A data trace-based power supervision report credibility checking method and system

Abstract

The present application relates to the technical field of power market data analysis, solves the problem that the prior art lacks a verification mechanism for the credibility of generated reports and the traceability of evidence, and is also prone to generating false content that contradicts the actual physical operating state, thereby affecting the credibility of the report, and particularly relates to a power regulation report credibility verification method based on data traceability, which collects and preprocesses multi-modal operating data of the power market to obtain baseline operating data, then performs multi-dimensional feature extraction thereon to generate a multi-modal basic feature vector, assigns a traceability identifier to each valid data entity in the multi-modal basic feature vector, and synchronizes the data entity and its corresponding data feature abstract to a distributed tamper-proof network to construct a traceability credential library. The present application improves the evidence traceability of the analyzed text, reduces the risk of tampering with the operating data during circulation, and improves the reliability of the analysis and periodic regulatory evaluation results of the market operation.

Description

Technical Field

[0001] This invention relates to the field of power market data analysis technology, and in particular to a method and system for verifying the credibility of power regulatory reports based on data traceability. Background Technology

[0002] Modern electricity market operation data exhibits significant characteristics such as exponential growth, multi-source heterogeneity, and high spatiotemporal coupling correlation. It encompasses structured market-clearing prices, nodal marginal prices, and generating unit output curves, as well as semi-structured or unstructured grid dispatch logs and monitoring system alarm texts. To meet increasingly complex regulatory needs, a standardized regulatory reporting and configuration center is being built. Through standardized data collection, cleaning, processing, and visualization, it provides comprehensive and systematic market operation analysis reports and supports the generation of customized thematic analysis reports and regular regulatory assessments.

[0003] Currently, most electricity market regulatory reports are generated using large models to produce fluent and structured texts. However, in practice, there is a general lack of verification mechanisms for the credibility and traceability of evidence in the generated reports. Furthermore, when processing large amounts of multimodal electricity market operation data, it is easy to generate false content that contradicts the actual physical operation status. False texts lacking underlying data evidence anchors reduce the credibility of electricity regulatory reports and affect the scientific and fair nature of electricity market decisions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for verifying the credibility of power regulatory reports based on data traceability. This solves the technical problems of existing technologies lacking a verification mechanism for the credibility of generated reports and the traceability of evidence, and being prone to generating false content that contradicts the actual physical operating state, thus affecting the credibility of the reports.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for verifying the credibility of power regulatory reports based on data traceability, comprising the following steps: S1. Collect and preprocess multimodal operation data of the power market to obtain baseline operation data, and then extract multidimensional features to generate multimodal basic feature vectors. S2. Assign a traceability identifier to each valid data entity in the multimodal basic feature vector, and synchronize it and its corresponding data feature summary to the distributed anti-tampering network to build a traceability credential library; S3. Construct a topological feature map based on multimodal basic feature vectors, and establish a routing mapping relationship between nodes in the topological feature map and corresponding traceability identifiers in the traceability certificate database; S4. Receive and parse the target instruction, and perform cross-modal semantic alignment and dimensional fusion of the topological feature map and the target instruction in a unified feature vector space to generate the target feature matrix; S5. Input the target feature matrix into the preset text generation model to generate initial analysis text containing operational status evaluation and quantitative indicators, and simultaneously extract feature correlation weights. S6. Based on the feature association weight and routing mapping relationship, reverse feature extraction is performed to extract the reference mapping set corresponding to the running status evaluation, and the consistency deviation parameter between the semantic vector of the target text sequence and the reference mapping set in the multi-dimensional feature space is calculated. S7. Determine whether the consistency deviation parameter is greater than the preset dynamic confidence threshold; If so, then local feature reconstruction of the initial analysis text is triggered based on the source fault nodes; If not, a state verification sequence is generated based on the reverse feature extraction path to obtain the verified power regulatory report text.

[0006] Furthermore, step S2 specifically includes the following steps: S21. Extract the spatial mapping vector from the multimodal basic feature vector, and combine it with the preset spatiotemporal benchmark to calculate and generate a source identifier containing temporal dependencies. The calculation formula is as follows: In the above formula, This serves as the source identifier for the i-th valid physical entity within time window t; This is a pre-defined one-way cryptographic hash function; It is an orthogonal projection matrix; Let be the multimodal basic feature vector of the i-th valid data entity within time window t; It is the XOR operator; This is the current time window reference stamp; This serves as the source identifier for the valid data entity in the previous time window t-1; S22. Based on the dynamic fluctuations of each valid data entity, calculate and generate a weighted data feature summary. The calculation formula is as follows: In the above formula, A summary of data features for the current period; This represents the total number of valid data entities within the current period. Let be the fluctuation gradient vector of the eigenvector of the i-th valid physical entity in the current period relative to the historical baseline; This is a preset scaling factor for volatility sensitivity; Let it be the first nonlinear activation function; S23. The traceability identifier and data feature summary are encapsulated in a structured manner and synchronously written into the distributed anti-tampering network through consensus nodes to build the underlying physical traceability credential library.

