A heterogeneous data processing method and system for energy big data

Abstract

The application discloses a kind of heterogeneous data processing method and system of energy big data, the method includes: obtaining different types of energy data and inputting pre-trained directed association graph model, obtain cross-type dynamic contribution flow network node forest;Obtain cross-type linkage report and build basic directional positioning dynamic correction chain through report analysis positioning and cause-effect analysis;The cross-type dynamic contribution flow network node forest and basic directional positioning dynamic correction chain are input into pre-trained positioning correction path model, and real-time positioning error correction adjustment is carried out on cross-type linkage report, and updated and adjusted cross-type linkage report is obtained.The application realizes cross-type linkage analysis and report real-time error correction of energy data, improves the accuracy and efficiency of energy data processing.

Description

Technical Field

[0001] This invention belongs to the field of energy data processing technology, and relates to a heterogeneous data processing method and system for energy big data. Background Technology

[0002] Current energy data centers aggregate heterogeneous data from multiple sources, covering structured, semi-structured, and unstructured forms. Significant semantic gaps exist between various energy subsystems, leading to constraints such as data silos, lack of dynamic correlation, and accumulated decision-making delays in cross-type data collaboration.

[0003] Existing energy data processing methods suffer from bottlenecks such as loss of value from heterogeneous data, distortion of dynamic contribution assessment, and lack of correction mechanisms. The fragmentation of energy correlation patterns makes it difficult to trace the origins of wind and solar power curtailment events. Traditional factor analysis ignores the cross-type transmission of primary and secondary indicators, resulting in significant prediction errors in correlations between different types of energy data. This forces adjustments to reports to rely solely on historical experience rules, failing to dynamically correct multi-energy chain imbalances. Consequently, it struggles to meet the requirements for accurate and self-correcting energy forecasts, and fails to provide accurate and timely analytical results for energy decision-making. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a heterogeneous data processing method and system for energy big data, which enables cross-type linkage analysis of energy data and real-time error correction of reports, thereby improving the accuracy and efficiency of energy data processing.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of this invention proposes a heterogeneous data processing method for energy big data, comprising:

[0007] Different types of energy data are acquired and input into a pre-trained directed graph model to obtain a cross-type dynamic contribution flow network node forest;

[0008] Obtain cross-type linked reports and construct a basic directed positioning dynamic correction chain through report parsing, positioning, and causal analysis;

[0009] The cross-type dynamic contribution flow network node forest and the basic directed positioning dynamic correction chain are input into the pre-trained positioning correction path model to perform real-time positioning error correction adjustment on the cross-type linkage report, thereby obtaining the updated and adjusted cross-type linkage report.

[0010] Preferably, the training process of the directed graph model includes:

[0011] Historical information on various types of energy is obtained and key variables are extracted to obtain a time series set of energy parameters and a set of associated change attributes for each type of energy.

[0012] Multi-stage time-series factor analysis was performed on the energy parameter time series set to obtain the main indicator variable and the corresponding co-indicator variable sequence, the instantaneous main contribution of the main indicator variable to the target result, the instantaneous co-contribution of the co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables.

[0013] Repeat the process of obtaining the instantaneous main contribution and instantaneous collaborative contribution to obtain the change path of the instantaneous main contribution and the change path of the instantaneous collaborative contribution. Combine the covariate correlation coefficients between collaborative indicator variables to construct a dynamic contribution flow network node tree corresponding to a single type of energy data.

[0014] A causal correlation analysis was performed on the aforementioned set of related change attributes to obtain the joint contribution factors of the main indicator variables to the target results;

[0015] Based on the joint contribution impact factor and the dynamic contribution flow network node tree corresponding to all types of energy data, a cross-type dynamic contribution flow network node forest is constructed.

[0016] By analyzing the acquisition process of the cross-type dynamic contribution flow network node forest, a reverse synthesis training function is constructed and reverse inference training is performed to obtain a trained directed association graph model.

[0017] Preferably, the step of acquiring historical information on various types of energy and extracting key variables to obtain a time series set of energy parameters and a set of associated change attributes for each type of energy includes:

[0018] The system acquires historical demand change information, cost change information, and related change information for each type of energy, extracts key variables, and obtains the time series set of energy parameters and the set of related change attributes for each type of energy.

[0019] Preferably, the step of performing multi-stage time-series factor analysis on the energy parameter time series set to obtain the main indicator variable and its corresponding co-indicator variable sequence, the instantaneous main contribution of the main indicator variable to the target result, the instantaneous co-contribution of the co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables includes:

[0020] Based on the time series set of energy parameters for each type of energy, the parameter loading matrix corresponding to continuous time points of each type of energy parameter is obtained through a multi-stage time series factor analysis model.

[0021] Statistical analysis is performed on the parameter loading matrix corresponding to each time point to obtain the sequence of each type of main indicator variable and its corresponding co-indicator variable, the instantaneous main contribution of each main indicator variable to the target result, the instantaneous co-contribution of each co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables.

[0022] Preferably, the process of repeatedly obtaining the instantaneous primary contribution and instantaneous collaborative contribution to obtain the change path of the instantaneous primary contribution and the change path of the instantaneous collaborative contribution, combined with the covariate correlation coefficients between collaborative indicator variables, constructs a dynamic contribution flow network node tree corresponding to a single type of energy data, including:

[0023] Repeat the process of obtaining instantaneous main contribution and instantaneous collaborative contribution to obtain the change path of instantaneous main contribution of each main indicator variable to the target result and the change path of instantaneous collaborative contribution of each collaborative indicator variable to the corresponding main indicator variable at continuous time points.

[0024] The target result is used as the target node of the first layer, the main indicator variable is used as the main contribution node of the second layer, and the sequence of collaborative indicator variables corresponding to each main indicator variable is used as the set of collaborative contribution nodes of the third layer.

[0025] Based on the main contributing node and the instantaneous main contribution change path to the target node corresponding to each type of energy data, the instantaneous collaborative contribution change path of each collaborative contributing node to the corresponding main indicator variable, and the covariate correlation coefficients between the same collaborative contributing node set, a dynamic contribution flow network node tree corresponding to a single type of energy data at continuous time points is constructed by combining a graph neural network with a graph database.

