step of performing privacy-preserving federated learning collaborative training of the initial global model, which is distributed by the federated learning server to the selected market entity's client in a trusted data space environment, until convergence, and then only uploading the encrypted global model parameters for aggregation to obtain the updated global model and deploying it to the federated learning server, specifically includes the following steps: S31. Task initialization and model distribution: The federated learning server (coordinator) defines the training task and initializes an improved Transformer model as the initial global model. At the same time, it distributes the structure and initial weights of the initial global model to all clients of the market entities authorized to participate in this training task through broadcast. S32. Local Model Training: After receiving the initial global model, each client trains it on its local private data. The training data includes not only its own data, but also relevant authorized data obtained in real time and on demand from the Trusted Data Space Service Platform. After training is completed, each market entity's client will receive a locally updated global model. S33. Model Parameter Upload and Aggregation: Each market entity's client encrypts the parameter updates of the locally updated global model and sends them to the federated learning server. Without decrypting the updates from individual clients, the federated learning server uses a secure aggregation protocol to calculate the sum of all parameter update values. Then, it applies the aggregated parameter update values ​​to the initial global model on the federated learning server to generate the updated global model. S34. Global Model Evaluation and Distribution: The federated learning server evaluates the performance of the updated global model through the validation set. When the performance no longer improves significantly in multiple consecutive iterations, or reaches the preset maximum number of iterations, it is determined that the updated global model has converged and training is terminated. Otherwise, the federated learning server distributes the updated global model to all participants and starts the next round of training. This process is repeated until the updated global model converges. S35. Deploy the final trained and updated global model to the federated learning server and access the trusted data space as a data service to form a prediction API for external service.

[0042] This embodiment involves a privacy-preserving federated learning process. In a trusted data space environment, it designs a specific federated learning process for collaboratively training and improving the Transformer model, achieving a balance between model performance, communication efficiency, and privacy security. The core algorithm adopts the classic federated averaging framework and is adapted for the characteristics of electricity time-series data. The entire process is a multi-round iterative process coordinated by a central server, such as... Figure 2 As shown, the detailed steps are as follows: (1) Process initialization phase, including: ① Global model definition and initialization: The federated learning server first defines the architecture of the improved Transformer model, including hyperparameters such as the number of encoder / decoder layers, the number of attention heads, and the dimension of the feedforward network. The federated learning server then uses a specific stochastic initialization strategy to initialize the weights of the global model. W 0.

[0043] ② Client registration and configuration: Eligible market participants (clients) register with the Federated Learning Server and declare the types of data they can provide (e.g., "can provide historical load data", "can provide unit status data"). The Federated Learning Server establishes a secure communication channel with each client and exchanges digital certificates, laying the foundation for subsequent secure transmission.

[0044] (2) The core iterative training phase includes: In the At the start of each round of training, the federated learning server randomly selects a subset of clients based on a specific selection strategy. The current global model weights Distribute to selected clients. Each client Before training begins, based on the needs of its training task, it obtains the required, authorized feature data from other participants in real time through the data access and dynamic authorization mechanism of the trusted data space. Each client... Use its local complete dataset ,by Using the initial weights, perform local model training, minimizing the locally defined loss function. The local update process typically executes over multiple epochs, as summarized by the following formula: ; in, It is the local learning rate.

[0045] After local training is complete, the client does not upload the entire model. Instead, it calculates the difference between the local weights and the initially received global weights. Before uploading, the client encrypts the difference or adds differential privacy noise to provide privacy protection and prevent the federated learning server from inferring sensitive information from the update.

[0046] The federated learning server collects updates from all selected clients. Then, secure aggregation is performed using the FedAvg algorithm aggregation formula, as follows: ; in, Represents each client Local dataset The amount of data, This represents the total amount of data, i.e. (Assume there are K clients); (3) Model convergence and deployment phase, including: The federated learning server continuously monitors the performance of the global model on the reserved validation set (such as the loss function value or R² score). When the performance no longer improves significantly over multiple iterations, or when the preset maximum number of iterations is reached, the model is considered to have converged, and training terminates.