[0007] Furthermore, step S3 specifically includes the following steps: S31. Construct a constraint mask matrix based on the equipment connection relationships of the power system, and combine it with the multimodal basic feature vectors to calculate the dynamic edge connection weights of the generated topological feature graph. The calculation formula is as follows: In the above formula, The dynamic edge connection weight between node u and node v at time window t; The preset projection weight matrix; This is a concatenation of the multimodal basic feature vectors corresponding to node u and node v; This is the second nonlinear activation function; The symbol for the Hadama product; For the constraint mask matrix; S32. Based on the local neighborhood determined by the dynamic edge connection weights and constraint mask matrix, the node features are aggregated and calculated to generate the node topological feature vector. The calculation formula is as follows: In the above formula, This is the node topology feature vector after node u is aggregated; Let be the set of first-order neighbor nodes of node u determined based on the constraint mask matrix; These are the topological degrees of nodes u and v, respectively. The feature aggregation weight matrix; S33. Based on the node topology feature vector and the source identification of the corresponding data entity in the source identification certificate database, calculate the routing mapping tensor used for cross-modal addressing. The calculation formula is as follows: In the above formula, For route mapping tensors; For cross-modal routing projection matrix; It is a tensor product operator; A predefined low-dimensional embedding function for mapping discrete characters to a continuous vector space; This serves as the source identifier for the data entity corresponding to node u within time window t; S34. Use the routing mapping tensor as the addressing coordinates in the multidimensional feature space to establish the cross-modal precise positioning and routing mapping relationship between the logical feature nodes in the topological feature map and the corresponding traceability identifiers in the traceability voucher database.

[0008] Furthermore, step S4 specifically includes the following steps: S41. Extract the instruction character sequence corresponding to the target analysis text sequence, input it into a preset semantic encoding network for high-dimensional latent space mapping of sequence features, and generate instruction semantic feature vectors with the same dimensional scale as the topological feature map through a cross-modal space projection operator. The calculation formula is as follows: In the above formula, This is the semantic feature vector of the instruction; Here is the weight matrix of the spatial projection operator; This is the sequence feature mapping function for a semantic coding network; A sequence of instruction characters; For layer normalization function; S42. Concatenate the topological feature vectors of each node to construct a global topological feature matrix. Using the instruction semantic feature vector as the query condition, and combining it with the preset cross-modal isolation mask vector, perform scaling dot product attention calculation on the global topological feature matrix to generate a semantic topological vector. The calculation formula is as follows: In the above formula, It is a semantic topological vector; It is a global topological feature matrix composed of the topological feature vectors of each node; The feature projection matrix of the graph node; This is the numerical scaling factor; The symbol for the Hadama product; A cross-modal isolation mask vector; S43. Based on semantic topological vectors, the global topological feature matrix is ​​weighted and aggregated to generate a graph feature vector. Then, a cross-modal feature weight vector is calculated using a preset adaptive gating function. The calculation formula is as follows: In the above formula, For the spectral feature vector; This is the transpose of the global topological feature matrix; The transpose of the semantic topological vector; This is a cross-modal feature weight vector; This is a preset feature mapping parameter matrix; To concatenate and splice the instruction semantic feature vector with the graph feature vector; It is the third nonlinear activation function; S44. Based on the cross-modal feature fusion weights, the graph feature vector and the instruction semantic feature vector are combined along the feature dimension to generate the target feature matrix. The calculation formula is as follows: In the above formula, The target feature matrix is ​​fused.