[0026] Preferably, the step of performing causal correlation analysis on the set of related change attributes to obtain the joint contribution factors of the main indicator variables to the target result includes:

[0027] Based on the set of associated change attributes of all types of energy parameters at each time point, the causal association analysis algorithm is used to obtain the main joint contribution of all main indicator variables corresponding to each target result to the same target result within a single time period, the association influence relationship and directional association influence direction among all main indicator variables corresponding to each target result, and the directional causal association coefficient between the sequences of synergistic indicator variables corresponding to different main indicator variables.

[0028] The joint contribution factor of each principal indicator variable to the same target result at a single time point is obtained by dividing the instantaneous principal contribution of a single principal indicator variable to the target result by the principal joint contribution of all principal indicator variables corresponding to each target result to the same target result within a single time period.

[0029] Preferably, the step of constructing a cross-type dynamic contribution flow network node forest based on the joint contribution impact factor and the dynamic contribution flow network node tree corresponding to all types of energy data includes:

[0030] Based on the joint contribution influence factor of each main indicator variable to the same target result at a single time point, the main contribution relationship connection between each main contribution node and the corresponding target node is constructed.

[0031] Based on the correlation and influence relationship and the direction of the directed correlation among all main indicator variables, a directed correlation connection between the main contributing nodes is constructed through a blockchain algorithm. At the same time, a cross-tree covariate correlation connection is constructed based on the directed causal correlation coefficient between the sequences of co-indicator variables corresponding to different main indicator variables.

[0032] The dynamic contribution flow network node trees corresponding to all types of energy data are timestamped at continuous time points with the constructed main contribution relationship connection, directed association connection and cross-tree covariate association connection. After alignment, the cross-type dynamic contribution flow network node forest is constructed by combining the topology space algorithm.

[0033] Preferably, the step of constructing a reverse synthesis training function and performing reverse inference training by analyzing the acquisition process of the cross-type dynamic contribution flow network node forest to obtain the trained directed association graph model includes:

[0034] Based on the instantaneous contribution mean square error, instantaneous main contribution change path loss, main index correlation cross-entropy loss, cross-tree covariate causal relationship loss, continuous timestamp alignment loss, complexity penalty for the construction of the number of main contributing nodes and corresponding collaborative contributing nodes, and constraint penalty function for the corresponding correlation strength in the cross-type dynamic contribution flow network node forest being less than the preset correlation coefficient, a reverse comprehensive training function is constructed.

[0035] Based on the inverse synthesis training function combined with the cross-type dynamic contribution flow network node forest, and through random forest combined with simulation algorithm, the trained directed association graph model is obtained by inverse reasoning training from the third layer collaborative contribution node set to the first layer target node.

[0036] Preferably, the training process of the localization correction path model includes:

[0037] By acquiring historical cross-type linkage report information and locating the key target results in the table through report parsing, we can obtain the sequence of key target results in the table and the relationship of synchronous changes in key target results, the position of key target results in the table and their contribution factors. Through causal analysis, we can obtain key target results with related changes and the corresponding causal relationship strength.

[0038] Based on the position of key target results in the table, key target results with related changes and the corresponding causal correlation strength, a training optimization directed localization dynamic correction chain is constructed, and it is mapped and similarly analyzed with the cross-type dynamic contribution flow network node forest to obtain the similarity consistency error.

[0039] By analyzing similarity consistency error, training and optimizing the corresponding parameter relationships between the directed localization dynamic correction chain and the cross-type dynamic contribution flow network node forest, a positive comprehensive loss function is constructed and forward training is performed to obtain the trained localization correction path model. Based on the loss results of the forward training, the directed association graph model is trained online in real time.

[0040] Preferably, the step of acquiring historical cross-type linkage report information and locating it through report parsing to obtain the sequence of key target results in the table and the relationship of synchronous changes in key target results, the position of key target results in the table and their contributing factors, and obtaining key target results with correlated changes and the corresponding causal correlation strength through causal analysis includes:

[0041] Obtain historical cross-type linkage report information, and through the preset report parsing and positioning model, obtain the sequence of key target results in the cross-type linkage report, the relationship between the synchronous change of the remaining key target results when each key target result changes, the position of each key target result in the cross-type linkage report, and the contribution impact factor corresponding to each key target result.

[0042] Based on the sequence of key target results and the relationship between the synchronous changes of the remaining key target results when each key target result changes, as well as the contribution and influence factors corresponding to each key target result, a causal analysis algorithm is used to obtain the key target results with correlated changes and the corresponding causal correlation strength.

[0043] Preferably, the step of constructing a directed localization dynamic correction chain based on the position of key target results in the table, key target results with correlational changes, and the corresponding causal correlation strength, and mapping and performing similarity analysis with the cross-type dynamic contribution flow network node forest to obtain similarity consistency error includes:

[0044] Based on the key target results with related changes and the corresponding causal relationship strength and the position of each key target result in the report, a training and optimization directed positioning dynamic correction chain is constructed by combining blockchain algorithms.

[0045] Each key objective result in the training optimization directed localization dynamic correction chain is mapped one-to-one with the target nodes in the cross-type dynamic contribution flow network node forest at the same time. At the same time, based on the contribution impact factor corresponding to each key objective result and the set of main contributing nodes and co-contributing nodes in the cross-type dynamic contribution flow network node forest at the same time point, a similarity algorithm is used to obtain the similarity coefficient between the contribution impact factor and the main contributing nodes and co-contributing nodes, and the similarity coefficient between the contribution impact factor and the co-contributing nodes.

[0046] Based on the obtained similarity coefficients, establish a similarity mapping between the contribution impact factor and the main contributing node and the co-contributing node, and construct a similarity consistency error.

[0047] Preferably, the step of constructing a positive comprehensive loss function by analyzing similarity consistency error, training and optimizing the corresponding parameter relationships between the directed localization dynamic correction chain and the cross-type dynamic contribution flow network node forest, and performing forward training to obtain the trained localization correction path model, and then training the directed association graph model online in real time based on the loss results of the forward training, includes:

[0048] Combining similarity consistency error, the error between the location key target result of the training and optimization directed positioning dynamic correction chain and the reverse target result of the same target node obtained by the cross-type dynamic contribution flow network node forest back reasoning, the error between the target result after the synchronous change of the remaining key target result corresponding to each key target result in the training and optimization directed positioning dynamic correction chain and the error between the reverse target result under the same target node as the remaining key target result obtained by the cross-type dynamic contribution flow network node forest back reasoning, and the positive comprehensive loss function constructed by the synchronous change delay of the remaining key target result, positive training is performed from the first layer corresponding to the target node to the third layer corresponding to the collaborative contribution node set to obtain the trained positioning correction path model;

[0049] The positive integrated loss result obtained at each step of the localization correction path model training is fed back to the directed relation graph model for synchronous reverse adjustment training, thereby obtaining a directed relation graph model trained in real time and synchronously online.