[0047] The final trained global model Deployed on the federated learning server and accessed as a data service in the trusted data space, it provides services as a prediction API. Participating entities can call it through the trusted data space connector. When it is necessary to use the global model for electricity price prediction, it can obtain the latest authorized data required in real time as input to the global model according to the data access and dynamic authorization mechanism to generate future electricity price predictions and assist entities in making pricing decisions.

[0048] like Figure 3 As shown, this embodiment designs a collaborative auxiliary decision-making framework for the electricity spot market based on a trusted data space, which includes four layers: a data resource layer, a data space layer, a collaborative computing layer, and an auxiliary decision-making application layer.

[0049] The data resource layer is mainly responsible for connecting with various market entities. The diagram shows power plant A, electricity sales company B, and power grid company C as examples. It can store their private data locally and access the trusted data space through the trusted data space connector.

[0050] The data space layer is built upon the national trusted data space technology architecture to construct a trusted data space service platform, providing core services such as identity authentication, smart contract management, metadata management, access connector management, and transaction clearing. All access to and use of data must comply with the data owner's preset permission policies.

[0051] The collaborative computing layer adopts federated learning as the core computing paradigm. A federated learning server coordinates multiple participants to train an improved Transformer model locally using their private data. Only encrypted model parameters (such as gradients) are uploaded for aggregation and global model updates, ensuring that the original data never leaves the domain.

[0052] The decision support application layer deploys the trained global model as a SaaS service, providing authorized users with high-precision multi-step time-series electricity price forecasts, risk warnings, and other services.

[0053] like Figure 4As shown, another preferred embodiment of this application also provides a collaborative auxiliary decision-making device for the electricity spot market based on a trusted data space, comprising: The data docking and storage module is used to dock with various market participants in the electricity spot market, enabling them to store their private data locally and connect to the trusted data space service platform of the trusted data space through the trusted data space connector; The data security supervision module is used to build a trusted data space service platform based on the national trusted data space technology architecture. It provides core services including identity authentication, smart contract management, metadata management, access connector management, and transaction clearing, and realizes data access and dynamic authorization, and trusted auditing based on blockchain, so that the access and use of all data meet the data owner's preset permission policy. The global model federated learning module is used in a trusted data space environment to perform privacy-preserving federated learning collaborative training of the initial global model issued by the federated learning server to the selected market entity's client. After the initial global model converges, only the encrypted global model parameters are uploaded for aggregation to obtain an updated global model, which is then deployed to the federated learning server. The initial global model adopts an improved Transformer model, which uses non-autoregressive decoding and power time-series-aware location encoding. The global model deployment module is used to deploy the updated global model as a SaaS service, providing authorized users with multi-step time-series electricity price forecasting and risk warning services to assist entities in making pricing decisions.

[0054] The power spot market collaborative auxiliary decision-making device based on trusted data space provided in this embodiment adopts the power spot market collaborative auxiliary decision-making method based on trusted data space in the above embodiments, which solves the technical problems of limited performance, insufficient global dependence, and lack of privacy and security protection in the prior art. Compared with the prior art, the beneficial effects of the power spot market collaborative auxiliary decision-making device based on trusted data space provided in this embodiment are the same as the beneficial effects of the power spot market collaborative auxiliary decision-making method based on trusted data space provided in the above embodiments. Moreover, the other technical features in the power spot market collaborative auxiliary decision-making device based on trusted data space are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0055] like Figure 5 As shown, a preferred embodiment of this example also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space in the above embodiment.

[0056] This embodiment provides an electronic device that employs the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space as described in the above embodiments. This addresses the technical problems of limited performance, insufficient global dependence, and lack of privacy and security guarantees in existing technologies. Compared with existing technologies, the beneficial effects of the electronic device provided in this embodiment are the same as those of the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space provided in the above embodiments. Furthermore, other technical features of the electronic device are the same as those disclosed in the methods of the above embodiments, and will not be elaborated upon here.