[0009] Furthermore, step S5 specifically includes the following steps: S51. Input the target feature matrix into the latent space feature extraction layer of the preset text generation model, and generate the dynamic query feature vector required for the current sequence node through autoregressive nonlinear transformation. The calculation formula is as follows: In the above formula, This is the dynamic query feature vector corresponding to the j-th sequence node; The hidden layer state output of the text generation model at time step j-1; The query projection weight matrix for the latent space feature extraction layer; S52. Based on the scaling dot product calculation of the dynamic query feature vector and the target feature matrix, extract the intermodal correlation weights of the current sequence node to the multimodal physical features. The calculation formula is as follows: In the above formula, The intermodal association weights generated for the j-th sequence node; The key projection weight matrix of the latent space feature extraction layer; Scalarize the feature dimension; The target feature matrix; S53. Based on the preset semantic truncation boundary, the sequence nodes generated by continuous decoding are spliced ​​in time to generate an initial analysis text containing running status evaluation and quantitative indicators. S54. Extract multiple sequence nodes that constitute the operational status assessment. Perform weighted pooling on the corresponding intermodal association weights based on the semantic saliency of each sequence node. Then, perform redundant connection filtering with a preset truncation mask to generate feature association weights corresponding to the operational status assessment. The calculation formula is as follows: In the above formula, The feature association weights for the m-th operating state evaluation; This is the set of time steps for the sequence nodes in the operational status assessment. The semantic saliency weight of the j-th sequence node; Use a preset truncation mask for valid features.

[0010] Furthermore, the consistency deviation parameter includes distance deviation and absolute value deviation. Step S6 specifically includes the following steps: S61. Based on the feature association weight and the routing mapping tensor, perform inverse projection in the feature space to obtain the addressing response intensity vector, and based on the preset source tracing response threshold, extract the reference mapping set that maps to the running status evaluation from the topological feature map in reverse. S62. Perform weighted aggregation of the node topological feature vectors in the reference mapping set based on response intensity to obtain the reference vector, and introduce a preset diagonal weight matrix. Calculate the high-dimensional feature distance between the semantic vector of the corresponding label text sequence and the reference vector in the multi-dimensional fusion space by evaluating the running state, so as to generate the distance deviation. S63. Extract quantitative indicators related to the operational status assessment from the initial analysis text, calculate the absolute error between them and the physical values ​​mapped by the reference mapping set, and generate the absolute numerical deviation by combining the allowable measurement error tolerance of the power grid equipment. S64. Use the hyperbolic tangent function to fuse and transform the distance deviation and the absolute numerical deviation to generate the consistency deviation parameter.

[0011] A power regulatory report credibility verification system based on data traceability includes a processor and a memory. The memory stores a computer program, which is implemented by the processor as a power regulatory report credibility verification method.

[0012] By employing the above technical solution, the present invention provides a method and system for verifying the credibility of power regulatory reports based on data traceability, which has at least the following beneficial effects: 1. This invention constructs an underlying traceability credential repository by assigning traceability identifiers to valid data entities and synchronizing them with data feature summaries to a distributed anti-tampering network. After successful verification, a status verification sequence is generated based on the reverse feature extraction path. This approach generates a clear and tamper-proof verification link for the report configuration center, improves the traceability of evidence in the analysis text, reduces the risk of operational data being tampered with during circulation, and enhances the reliability of market operation analysis and periodic regulatory assessment results.

[0013] 2. This invention performs reverse feature extraction based on feature association weights and routing mapping relationships, and calculates the consistency deviation parameter between the semantic vector of the target text sequence and the reference mapping set in the multi-dimensional feature space. This achieves accurate verification from natural language evaluation to objective physical data, improves the credibility of power regulatory reports, reduces market misleading caused by false content that contradicts the actual physical operating status, and enhances the scientific nature and fairness of power market decision-making.

[0014] 3. This invention constructs a topological feature map and performs cross-modal semantic alignment and dimensional fusion between the map and the target instruction within a unified feature vector space. This reduces semantic deviations caused by simple splicing of heterogeneous modal data, reduces the limitations of isolated text features affecting the depth of global analysis, improves the comprehensive analysis capability of complex features, and ensures the output quality of customized power supervision reports.