[0050] A second aspect of this invention proposes a heterogeneous data processing system for energy big data, comprising:

[0051] The directed association module is used to acquire different types of energy data and input them into a pre-trained directed association graph model to obtain a cross-type dynamic contribution flow network node forest.

[0052] The causal analysis module is used to obtain cross-type linked reports and construct a basic directed positioning dynamic correction chain through report parsing, positioning, and causal analysis.

[0053] The positioning correction module is used to input the cross-type dynamic contribution flow network node forest and the basic directed positioning dynamic correction chain into the pre-trained positioning correction path model, and to perform real-time positioning error correction adjustment on the cross-type linkage report to obtain the updated and adjusted cross-type linkage report.

[0054] A third aspect of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0056] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0057] This invention forms a cross-type dynamic contribution flow network node forest through a pre-trained directed correlation graph model and a localization correction path model, realizing deep fusion and dynamic correlation analysis of multi-source heterogeneous energy data. It can eliminate data silos, establish a reliable correlation mapping of cross-type energy data, accurately capture the nonlinear time-varying law of multi-energy coupling, and realize cross-type transmission analysis of the dynamic contribution of the main-co-indicators.

[0058] This invention enhances the learning ability of the directed correlation graph model to the correlation patterns of energy data by integrating training functions and back reasoning training. Combined with the positioning correction path model, it realizes real-time error correction of cross-type linkage reports and dynamically corrects the chain imbalance of multiple energy sources.

[0059] This invention utilizes an online synchronous training mechanism to adapt to dynamic changes in the energy system, providing accurate analysis results for energy forecasting with interconnected self-correcting capabilities, thereby improving the accuracy and timeliness of energy decision-making. Attached Figure Description

[0060] Figure 1 This is a flowchart of a heterogeneous data processing method for energy big data according to the present invention.

[0061] Figure 2 This is a flowchart illustrating the implementation of a heterogeneous data processing method for energy big data according to the present invention.

[0062] Figure 3 This is the architecture diagram for constructing the directed association graph model of the present invention;

[0063] Figure 4 This is a diagram illustrating the architecture of the location correction path model for this invention.

[0064] Figure 5 This is a schematic diagram of a heterogeneous data processing system for energy big data according to the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0066] Embodiment 1 of this invention provides a heterogeneous data processing method for energy big data. It acquires different types of energy data and constructs a directed association graph model using an association algorithm and graph database, thereby obtaining a cross-type dynamic contribution flow network node forest. Simultaneously, it acquires cross-type linkage reports, extracting key target results with correlational changes and causal correlation strengths. A directed positioning dynamic correction chain is constructed using blockchain algorithms and location information. Finally, a positioning correction path model is built based on the cross-type dynamic contribution flow network node forest, the directed positioning dynamic correction chain, and a text positioning algorithm. This enables real-time positioning error correction and adjustment of the cross-type linkage reports, resulting in updated cross-type linkage reports. These updated reports are used for real-time positioning correction of multi-source heterogeneous data in various energy fields such as coal, oil, gas, electricity, and heat. Figures 1-2 As shown, it includes:

[0067] S1: Obtain different types of energy data and input them into a pre-trained directed graph model to obtain a cross-type dynamic contribution flow network node forest;

[0068] More preferably, different types of energy data are acquired, and a directed association graph model constructed by association algorithms and graph databases is combined to obtain a cross-type dynamic contribution flow network node forest;

[0069] The directed relational graph model is constructed and trained by combining similar energy data with a relational algorithm and a graph database, such as... Figure 3 As shown, the process includes:

[0070] (1) Obtain information on various types of energy and extract key variables to obtain the time series set of energy parameters and the set of associated change attributes for each type of energy:

[0071] It acquires historical demand change information, cost change information, and related change information of various types of energy, and extracts key variables through a preset variable analysis model to obtain the time series set of parameters and related change attribute set of each type of energy.

[0072] It should be further explained that the specific process for obtaining the time series set of each type of energy parameter and the associated change attribute set in this embodiment includes:

[0073] First, clean and preprocess heterogeneous data from multiple sources such as coal, oil, gas, electricity, and heat. For missing values, use time series interpolation or rule filling based on domain knowledge. For outliers, use isolated forest or statistical thresholding methods to detect and correct them.

[0074] Second, wavelet transform is used to perform multi-resolution analysis on time series data to extract features at different time scales;

[0075] Third, mutual information and random forest feature importance scores are used to screen variables that are strongly correlated with energy demand and cost;

[0076] Fourth, the Granger causality test is used to identify the direction and strength of causal relationships between variables. For the set of associated change attributes, the dynamic conditional covariate correlation coefficient is used to capture the time-varying correlation between parameters. Combined with the Copula function, the nonlinear dependency structure is characterized, and finally, a time series set of energy parameters and a set of associated change attributes containing multi-dimensional attributes such as energy price fluctuations and supply and demand relationships are formed.

[0077] (2) Perform multi-stage time-series factor analysis on the energy parameter time series set to obtain the main indicator variable and the corresponding co-indicator variable sequence, the instantaneous main contribution of the main indicator variable to the target result, the instantaneous co-contribution of the co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables:

[0078] Based on the time series set of each type of energy parameter, the parameter load matrix corresponding to the continuous time points of each type of energy parameter is obtained through a multi-stage time series factor analysis model.

[0079] The implementation process of the multi-stage time-series factor analysis model in this embodiment includes: performing time continuity checks and outlier processing on the time series sets of energy parameters for each type of energy to obtain standardized energy parameter time series sequences; dividing the standardized energy parameter time series sequences into continuous time windows according to preset time intervals, performing Kaiser criterion checks and factor rotation processing on the energy parameter time series sequences within each time window to determine the number of common factors within the corresponding time window and the correlation between each common factor and the energy parameter; based on the correlation between the common factors and the energy parameter within each time window, calculating the loading coefficients of the energy parameter on the common factors within each time window using the maximum likelihood estimation method, arranging the loading coefficients within each time window according to the correspondence between the energy parameter and the common factor to obtain the initial parameter loading matrix corresponding to each time window; performing smoothness correction on the initial parameter loading matrices of adjacent time windows to eliminate abrupt changes in loading coefficients caused by time window division, and finally obtaining the parameter loading matrix corresponding to each type of energy parameter at continuous time points.