[0057] like Figure 6 As shown in the preferred embodiment, this embodiment also provides a computer device, which may be a terminal or a liveness detection server, and its internal structure diagram may be as follows. Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface is used to communicate with other external computer devices via a network connection. When the computer program is executed by the processor, it implements the steps of the aforementioned collaborative decision-making support method for the electricity spot market based on trusted data space.

[0058] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the solution of this embodiment, and does not constitute a limitation on the computer device to which the solution of this embodiment is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0059] The computer equipment provided in this application adopts the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space in the above embodiments, which solves the technical problems of limited performance, insufficient global dependence, and lack of privacy and security protection in the prior art. Compared with the prior art, the beneficial effects of the computer equipment provided in this embodiment are the same as the beneficial effects of the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space provided in the above embodiments. Moreover, the other technical features in the electronic equipment are the same as the features disclosed in the method of the above embodiments, and will not be repeated here.

[0060] A preferred embodiment of this example also provides a storage medium, which includes a stored program that, when the program is executed, controls the device where the storage medium is located to perform the steps of the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space described in the above embodiment.

[0061] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0062] If the functions described in this embodiment are implemented as software functional units and sold or used as independent products, they can be stored in one or more computing device-readable storage media. Based on this understanding, the parts of this embodiment that contribute to the prior art or the technical solution can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computing device (which may be a personal computer, server, mobile computing device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this embodiment. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0063] Those skilled in the art will understand that the embodiments of this example can be provided as methods, systems, or computer program products. Therefore, this example can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this example can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in this example can be implemented using various computer languages, such as the object-oriented programming language C++ and the embedded programming language C.

[0064] This embodiment is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this embodiment. It should 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, generate instructions for implementing the flowchart illustrations. 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 stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] 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.

[0067] This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described collaborative auxiliary decision-making method for the electricity spot market based on a trusted data space.

[0068] The computer program product provided in this embodiment solves the technical problems of high testing and experimentation costs, delays, limited applicability, and slowdowns in scientific research in existing technologies. Compared with the prior art, the beneficial effects of the computer program product provided in this embodiment are the same as those of the collaborative auxiliary decision-making method for the electricity spot market based on trusted data space provided in the above embodiments, and will not be repeated here.

[0069] In summary, the key innovations of this application include: An improved Transformer model is designed by combining non-autoregressive decoding with power time-series-aware location encoding to enhance prediction efficiency and accuracy.

[0070] An innovative "Federal Electricity Price Forecasting Framework Based on Trusted Data Space" is designed. Under the premise that the data does not leave the domain, a global Transformer forecasting model is collaboratively trained in the trusted data space by exchanging model parameters or encrypting intermediate features.

[0071] This innovative design utilizes a dynamic, authorized high-value feature usage mechanism based on a trusted data space. By authorizing the Transformer model, it enables temporary "queries" of precise power prediction data for a specific renewable energy power plant or the flexible load adjustment potential of an industrial user within the trusted data space for a specific time period and for a specific prediction task. Furthermore, it introduces a blockchain-based trusted traceability mechanism, recording information such as the data source, usage permissions, model version, and prediction results used for each prediction task on the blockchain via smart contracts. This creates an immutable audit trail, resolving the "black box" trust issue of AI models.

[0072] Obviously, those skilled in the art can make various modifications and variations to this embodiment without departing from the spirit and scope of this embodiment. Therefore, if these modifications and variations of this embodiment fall within the scope of the claims of this embodiment and their equivalents, this embodiment is also intended to include these modifications and variations.

Claims

1. A collaborative auxiliary decision-making method for the electricity spot market based on a trusted data space, characterized in that, Including the following steps: S1. Connect with various market participants in the electricity spot market, enabling them to store their private data locally and access the Trusted Data Space service platform through the Trusted Data Space Connector; S2. Based on the national trusted data space technology architecture, build a trusted data space service platform to provide core services including identity authentication, smart contract management, metadata management, access connector management, and transaction clearing, realize data access and dynamic authorization, and trusted auditing based on blockchain, so that the access and use of all data meet the data owner's preset permission policy. S3. In a trusted data space environment, the initial global model is distributed by the federated learning server to the selected market entity's client for privacy-preserving federated learning collaborative training until convergence. Then, only the encrypted global model parameters are uploaded for aggregation to obtain the updated global model, which is then deployed to the federated learning server. The initial global model adopts an improved Transformer model, which uses non-autoregressive decoding and power time-series sensing location encoding. S4. Deploy the updated global model as a SaaS service to provide authorized users with multi-step time-series electricity price forecasting and risk warning services, assisting entities in making pricing decisions.

2. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 1, characterized in that, Step S2, specifically includes the following steps: S201. Metadata Publishing and Registration: Data providers publish their shareable data resources in the form of metadata to the metadata management module of the Trusted Data Space Service Platform through their local Trusted Data Space Connector. The metadata only describes the basic information of the data and does not contain any specific sensitive data values. S202. Usage Policy Definition: When publishing metadata, provide: attach a machine-readable usage policy to the data, which defines the authorized usage conditions of the data in the form of rules; S203, Dynamic Authorization and Contract Agreement: When a data user discovers the required data through the directory service, it initiates a data access request to the provider's trusted data space connector. The provider's trusted data space connector verifies the user's digital identity certificate and evaluates whether its request complies with the preset usage policy. If the verification is successful, the trusted data space connectors of both parties will automatically negotiate and generate a legally binding smart contract, recording the details of this authorization. After the smart contract is uploaded to the blockchain, a connection is established between the connectors.

3. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 1, characterized in that, In step S2, the blockchain-based trusted audit specifically includes the following steps: S211. Key events are recorded on the blockchain, including defining key behaviors, identifying key behaviors during system operation, converting them into hash values, and recording them on the blockchain. S212. Construct a trust chain, including linking key on-chain behaviors in chronological order into a complete and tamper-proof trust chain through the inherent logic of the blockchain.

4. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 3, characterized in that, In step S211, the key actions include data registration events, contract completion events, and model usage time. The data registration events include publishing metadata and usage policies; the contract completion events include granting data usage authorization and recording the hash value of the smart contract; and the model usage time includes recording the data credentials, model version number, and hash value of the prediction result called for this prediction when the finally trained model is used for a prediction task.

5. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 3, characterized in that, In step S212, after the trust chain is established, any participant is allowed to independently verify the following: whether a prediction result comes from a compliantly trained model, whether the model legally used authorized data during training, and whether the entire process complies with the usage policies set by each data provider.

6. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 1, characterized in that, In step S3, the improved Transformer model includes improvements to the model structure and output layer, and optimization of the decoder. The improvements to the model structure and output layer include introducing a dedicated fully connected layer as a feature focusing layer after the high-dimensional sequence representation output by the decoder. This layer maps high-dimensional features rich in temporal dependency information to a low-dimensional feature space focused on electricity price prediction. A linear projection layer is used as the final linear output layer. The activation function LeakyReLU is restricted to the period after the feature focusing layer and before the linear output layer. The optimization of the decoder includes using a learnable query matrix for input initialization to form a non-autoregressive decoder input mechanism, employing fully parallel forward computation during computation, and mapping the entire prediction sequence at once through a shared linear projection layer during output.

7. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 1, characterized in that, The improvements to the Transformer model include input representation and preprocessing of multi-source heterogeneous data, adaptation of location encoding and power time series, and improvements to the model structure and output layer. The input representation and preprocessing of the multi-source heterogeneous data includes numerical continuous feature preprocessing, non-numerical state feature preprocessing, and sequence construction and alignment. Numerical continuous features, including historical electricity prices, load, wind speed, and temperature, undergo local normalization. Non-numerical state features, including unit start-up / shutdown status, date type, and time period labels, cannot be directly input into the model; therefore, an "embedding layer" is used to map them into dense vector representations. Sequence construction and alignment concatenate all processed features, including normalized numerical features and embedded categorical features, according to time points to form a unified feature vector, and construct the input sequence according to the set historical window length. Each of them x t They all contain all available multi-dimensional information at that moment; The adaptation of the location coding to the power timing sequence retains the original sine and cosine coding for the basic location coding to capture the relative location information within the sequence; for the time-aware location coding, explicit time features are embedded and used as a supplement to or partial replacement of the location coding. The improvements to the model structure and output layer include: introducing a dedicated fully connected layer as a feature focusing layer after the high-dimensional sequence representation output by the decoder, mapping the high-dimensional features rich in time-dependent information to a low-dimensional feature space focused on electricity price prediction, effectively filtering redundant information; using a linear projection layer as the final linear output layer to directly map the compressed feature vector to the continuous electricity price prediction value at a future set time point; limiting the activation function LeakyReLU to after the feature focusing layer and before the linear output layer; optimizing the decoder input by using a learnable query matrix for input initialization to form a non-autoregressive decoder input mechanism, employing fully parallel forward computation during computation, and mapping the decoder output to the entire prediction sequence at once through a shared linear projection layer.

8. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 7, characterized in that, During local normalization, each client calculates the mean and standard deviation of its own features locally and performs Z-score normalization. ； in, and These are the mean and standard deviation of the client's local training data, respectively. The formula for adapting the location code to the power timing sequence is as follows: ； in, Represents the absolute position index in the sequence; This represents the enhanced position encoding vector; Represents a linear transformation; This represents a vector concatenation operation; The standard sine and cosine position encoding vector; Represents an embedding operation. Representing the The embedding vector corresponding to the hour; Representing the The embedding vector corresponding to the sky.

9. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 7, characterized in that, Input initialization is performed using a learnable query matrix: ； in, n steps Indicates the length of the predicted sequence; Indicates the dimension of the hidden layer of the model; This represents a trainable parameter matrix, which is randomly initialized at the start of training and optimized along with other parameters of the model during training. The fully parallel forward computation of the decoder is as follows: ； in, Decoder (•) represents the decoding module, which includes multi-head self-attention, encoder-decoder attention, and feedforward network; , represents the hidden state matrix output by the decoder, where each row contains prediction information for the corresponding future time step i; When the decoder outputs, the hidden state matrix is... The entire predicted sequence is mapped in one step using a shared linear projection layer, as shown in the formula: ； in, This represents the weight matrix of the output layer; This represents the entire prediction sequence generated at once.

10. The collaborative auxiliary decision-making method for the electricity spot market based on trusted data space according to claim 1, characterized in that, In step S3, the step of performing privacy-preserving federated learning collaborative training of the initial global model, which is distributed by the federated learning server to the selected market entity's client in a trusted data space environment, until convergence, and then only uploading the encrypted global model parameters for aggregation to obtain the updated global model and deploying it to the federated learning server, specifically includes the following steps: S31. Task Initialization and Model Distribution: The federated learning server defines the training task and initializes an improved Transformer model as the initial global model. At the same time, it distributes the structure and initial weights of the initial global model to all clients of the market entities authorized to participate in this training task through broadcasting. S32. Local Model Training: After receiving the initial global model, each client trains it on its local private data. The training data includes not only its own data, but also relevant authorized data obtained in real time and on demand from the Trusted Data Space Service Platform. After training is completed, each market entity's client will receive a locally updated global model. S33. Model Parameter Upload and Aggregation: Each market entity's client encrypts the parameter updates of the locally updated global model and sends them to the federated learning server. Without decrypting the updates from individual clients, the federated learning server uses a secure aggregation protocol to calculate the sum of all parameter update values. Then, it applies the aggregated parameter update values ​​to the initial global model on the federated learning server to generate the updated global model. S34. Global Model Evaluation and Distribution: The federated learning server evaluates the performance of the updated global model through the validation set. When the performance no longer improves significantly in multiple consecutive iterations, or reaches the preset maximum number of iterations, it is determined that the updated global model has converged and training is terminated. Otherwise, the federated learning server distributes the updated global model to all participants and starts the next round of training. This process is repeated until the updated global model converges. S35. Deploy the final trained and updated global model to the federated learning server and access the trusted data space as a data service to form a prediction API for external service.

Images