[0015] 4. This invention constructs an adaptive feedback closed loop based on a dynamic confidence threshold. When the consistency deviation parameter is greater than the preset dynamic confidence threshold, the local feature reconstruction of the initial analysis text is triggered based on the source fault node. This method reduces unnecessary redundant calculations, reduces the generation delay of regulatory reports, and improves the report generation efficiency of the regulatory reporting configuration center. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the credibility verification method of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0018] Existing technologies for generating electricity market regulatory reports generally lack verification mechanisms for the credibility and traceability of evidence. Furthermore, when processing large amounts of multimodal electricity market operation data, they are prone to generating false content that contradicts the actual physical operating conditions, thus affecting the credibility of electricity regulatory reports. To improve the traceability of evidence in electricity market regulatory reports and reduce false content that contradicts the actual physical operating conditions, this invention proposes a data traceability-based method for verifying the credibility of electricity regulatory reports. Figure 1 As shown, the method includes the following steps: S1. Collect multimodal power operation data from the power market. This data includes structured SCADA system measurement data (such as voltage and active power) and transaction gateway clearing data, as well as unstructured dispatch log text. Failure to align heterogeneous data will result in severe feature misalignment during data fusion, affecting the accuracy of subsequent topological relationships in map construction and reducing system data processing efficiency. Based on a pre-defined spatiotemporal benchmark, Kalman filtering or moving average denoising algorithms are used for noise filtering, uniformly mapping scattered data into high-dimensional continuous feature vectors to obtain benchmark operation data. Then, multi-dimensional feature extraction is performed on the benchmark operation data to generate multimodal basic feature vectors with a unified spatiotemporal sequence, providing a standardized data structure for subsequent calculations.

[0019] S2. Traditional power data is easily tampered with when accessed, which can lead to the concealment and alteration of key data or transaction instructions, affecting the integrity of the evidence chain used as the basis for penalties in regulatory reports. To address this, a traceability identifier is assigned to each valid data entity in the multimodal basic feature vector. Valid data entities include valid physical entities that characterize the state of power grid topology nodes. These entities and their corresponding data feature summaries are synchronized to a distributed anti-tampering network to build an underlying physical traceability certificate library. This improves the traceability of evidence in the analysis text and reduces the risk of operational data being tampered with during the transfer process.

[0020] In this embodiment, as a preferred implementation method, the specific implementation steps are as follows: By extracting the spatial mapping vector from the multimodal basic feature vector and combining it with a preset spatiotemporal benchmark, a traceability identifier containing temporal dependencies is calculated and generated. Based on the dynamic fluctuations of each valid data entity, a weighted data feature summary is calculated and generated. The traceability identifier and the data feature summary are structurally encapsulated and synchronously written into a distributed anti-tampering network through consensus nodes to build an underlying physical traceability credential library.

[0021] The formulas for calculating the source identifier and the data feature summary are as follows: In the above formula, This serves as the source identifier for the i-th valid physical entity within time window t; This is a pre-defined one-way cryptographic hash function; This is the orthogonal projection matrix used for feature dimensionality reduction and identity decoupling; Let be the multimodal basic feature vector of the i-th valid data entity within time window t; It is the XOR operator; This is the current time window reference stamp; This serves as the source identifier for the valid data entity in the previous time window t-1; A summary of data features for the current period; This represents the total number of valid data entities within the current period. Let be the fluctuation gradient vector of the eigenvector of the i-th valid physical entity in the current period relative to the historical baseline; This is a preset scaling factor for volatility sensitivity; is the first nonlinear activation function, used to map abnormal fluctuation states to summary aggregation weights.

[0022] S3. To avoid the impact of existing technologies on the accuracy of subsequent state assessment and tracing when processing complex power data, a topological feature map representing the real-time operating state of the power system is constructed based on multimodal basic feature vectors. A routing mapping relationship between nodes in the topological feature map and corresponding tracing identifiers in the tracing voucher database is established. By establishing a cross-modal addressing coordinate system with guiding significance, it is beneficial for subsequent large models to perform accurate reverse feature extraction based on attention weights.

[0023] In this embodiment, as a preferred implementation method, the specific implementation steps are as follows: Based on the equipment connection relationships of the power system, a constraint mask matrix is ​​constructed. Combined with multimodal basic feature vectors, the dynamic edge connection weights of the generated topological feature graph are calculated using the following formula: In the above formula, The dynamic edge connection weight between node u and node v at time window t; The preset projection weight matrix; This is a concatenation of the multimodal basic feature vectors corresponding to node u and node v; This is the second nonlinear activation function; The symbol for the Hadama product; To constrain the mask matrix, if node u and node v are physically connected, its value is 1, otherwise it is 0; Based on the local neighborhood determined by the dynamic edge connection weights and constraint mask matrix, node features are aggregated and calculated to generate a node topological feature vector that incorporates the global topological state. The calculation formula is as follows: In the above formula, This is the node topology feature vector after node u is aggregated; Let be the set of first-order neighbor nodes of node u determined based on the constraint mask matrix; These are the topological degrees of nodes u and v, respectively, used for normalization during the aggregation process; The feature aggregation weight matrix; Based on the node topological feature vectors and the corresponding data entity's traceability identifier in the traceability certificate database, a routing mapping tensor for cross-modal addressing is calculated. This routing mapping tensor is used as the addressing coordinates in the multi-dimensional feature space to establish the precise cross-modal localization and routing mapping relationship between logical feature nodes in the topological feature map and their corresponding traceability identifiers in the traceability certificate database. The formula for calculating the routing mapping tensor is as follows: In the above formula, For route mapping tensors; For cross-modal routing projection matrix; It is a tensor product operator used for cross-fusion of spatial features; A predefined low-dimensional embedding function for mapping discrete characters to a continuous vector space; This is the source identifier for the data entity corresponding to node u within time window t.