[0080] Statistical analysis is performed on the parameter loading matrix corresponding to each time point to obtain the sequence of each type of main indicator variable and its corresponding co-indicator variable, the instantaneous main contribution of each main indicator variable to the target result, the instantaneous co-contribution of each co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables.

[0081] It should be further explained that the specific process of obtaining each type of main indicator variable and its corresponding co-indicator variable sequence, the instantaneous main contribution of each main indicator variable to the target result, the instantaneous co-contribution of each co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficients between the co-indicator variables in this embodiment includes:

[0082] First, based on the original multi-source heterogeneous data, the parameter loading matrix is ​​factor-rotated by the maximum likelihood estimation method. The number of principal components is determined by combining the eigenvalue greater than 1 criterion and scree plot test. Variables with absolute values ​​of loading coefficients greater than 0.8 are selected as principal index variables, and the remaining variables constitute the sequence of co-index variables.

[0083] Second, based on the main indicator variables and target result data, the standardized regression coefficients are calculated by partial least squares regression, and the significance of the coefficients is verified by the Bootstrap resampling method to obtain the instantaneous main contribution of the main indicator variables to the target result (i.e., the standardized regression coefficients).

[0084] Third, based on the data of the co-indicator variables and the main indicator variables, the path coefficients are calculated by constructing a structural equation model. Combined with multiple sets of confirmatory factor analysis, it is ensured that both RMSEA (root mean square of approximation error) and CFI (comparison fit index) meet the corresponding set thresholds, so as to obtain the instantaneous co-contribution of the co-indicator variables to the main indicator variables (i.e., path coefficients).

[0085] Fourth, based on the sequence of co-indicator variables, the time-varying correlation matrix is ​​calculated using the cosine correlation algorithm, and the correlation pairs with a significance greater than 0.95 are selected by combining the Bayesian model averaging method to obtain the significant covariate correlation coefficient matrix among the co-indicator variables;

[0086] Fifth, based on the above instantaneous principal contribution, instantaneous co-contribution, and covariate correlation coefficient matrix, a structured feature set is constructed that includes the hierarchical relationship between principal and co-indicators, the contribution matrix, and related networks.

[0087] (3) Repeat the process of obtaining the instantaneous main contribution and instantaneous collaborative contribution to obtain the change path of the instantaneous main contribution and the change path of the instantaneous collaborative contribution. Combine the covariate correlation coefficients between the collaborative index variables to construct the dynamic contribution flow network node tree corresponding to a single type of energy data:

[0088] Repeat the above process of obtaining instantaneous main contribution and instantaneous collaborative contribution to obtain the change path of instantaneous main contribution of each main indicator variable to the target result and the change path of instantaneous collaborative contribution of each collaborative indicator variable to the corresponding main indicator variable at continuous time points.

[0089] It should be further explained that the specific process of obtaining the instantaneous main contribution change path of each main indicator variable to the target result and the instantaneous collaborative contribution change path of each coordinating indicator variable to the corresponding main indicator variable at continuous time points in this embodiment includes:

[0090] First, based on historical data, the data is segmented by setting a sliding time window with a length of N time points and a step size of 1. The instantaneous contribution calculation process is repeated for each window, and the contribution value is smoothed by local weighted regression to obtain a continuous contribution sequence after noise suppression.

[0091] Second, based on the instantaneous principal contribution of each principal indicator variable to the target result at continuous time points, a Hidden Markov Model is used to model the state and divide it into high, medium, and low states. The state duration and transition probability are calculated to obtain the state transition pattern of the principal indicator contribution. It should be further noted that the high, medium, and low states are divided by those skilled in the art based on historical information and corresponding specific needs, and will not be elaborated here. When constructing the dynamic contribution flow network node tree corresponding to a single type of energy data, it is necessary to integrate the temporal features based on the principal contributing nodes corresponding to each type of energy data and the instantaneous principal contribution change path to the target node. The state transition pattern of the principal indicator contribution (duration and transition probability of the high, medium, and low states) is one of the core dynamic features of the instantaneous principal contribution change path, and its specific utilization method is as follows:

[0092] The temporal features of the main contributing nodes are input into the graph neural network: When capturing temporal features through graph convolutional networks combined with gated recurrent units, the state transition patterns are transformed into quantified features (such as the transition probability from high to medium state, the average duration of a certain state), which together with the instantaneous numerical change of the main contribution constitute the dynamic attributes of the main contributing nodes. This helps the model to learn the evolution of the contribution of the main index at continuous time points more accurately, and avoids feature loss caused by relying solely on numerical fluctuations and ignoring the trend of state changes.

[0093] Optimize edge weight calculation in the dynamic contribution flow network node tree: When obtaining the dynamic weight matrix between nodes through the Transformer multi-head attention mechanism, the state transition mode serves as a reference for attention allocation. For example, when the contribution of the main indicator transitions from a high state to a medium state with a high probability of transition, the decay rate of the edge weight between the main contributing node and the target node will be adjusted accordingly. This ensures that the edge weight not only reflects the instantaneous contribution level but also the transition trend of the contribution state, enhancing the node tree's ability to represent the dynamic changes in the contribution of the main indicator.

[0094] Ensuring the temporal continuity of multi-version graph structure indexes: When constructing multi-version graph structure indexes using timestamps as index keys through the Neo4j graph database, state transition patterns are bound to the node tree structure at the corresponding time points for storage. For example, high → low state transition records within a certain time window are associated with the node tree of that window. This allows subsequent queries or backtracking of changes in the contribution of the main indicator to quickly locate the node tree version corresponding to key time nodes (such as state mutation points) through the state transition pattern. This ensures that the node tree can fully present the evolution process of the main indicator's contribution from high to medium to low states, rather than an isolated time point structure.

[0095] Third, based on the instantaneous collaborative contribution of each collaborative indicator variable to the corresponding main indicator variable at continuous time points, the temporal features are extracted by the variational autoencoder and the mutation points are screened according to the reconstruction error to obtain the feature sequence marked by the key change points.

[0096] Fourth, based on long-term energy data of different types, the trend term and seasonal term are separated by a seasonal decomposition algorithm, and the missing time points are interpolated and filled by Kalman filtering to obtain a deseasonalized contribution trend sequence.

[0097] Fifth, based on the temporal continuity of the segmented paths of the deseasoned contribution trend sequence, the dynamic time warping (DTW) algorithm is used to align the cross-window contribution changes, thereby obtaining the dynamic change paths of the contribution of the main indicator variable and the co-indicator variable that integrate the trend component, the periodic component, and the mutation point marker.

[0098] It needs further explanation that the periodic component is obtained from the preceding time-series preprocessing steps for the instantaneous main contribution of the main indicator variable. Specifically, the second step has already obtained the instantaneous main contribution of each main indicator variable to the target result at continuous time points (i.e., the main contribution time series) through Hidden Markov Model modeling. The deseasonalized contribution trend series in the fifth step needs to be generated based on this main contribution time series through deseasonalization preprocessing. During the deseasonalization process, the original main contribution time series is decomposed into two parts: one part is the trend component after removing periodic fluctuations (i.e., the main body of the deseasonalized contribution trend series), and the other part is the separated periodic component with a fixed time interval pattern (such as seasonal cycles, time period cycles, and other periodic fluctuation characteristics). The periodic component obtained from this decomposition is the periodic component corresponding to the fusion of the trend component, periodic component, and abrupt change point marker. The process of obtaining the periodic component in this embodiment is as follows:

[0099] Based on the seasonal terms separated from long-term energy data of different types using the seasonal decomposition algorithm in the fourth step, and the long-term energy data without missing time points obtained by interpolating and completing the missing time points using Kalman filtering based on the long-term energy data of different types, the time series information of the instantaneous main contribution of the main indicator variable to the target result and the instantaneous collaborative contribution of the co-indicator variable to the corresponding main indicator variable is obtained. The fixed period type reflected by the seasonal terms is determined by calculating the repetition time interval of the seasonal term fluctuations and the time difference between adjacent fluctuation peaks and troughs through statistical analysis. By taking the seasonal term with the clearly defined fixed period type as the periodic component, the periodic component that can be used to construct the dynamic change path of the contribution of the main indicator variable and the co-indicator variable can be used for subsequent construction of the fusion trend component, periodic component and the mutation point marker.

[0100] The target result is used as the target node of the first layer, the main indicator variable is used as the main contribution node of the second layer, and the sequence of collaborative indicator variables corresponding to each main indicator variable is used as the set of collaborative contribution nodes of the third layer.

[0101] Based on the main contributing node and the instantaneous main contribution change path to the target node corresponding to each type of energy data, the instantaneous collaborative contribution change path of each collaborative contributing node to the corresponding main indicator variable, and the covariate correlation coefficients between the same collaborative contributing node set, a dynamic contribution flow network node tree corresponding to a single type of energy data at continuous time points is constructed by combining a graph neural network with a graph database.

[0102] It should be further explained that the specific implementation process of obtaining the dynamic contribution flow network node tree corresponding to a single type of energy data at continuous time points in this embodiment includes:

[0103] First, a three-layer node set is constructed based on the target result, main indicator variables, and co-indicator variables. The dynamic weight matrix between nodes is obtained by using the main contributing node and the instantaneous main contribution path to the target node, the instantaneous co-contribution path of the co-contributing node, and the covariate correlation coefficient as edge features input to the Transformer multi-head attention mechanism.

[0104] Second, based on the weight matrix and the initial features of the nodes, temporal features are captured by combining a graph convolutional network with a time-gated recurrent unit, and then neighborhood information is weighted and aggregated through an attention mechanism to obtain updated node features.

[0105] Third, based on the node features and edge weights at continuous time points, the graph structure is aligned using the dynamic time warping algorithm, and the topological sorting algorithm is used to eliminate cyclic dependencies, thereby obtaining the topological structure of the directed acyclic graph.

[0106] Fourth, based on this directed acyclic graph, nodes and edges are stored in the Neo4j graph database, and a multi-version graph structure index is constructed using timestamps as index keys. Finally, a single-type energy dynamic contribution flow network node tree containing three-layer node hierarchy, dynamic edge weights, and time-series evolution characteristics is obtained.

[0107] (4) Conduct causal correlation analysis on the aforementioned set of related change attributes to obtain the joint contribution factors of the main indicator variables to the target results:

[0108] Based on the set of associated change attributes of all types of energy parameters at each time point, the causal association analysis algorithm is used to obtain the main joint contribution of all main indicator variables corresponding to each target result to the same target result within a single time period, the association influence relationship and directional association influence direction among all main indicator variables corresponding to each target result, and the directional causal association coefficient between the sequences of synergistic indicator variables corresponding to different main indicator variables.

[0109] The joint contribution factor of each principal indicator variable to the same target result at a single time point is obtained by dividing the instantaneous principal contribution of a single principal indicator variable to the target result by the principal joint contribution of all principal indicator variables corresponding to each target result to the same target result within a single time period.

[0110] It should be further explained that the specific process for obtaining the joint contribution factor of each principal indicator variable to the same target result at a single time point in this embodiment includes:

[0111] First, based on multiple energy principal indicator variables at a single time point, a structural equation model is used for joint modeling, and the standardized path coefficients are solved by maximum likelihood estimation to obtain the direct contribution of each principal indicator variable to the target result.

[0112] Second, based on the set of principal indicator variables, conditional dependencies are inferred through Bayesian networks. The D-separation criterion is used to verify the direction of directed association influence, and the association strength is quantified by combining transfer entropy to obtain the directed association matrix among principal indicators. Here, conditional dependency refers to the association influence relationship and the direction of directed association influence among all principal indicator variables corresponding to each target result. Specifically, when constructing the joint contribution influence factor of principal indicator variables to the target result, the analysis was clearly focused on all principal indicator variables corresponding to each target result within a single time period. The set of principal indicator variables here specifically refers to all principal indicator variables corresponding to the target result. The conditional dependency relationship inferred through Bayesian networks is essentially an analysis of the association characteristics (i.e., association influence relationship) between principal indicator variables within the set under the constraint of the target result. Then, the directionality of this association (i.e., the direction of directed association influence) is verified by using the D-separation criterion. The directed association matrix among principal indicators formed after combining transfer entropy to quantify the association strength is a quantitative representation of the association influence relationship and the direction of directed association influence among all principal indicator variables corresponding to each target result. The two are completely consistent in terms of analysis object and association dimension.