[0024] S4. Since natural language commands and physical topology maps belong to completely different heterogeneous data modalities, direct vector concatenation will cause semantic shifts and feature space misalignments, affecting the accuracy of perceiving the real physical state and generating extremely high redundant floating-point operations, thereby reducing the system's processing efficiency. By receiving and parsing the trigger commands of the target analysis text sequence, i.e., the target commands, cross-modal semantic alignment and dimensional fusion of the topology feature map and the target commands are performed in a unified feature vector space to generate a target feature matrix. This allows the system to adaptively adjust the proportion of text intent in the target feature matrix based on the support strength of the physical topology map data, thereby reducing modal noise interference caused by excessive dominance of natural language and eliminating the negative impact of simple vector concatenation on the integrity of feature expression.

[0025] In this embodiment, as a preferred implementation method, the specific implementation steps are as follows: Extract the instruction character sequence corresponding to the target analysis text sequence, input it into a pre-defined semantic encoding network for high-dimensional latent space mapping of sequence features, and generate instruction semantic feature vectors at the same dimensional scale as the topological feature map through a cross-modal space projection operator. The calculation formula is as follows: In the above formula, This is the semantic feature vector of the instruction; This is the weight matrix for the spatial projection operator used for cross-modal dimension alignment; This is the sequence feature mapping function for a semantic coding network; A sequence of instruction characters; This is the layer normalization function.

[0026] The topological feature vectors of each node are concatenated to construct a global topological feature matrix. Using the instruction semantic feature vector as the query condition, and combined with a preset cross-modal isolation mask vector, a scaling dot product attention calculation is performed on the global topological feature matrix to generate a semantic topological vector representing the spatial mapping weights. The calculation formula is as follows: In the above formula, For semantic topology vectors, each dimension represents the feature alignment weight of the target instruction with different physical nodes in the graph; It is a global topological feature matrix composed of the topological feature vectors of each node; The feature projection matrix of the graph node; This is a numerical scaling factor used to scale the vector dot product to prevent numerical saturation. The symbol for the Hadama product; It is a cross-modal isolation mask vector used to force the alignment weights of nodes that do not match the physical properties represented by the target instruction to zero through a hard filtering mechanism.

[0027] The global topological feature matrix is ​​weighted and aggregated based on semantic topological vectors to generate a graph feature vector. A cross-modal feature weight vector is calculated using a pre-defined adaptive gating function. Then, the graph feature vector and the instruction semantic feature vector are combined along the feature dimension according to the cross-modal feature fusion weights to generate the target feature matrix. The calculation formula is as follows: In the above formula, The aggregated feature vectors of the graph; This is the transpose of the global topological feature matrix; The transpose of the semantic topological vector; This is a cross-modal feature weight vector; This is a preset feature mapping parameter matrix; To concatenate and splice the instruction semantic feature vector with the graph feature vector; As the third non-linear activation function, the Sigmoid function is selected to map the feature fusion weights to the interval between 0 and 1. The target feature matrix is ​​fused.

[0028] S5. Input the target feature matrix into the preset text generation model to generate initial analysis text containing operational status assessment and quantitative indicators, and simultaneously extract feature association weights representing cross-modal mapping relationships. This method not only automatically and forcibly outputs multi-dimensional quantitative traceability evidence supporting the assessment while generating initial analysis text containing operational status assessment and quantitative indicators, but also eliminates the risks caused by the uninterpretability of natural language generation.

[0029] In this embodiment, as a preferred implementation method, the specific implementation steps are as follows: The target feature matrix is ​​input into the latent space feature extraction layer of the preset text generation model. Through autoregressive nonlinear transformation, the dynamic query feature vector required by the current sequence node is generated. Based on the scaling dot product calculation of the dynamic query feature vector and the target feature matrix, the intermodal correlation weights of the current sequence node with the multimodal physical features are extracted.

[0030] Specifically, the formulas for calculating the dynamic query feature vector and the intermodal association weights are as follows: In the above formula, This is the dynamic query feature vector corresponding to the j-th sequence node; The hidden layer state output of the text generation model at time step j-1; The query projection weight matrix for the latent space feature extraction layer; The intermodal correlation weights generated for the j-th sequence node are used to characterize the focused distribution of this sequence node on different power grid characteristic nodes; The key projection weight matrix of the latent space feature extraction layer; Scale the feature dimension for smooth gradient updates to a scalar; The target feature matrix.

[0031] Based on a preset semantic truncation boundary, sequential nodes generated by continuous decoding are temporally concatenated to generate initial analysis text containing operational status assessment and quantitative indicators. Multiple sequence nodes constituting the operational status assessment are extracted. Based on the semantic saliency of each sequence node, the corresponding intermodal association weights are weighted and pooled. The pooled weights are then redundantly connected and filtered with a preset truncation mask to generate feature association weights corresponding to the operational status assessment. The calculation formula is as follows: In the above formula, The feature association weights for the m-th operating state evaluation; This is the set of time steps for the sequence nodes in the operational status assessment. The semantic saliency weight of the j-th sequence node; A preset effective feature truncation mask is used to force weight elements in the weight distribution that are below a preset importance threshold to zero, thereby reducing the memory usage and computing power consumption of subsequent reverse addressing.

[0032] S6. Traditional monitoring systems typically use simple text matching based on shallow rules to verify conclusions, which easily deviates from underlying objective physical laws and affects the accuracy of verification results. This system, however, uses reverse feature extraction based on feature association weights and routing mapping relationships to extract a reference mapping set corresponding to the operational status assessment. It then calculates the consistency deviation parameters between the semantic vector of the target text sequence and the reference mapping set in a multi-dimensional feature space, including distance and absolute numerical deviations. This reduces unnecessary redundant calculations, lowers the delay in generating analysis reports, and improves the overall computing efficiency of computer hardware.

[0033] In this embodiment, as a preferred implementation method, the specific implementation steps are as follows: Inverse projection is performed in the feature space based on feature association weights and routing mapping tensors to obtain the addressing response intensity vector. Then, a reference mapping set that maps to the running status assessment is extracted from the topological feature map based on a preset source tracing response threshold.

[0034] The node topological feature vectors in the reference mapping set are weighted and aggregated based on response intensity to obtain the reference vector. A pre-set diagonal weight matrix is ​​introduced, and the distance between the semantic vector of the corresponding label text sequence and the reference vector in the multi-dimensional fusion space is calculated by evaluating the running state to generate the distance deviation.

[0035] Based on the quantitative indicators extracted from the initial analysis text and associated with the operational status assessment, the absolute error is calculated by comparing them with the underlying real physical values ​​mapped by the reference mapping set, and the absolute numerical deviation is generated by combining the allowable measurement error tolerance of the power grid equipment.

[0036] Specifically, the formulas for calculating distance deviation and absolute numerical deviation are as follows: in, This refers to distance deviation; As a reference vector; The semantic vector of the target text sequence; This is a pre-defined diagonal weight matrix used for scaling the importance of feature dimensions; This represents the absolute numerical deviation. These are the extracted quantitative indicators; The underlying real physical values ​​obtained from the underlying physical traceability credential repository; To introduce the tolerance of measurement error to the engineering fault boundary.

[0037] Next, the distance deviation and absolute numerical deviation are fused and transformed using the hyperbolic tangent function to generate the consistency deviation parameter. The calculation formula is as follows: in, This is the consistency deviation parameter; This represents the absolute numerical deviation. This refers to distance deviation; This is the adaptive adjustment coefficient; This is the distance deviation scaling matrix; It is a hyperbolic tangent nonlinear transformation function.

[0038] S7. Compare the consistency deviation measurement parameter with the preset dynamic confidence threshold; if the consistency deviation measurement parameter is greater than the preset dynamic confidence threshold, generate a feedback adjustment signal based on the source fault node to trigger local feature reconstruction for the initial analysis text; if the consistency deviation measurement parameter is less than or equal to the preset dynamic confidence threshold, generate a state verification sequence containing the state mapping benchmark based on the reverse feature extraction path, and generate a verified power supervision report text.

[0039] The present invention also provides a power regulatory report credibility verification system based on data traceability, including a processor and a memory, wherein the memory is used to store a computer program, and the computer program is implemented by the processor to verify the credibility of power regulatory reports.

[0040] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application 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.