[0113] Third, based on the sequence of synergistic index variables, the causal correlation coefficient is estimated by combining Granger causality test with time-varying parameter vector autoregression model and Markov chain Monte Carlo sampling, and the directed causal correlation coefficient matrix across synergistic indicators is obtained.

[0114] Fourth, based on the direct contribution of each principal indicator variable and the interaction effect among the principal indicator variables, the total contribution of the target result is allocated to each principal indicator variable through the Shapley value decomposition method to obtain the principal joint contribution considering the interaction effect.

[0115] Fifth, based on the individual contribution and joint contribution of a single principal indicator variable, a standardized set of joint contribution influence factors, including the relative importance of the principal indicator, the direction of correlation, and the strength of causality, is obtained by calculating the ratio and using the bootstrap method to correct for bias.

[0116] (5) Based on the joint contribution impact factor and the dynamic contribution flow network node tree corresponding to all types of energy data, construct a cross-type dynamic contribution flow network node forest:

[0117] Based on the joint contribution influence factor of each main indicator variable to the same target result at a single time point, the main contribution relationship connection between each main contribution node and the corresponding target node is constructed.

[0118] Based on the correlation and influence relationship and the direction of the directed correlation among all main indicator variables, a directed correlation connection between the main contributing nodes is constructed through a blockchain algorithm. At the same time, a cross-tree covariate correlation connection is constructed based on the directed causal correlation coefficient between the sequences of co-indicator variables corresponding to different main indicator variables.

[0119] The dynamic contribution flow network node trees corresponding to all types of energy data are timestamped at continuous time points with the constructed main contribution relationship connection, directed association connection and cross-tree covariate association connection. After alignment, the cross-type dynamic contribution flow network node forest is constructed by combining the topology space algorithm.

[0120] It should be further explained that the detailed acquisition process of the cross-type dynamic contribution flow network node forest in this embodiment includes:

[0121] First, based on the joint contribution influence factor of each main indicator variable at a single time point, the main contribution relationship connection is obtained by mapping it to the directed edge weight between the target node and the main contribution node and performing normalization.

[0122] Second, based on the causal relationship between the main contributing nodes inferred from the Bayesian network, the association direction is recorded through the consortium blockchain framework and verified using smart contracts. Combined with the Merkle tree to store historical information on association strength, the directed association connection between the main contributing nodes is obtained.

[0123] Third, based on the directed causal correlation coefficients output by the time-varying parameter vector autoregressive model, the time series of different energy types are aligned through the dynamic time warping algorithm to obtain time-consistent cross-tree covariate correlation connections.

[0124] Fourth, based on the node tree structure of each energy type, a time offset pattern is learned by combining a sliding window matching algorithm with a bidirectional long short-term memory network, and an attention mechanism is introduced to dynamically adjust the alignment weights to obtain a multi-source node tree set with timestamp alignment.

[0125] Fifth, based on the timestamp-aligned node tree, the nodes are mapped to a high-dimensional space by combining topological space algorithms with graph embedding algorithms (such as Node2Vec). The k-nearest neighbor algorithm is used to obtain cross-tree node similarity and combined with hierarchical clustering to obtain a cross-type node forest with multi-scale topological structure.

[0126] Sixth, based on the dynamic features of cross-type node forests and time-series alignment in multi-scale topology, a three-dimensional node forest containing energy type, time series and causal relationship is obtained by integrating the contribution changes in the time dimension and the topological relationship in the spatial dimension through a spatiotemporal graph convolutional network, realizing the visualization and dynamic analysis of contribution flow of cross-type energy data.

[0127] (6) By analyzing the acquisition process of the cross-type dynamic contribution flow network node forest, a reverse synthesis training function is constructed and reverse inference training is performed to obtain the trained directed association graph model:

[0128] Based on the instantaneous contribution mean square error, instantaneous main contribution change path loss, main index correlation cross-entropy loss, cross-tree covariate causal relationship loss, continuous timestamp alignment loss, complexity penalty for constructing the number of main contributing nodes and corresponding co-contributing nodes during the acquisition process of the cross-type dynamic contribution flow network node forest, and the constraint penalty function for the corresponding correlation strength in the cross-type dynamic contribution flow network node forest being less than the preset correlation coefficient, a reverse comprehensive training function is constructed (constructed by adding the errors, losses, and penalties listed above); wherein, the correlation strength refers to the correlation strength between the corresponding connections between nodes in the cross-type dynamic contribution flow network node forest, one being the main contributing node The strength of the principal contribution relationship between a node and its corresponding target node is determined based on the joint contribution factor of each principal indicator variable to the same target result at a single time point, reflecting the degree of correlation between the principal indicator variables and the target result. Secondly, the strength of the directed correlation between principal contributing nodes is determined based on the correlation influence relationship and the direction of the directed correlation influence between principal indicator variables of all types of energy, reflecting the mutual influence strength between principal indicator variables of different types of energy. Thirdly, the strength of the cross-tree covariate correlation is determined based on the directed causal correlation coefficient between the sequences of corresponding co-indicator variables of different principal indicator variables, reflecting the degree of cross-type causal correlation between co-indicator variables of different types of energy. These three types of correlation strength together constitute the quantitative index of the strength of various connection relationships in the cross-type dynamic contribution flow network node forest. When the strength of a certain correlation is less than the preset correlation coefficient, a corresponding constraint penalty will be triggered to optimize the effectiveness of the network structure.

[0129] It should be further explained that the instantaneous contribution mean square error in this embodiment is constructed based on the sum of squared differences between the partial least squares regression (PLS) predicted value and the actual contribution, and is used to quantify the prediction accuracy of the contribution of the main and co-indicators and avoid the estimation bias of the contribution of key variables.

[0130] The instantaneous main contribution change path loss is a weighted sum of the dynamic time warping distance and the hidden state difference of the long short-term memory network, used to constrain the smoothness of the contribution path at continuous time points and capture the temporal evolution pattern of energy data.