[0041] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Since the above embodiments are substantially similar to the method embodiments, their descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0042] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for verifying the credibility of power regulatory reports based on data traceability, characterized in that, The method includes the following steps: S1. Collect and preprocess multimodal power operation data to obtain baseline operation data, and extract multidimensional features to generate multimodal basic feature vectors. S2. Assign a traceability identifier to each valid data entity in the multimodal basic feature vector and synchronize it with the corresponding data feature summary to the distributed anti-tampering network to build a traceability credential library; S3. Construct a topological feature map based on multimodal basic feature vectors, and establish a routing mapping relationship between nodes in the topological feature map and corresponding traceability identifiers in the traceability certificate database; S4. Receive and parse the target instruction, and perform cross-modal semantic alignment and dimensional fusion of the topological feature map and the target instruction in the feature vector space to generate the target feature matrix; S5. Input the target feature matrix into the preset text generation model to generate initial analysis text containing operational status evaluation and quantitative indicators, and simultaneously extract feature correlation weights. S6. Based on the feature association weight and routing mapping relationship, reverse feature extraction is performed to extract the reference mapping set corresponding to the running status evaluation, and the consistency deviation parameter between the semantic vector of the target text sequence and the reference mapping set in the multi-dimensional feature space is calculated. S7. Determine whether the consistency deviation parameter is greater than the preset dynamic confidence threshold; If so, then the local feature reconstruction of the initial analysis text is triggered based on the source fault nodes; If not, a state verification sequence is generated based on the reverse feature extraction path to obtain the verified power regulatory report text.

2. The credibility verification method according to claim 1, characterized in that, Step S2 specifically includes the following steps: S21. Extract the spatial mapping vector from the multimodal basic feature vector, and combine it with the preset spatiotemporal benchmark to calculate and generate a source identifier containing temporal dependencies. The calculation formula is as follows: In the above formula, This serves as the source identifier for the i-th valid physical entity within time window t; This is a pre-defined one-way cryptographic hash function; It is an orthogonal projection matrix; Let be the multimodal basic feature vector of the i-th valid data entity within time window t; It is the XOR operator; This is the current time window reference stamp; This serves as the source identifier for the valid data entity in the previous time window t-1; S22. Based on the dynamic fluctuations of each valid data entity, calculate and generate a weighted data feature summary. The calculation formula is as follows: In the above formula, A summary of data features for the current period; This represents the total number of valid data entities within the current period. Let be the fluctuation gradient vector of the eigenvector of the i-th valid physical entity in the current period relative to the historical baseline; This is a preset scaling factor for volatility sensitivity; Let it be the first nonlinear activation function; S23. The traceability identifier and data feature summary are encapsulated in a structured manner and synchronously written into the distributed anti-tampering network through consensus nodes to build the underlying physical traceability credential library.

3. The credibility verification method according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31. Construct a constraint mask matrix based on the equipment connection relationships of the power system, and combine it with the multimodal basic feature vectors to calculate the dynamic edge connection weights of the generated topological feature graph. The calculation formula is as follows: In the above formula, The dynamic edge connection weight between node u and node v at time window t; The preset projection weight matrix; This is a concatenation of the multimodal basic feature vectors corresponding to node u and node v; This is the second nonlinear activation function; The symbol for the Hadama product; For the constraint mask matrix; S32. Based on the local neighborhood determined by the dynamic edge connection weights and constraint mask matrix, the node features are aggregated and calculated to generate the node topological feature vector. S33. Based on the node topology feature vector and the source identification of the corresponding data entity in the source identification certificate database, calculate the routing mapping tensor used for cross-modal addressing. The calculation formula is as follows: In the above formula, For route mapping tensors; For cross-modal routing projection matrix; It is a tensor product operator; A predefined low-dimensional embedding function for mapping discrete characters to a continuous vector space; This serves as the source identifier for the data entity corresponding to node u within time window t; S34. Use the routing mapping tensor as the addressing coordinates in the multidimensional feature space to establish the cross-modal precise positioning and routing mapping relationship between the logical feature nodes in the topological feature map and the corresponding traceability identifiers in the traceability voucher database.

4. The credibility verification method according to claim 3, characterized in that, The formula for calculating the node topological feature vector is as follows: In the above formula, This is the node topology feature vector after node u is aggregated; Let be the set of first-order neighbor nodes of node u determined based on the constraint mask matrix; These are the topological degrees of nodes u and v, respectively. This is the feature aggregation weight matrix.