[0131] The cross-entropy loss of the main index correlation is constructed by the cross-entropy of the correlation probability distribution inferred by the Bayesian network and the actual causal relationship label. It is used to optimize the identification accuracy of the directed correlation between the main index variables and ensure the logical rationality of the energy transmission path.

[0132] Cross-tree covariate causality loss is the Huber loss that uses the transit entropy estimate and Granger causality test results to balance the estimation bias of causal strength between co-indicator variables and the robustness of outliers.

[0133] The continuous timestamp alignment loss is constructed based on the smoothed L1 loss of the time offset predicted by the bidirectional long short-term memory network and the actual alignment error, and is used to ensure the consistency of the time dimension of multi-energy data.

[0134] The complexity penalty term is constructed by L1 regularization of the number of main contributing nodes and co-contributing nodes and sparse constraints on the edge weights of the graph network. It is used to avoid model overfitting and improve the generalization ability of cross-type energy data.

[0135] The association strength constraint penalty is an exponential penalty imposed on associated edges that are below a preset threshold. It is used to force the preservation of strong associations, simplify the node forest topology, and improve computational efficiency.

[0136] Based on the back-comprehension training function combined with the cross-type dynamic contribution flow network node forest (i.e., training the cross-type dynamic contribution flow network node forest based on the back-comprehension training function), and through random forest combined with simulation algorithm, the trained directed association graph model is obtained through back-inference training from the collaborative contribution node set of the third layer to the target node of the first layer.

[0137] This process, through multi-level dynamic graph modeling and spatiotemporal fusion mechanisms, has for the first time achieved unified representation and deep coupling of heterogeneous energy data from multiple sources, including coal, oil, gas, electricity, and heat. In particular, the three-layer node tree structure based on graph neural networks (target node - main contributing node - set of co-contributing nodes), combined with dynamic time warping and gated loop units, effectively captures the evolution law of the contribution path of the main and co-contributing indicators. The joint contribution influence factor calculated by Bayesian networks and Shapley value decomposition innovatively quantifies the interaction effect between multiple energy main indicator variables, breaking through the limitation of incomparable contribution in traditional single-type energy analysis. Secondly, the blockchain-enabled relational storage mechanism, through smart contracts to verify the causal relationship direction and Merkle tree historical version tracing, constructs an immutable directed relational connection, ensuring the credibility of cross-type transmission paths. In the topological space embedding stage, the graph embedding algorithm is fused with k-nearest neighbor clustering to map multidimensional energy nodes to a high-dimensional continuous space, solving the problem of spatial heterogeneity in heterogeneous energy systems. Third, the three-dimensional modeling capability of spatiotemporal graph convolutional networks enables the visual tracking of wave propagation paths, providing a technical foundation for cross-system source tracing of complex events such as wind and solar power curtailment.

[0138] S2: Obtain cross-type linked reports and construct a basic directed positioning dynamic correction chain through report parsing, positioning, and causal analysis;

[0139] More preferably, a cross-type linkage report is obtained. Based on the key target results with related changes and corresponding causal relationship strength obtained by the entity relationship extraction algorithm in the cross-type linkage report, a basic directed positioning dynamic correction chain is constructed by combining the blockchain algorithm with the position of each key target result in the report.

[0140] The basic directed positioning dynamic correction chain is constructed by combining directed correlation curves with blockchain algorithms. One implementation method in this embodiment is as follows:

[0141] (1) Based on cross-type linked reports, text content is extracted by OCR technology and key target results are located by NLP entity recognition algorithm to obtain a set of key target results containing location coordinates;

[0142] For example, the key target results in this embodiment are the key parameters in the report and the corresponding result data; for example, the current electricity price, and the coal price or coal demand corresponding to the current electricity price.

[0143] (2) Based on the set of key target results, the graph neural network algorithm combined with semantic analysis technology is used to identify the correlation and change relationship between key target results, and the gradient boosting tree is used to calculate the causal correlation strength to obtain key target result pairs with correlation and change relationship and the corresponding causal strength matrix.

[0144] (3) Based on the causal strength matrix, each pair of key target results and their causal correlation strength are converted into a unique hash value through the hash algorithm of the blockchain. Combined with the smart contract to automatically verify the validity of the correlation change relationship, the set of correlation relationships stored on the chain is obtained.

[0145] (4) Based on the cross-type linkage report layout analysis results, the physical coordinates of the key target results in the report are converted into topological structure vectors through relative position coding technology, and the hierarchical relationship is identified by spatial clustering algorithm to obtain the position coding vector of the key target results;

[0146] (5) Based on the set of association relationships and location encoding vectors stored on the chain, a causal relationship network between key target results is constructed by the directed acyclic graph algorithm. The time series is aligned by the dynamic time warping algorithm to obtain a directed positioning dynamic network. Based on the directed positioning dynamic network, the integrity of the network is verified by the zero-knowledge proof algorithm under the premise of protecting data privacy. Multi-party consensus is achieved by combining the threshold signature algorithm. Finally, a basic directed positioning dynamic correction chain containing causal relationships, location encoding and time dimensions is obtained.

[0147] It should be noted that the cross-type linked reports (reports linked by different energy sources and different information) in this embodiment are generated using existing reporting software, such as FineReport, AnyReport and other professional reporting tools, which will not be elaborated on here.

[0148] S3: Input the cross-type dynamic contribution flow network node forest and the basic directed positioning dynamic correction chain into the pre-trained positioning correction path model, and perform real-time positioning error correction adjustment on the cross-type linkage report to obtain the updated and adjusted cross-type linkage report.

[0149] More preferably, based on the cross-type linkage report, the cross-type linkage report is adjusted in real time by a positioning correction path model constructed by the cross-type dynamic contribution flow network node forest, the directed positioning dynamic correction chain and the text positioning algorithm, so as to obtain the updated and adjusted cross-type linkage report.

[0150] The positioning correction path model is constructed by combining cross-type linked reports with a directed positioning dynamic correction chain, such as... Figure 4 As shown, the process includes:

[0151] 1) Obtain cross-type linked report information and locate key target results in the table through report parsing. This includes understanding the sequence of key target results, their synchronous changes, their positions within the table, and their contributing factors. Furthermore, causal analysis is used to identify key target results with correlated changes and the corresponding causal relationship strength.