5. The credibility verification method according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. Extract the instruction character sequence corresponding to the target analysis text sequence, input it into a preset semantic encoding network for high-dimensional latent space mapping of sequence features, and generate instruction semantic feature vectors with the same dimensional scale as the topological feature map through a cross-modal space projection operator. The calculation formula is as follows: In the above formula, This is the semantic feature vector of the instruction; Here is the weight matrix of the spatial projection operator; This is the sequence feature mapping function for a semantic coding network; A sequence of instruction characters; For layer normalization function; S42. Concatenate the topological feature vectors of each node to construct a global topological feature matrix. Using the instruction semantic feature vector as the query condition, and combined with the preset cross-modal isolation mask vector, perform scaling dot product attention calculation on the global topological feature matrix to generate semantic topological vectors. S43. Based on semantic topological vectors, the global topological feature matrix is ​​weighted and aggregated to generate a graph feature vector. Then, a cross-modal feature weight vector is calculated using a preset adaptive gating function. The calculation formula is as follows: In the above formula, For the spectral feature vector; This is the transpose of the global topological feature matrix; The transpose of the semantic topological vector; This is a cross-modal feature weight vector; This is a preset feature mapping parameter matrix; To concatenate and splice the instruction semantic feature vector with the graph feature vector; It is the third nonlinear activation function; S44. Based on the cross-modal feature fusion weights, the graph feature vector and the instruction semantic feature vector are combined along the feature dimension to generate the target feature matrix. The calculation formula is as follows: In the above formula, The target feature matrix is ​​fused.

6. The credibility verification method according to claim 5, characterized in that, The formula for calculating the semantic topological vector is: In the above formula, It is a semantic topological vector; It is a global topological feature matrix composed of the topological feature vectors of each node; The feature projection matrix of the graph node; This is the numerical scaling factor; The symbol for the Hadama product; This is a cross-modal isolation mask vector.

7. The credibility verification method according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51. Input the target feature matrix into the latent space feature extraction layer of the preset text generation model, and generate the dynamic query feature vector required for the current sequence node through autoregressive nonlinear transformation. The calculation formula is as follows: In the above formula, This is the dynamic query feature vector corresponding to the j-th sequence node; The hidden layer state output of the text generation model at time step j-1; The query projection weight matrix for the latent space feature extraction layer; S52. Based on the scaling dot product calculation of the dynamic query feature vector and the target feature matrix, extract the intermodal correlation weights of the current sequence node to the multimodal physical features. The calculation formula is as follows: In the above formula, The intermodal association weights generated for the j-th sequence node; The key projection weight matrix of the latent space feature extraction layer; Scalarize the feature dimension; The target feature matrix; S53. Based on the preset semantic truncation boundary, the sequence nodes generated by continuous decoding are spliced ​​in time to generate an initial analysis text containing running status evaluation and quantitative indicators. S54. Extract multiple sequence nodes that constitute the operation status assessment, perform weighted pooling on the corresponding intermodal association weights based on the semantic saliency of each sequence node, and perform redundant connection filtering on the pooled weights and the preset truncation mask to generate feature association weights corresponding to the operation status assessment.

8. The credibility verification method according to claim 7, characterized in that, The formula for calculating the feature association weight is as follows: In the above formula, The feature association weights for the m-th operating state evaluation; This is the set of time steps for the sequence nodes in the operational status assessment. The semantic saliency weight of the j-th sequence node; Use a preset truncation mask for valid features.

9. The credibility verification method according to claim 1, characterized in that, The consistency deviation parameters include distance deviation and absolute value deviation. Step S6 specifically includes the following steps: S61. Based on the feature association weight and the routing mapping tensor, perform inverse projection in the feature space to obtain the addressing response intensity vector, and based on the preset source tracing response threshold, extract the reference mapping set that maps to the running status evaluation from the topological feature map in reverse. S62. Perform weighted aggregation of the node topological feature vectors in the reference mapping set based on response intensity to obtain the reference vector, and introduce a preset diagonal weight matrix. Calculate the high-dimensional feature distance between the semantic vector of the corresponding label text sequence and the reference vector in the multi-dimensional fusion space by evaluating the running state, so as to generate the distance deviation. S63. Extract quantitative indicators related to the operational status assessment from the initial analysis text, calculate the absolute error between them and the physical values ​​mapped by the reference mapping set, and generate the absolute numerical deviation by combining the allowable measurement error tolerance of the power grid equipment. S64. Use the hyperbolic tangent function to fuse and transform the distance deviation and the absolute numerical deviation to generate the consistency deviation parameter.

10. A system for verifying the credibility of power regulatory reports as described in any one of claims 1-9, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, which, when executed by the processor, implements the power regulatory report credibility verification method as described in any one of claims 1-9.

Images