[0152] The system acquires historical cross-type linkage report information and, through a pre-defined report parsing and localization model (based on Faster R-CNN and OCR / NLP fusion), obtains the sequence of key target results in the cross-type linkage reports, the relationship between changes in each key target result and the synchronous changes in the remaining key target results, the position of each key target result in the cross-type linkage reports, and the contribution factor corresponding to each key target result. The specific implementation process includes:

[0153] First, based on the report image data, Faster R-CNN is used to detect the table region and OCR algorithm is used to extract the text content. NLP named entity recognition algorithm is used to locate key target results and obtain a set of key target results containing location coordinates.

[0154] Second, based on the time series data of key target results, the dependence strength is evaluated by calculating the time series mutual information matrix, the direction of change causality is identified by combining Granger causality test, and then a directed linkage graph is constructed using Bayesian network algorithm to obtain the linkage relationship matrix between key target results.

[0155] Third, based on the report layout analysis results, the position encoding vector of the key target results is obtained by mapping the table cell coordinates to the pixel space and using relative position encoding to represent the hierarchical structure; based on the key target results and related indicator data, the feature importance is calculated by gradient boosting tree, and the causal path graph is constructed by combining structural equation model and the path coefficients are standardized to obtain the direct contribution of each indicator.

[0156] Fourth, based on the direct contribution calculation results, the contribution of interaction effects is allocated through the Shapley value decomposition method. LASSO regularization constraint is introduced to estimate SEM parameters and the strength is optimized through cross-validation to obtain the contribution influence factor for processing high-dimensional sparse data.

[0157] Fifth, based on the characteristics of nonlinear relationships, Gaussian process regression (GPR) is used to fit the function mapping between indicators, and combined with kernel function selection (such as RBF kernel) to capture complex dependencies, so as to obtain the contribution factor tensor containing nonlinear effects.

[0158] Sixth, based on the linkage matrix, location encoding vector, and contribution influence factor tensor, a structured feature set containing three-dimensional information of indicators, time, and target results is constructed to support the accurate calculation of subsequent positioning correction paths.

[0159] Based on the sequence of key target results and the relationship between the synchronous changes of the remaining key target results when each key target result changes, as well as the contribution and influence factors corresponding to each key target result, a causal analysis algorithm is used to obtain the key target results with correlated changes and the corresponding causal correlation strength.

[0160] 2) Based on the position of key target results in the table, the key target results with correlational changes, and the corresponding causal correlation strength, a training and optimization directed localization dynamic correction chain is constructed, and mapped and similarity analyzed with the cross-type dynamic contribution flow network node forest to obtain the similarity consistency error:

[0161] Based on the key target results with correlated changes and the corresponding causal correlation strength, the blockchain algorithm constructs a directed positioning dynamic correction chain with the position of each key target result in the report; it should be further explained that the specific construction process of the directed positioning dynamic correction chain in this embodiment includes:

[0162] First, based on the linkage matrix of the key target result sequence, key target result pairs with causal association strength greater than a threshold are selected to form a set of directed edges; it should be further noted that the linkage matrix in this embodiment is generated by Granger causality test and Bayesian network.

[0163] Second, causal relationships are notified through a consortium blockchain architecture, the directionality of edges is verified through smart contracts, and historical values ​​of association strength at each timestamp are stored using a Merkle tree to ensure traceability.

[0164] Third, the pixel coordinates obtained from the report layout analysis are converted into topological graph node codes, and a multi-dimensional position vector is generated by a relative position embedding algorithm. The neighborhood position features are aggregated through a graph attention network to strengthen spatial correlation constraints. In this embodiment, the causal correlation strength is updated in real time through a dynamic conditional correlation model, and temporal features are extracted by a variational autoencoder to detect anomalies and adjust weights for sudden changes in causal relationships.

[0165] Fourth, in the chain structure construction stage, this embodiment uses key target results as nodes and causal relationships as edges, generates a directed acyclic chain through a topological sorting algorithm, introduces time-dimensional gated cyclic units to capture the temporal delay of causal transmission in the chain, and uses the PBFT consensus algorithm to ensure the consistency of the chain structure across nodes, and realizes the dynamic update of the chain through smart contracts to achieve directed positioning dynamic correction.

[0166] Fifth, when a new relationship change is detected, the chain structure is recalculated. The chain hierarchy is optimized through the Louvain community discovery algorithm, and finally a directed positioning dynamic correction chain containing spatiotemporal location features, dynamic causal weights and time sequence transmission delays is formed to perform real-time positioning and error correction for anomalies in cross-type linkage reports.

[0167] It should be noted that the construction of the directed positioning dynamic correction chain in S2 and S3 belongs to a progressive support relationship of basic construction and training optimization. In S2, the basic construction of the input layer needs to focus on the extraction of key information and basic correlation modeling of the report itself. Its directed positioning dynamic correction chain is based on real-time cross-type linkage reports. It uses OCR text extraction and NLP entity recognition to locate key target results, and combines graph neural network analysis of causal correlation strength and blockchain evidence relationship to build the basic correction chain. The core purpose is to extract the location, causal correlation and time sequence features of key target results from the report to form the core input data of the positioning correction path model. It does not combine the cross-type dynamic contribution flow network node forest. In contrast, S3 is the model layer optimization construction. It needs to combine the dynamic contribution relationship of the node forest to optimize the accuracy and spatiotemporal adaptability of the chain. Its directed positioning dynamic correction chain is based on the linkage relationship matrix of historical cross-type linkage reports in the positioning correction path model training scenario. It further combines the mapping analysis of the cross-type dynamic contribution flow network node forest (similarity algorithm, error calculation), optimizes the hierarchical structure through topological sorting and PBFT. The consensus algorithm ensures consistency, and the gated loop unit captures the timing delay to construct the training optimization and correction chain. The core purpose is to provide a precise carrier for the model to calculate similarity consistency error and construct a positive comprehensive loss function, thus serving model training and error correction.

[0168] Each key objective result in the training optimization directed localization dynamic correction chain is mapped one-to-one with the target nodes in the cross-type dynamic contribution flow network node forest at the same time. At the same time, based on the contribution impact factor corresponding to each key objective result and the set of main contributing nodes and co-contributing nodes in the cross-type dynamic contribution flow network node forest at the same time point, a similarity algorithm is used to obtain the similarity coefficient between the contribution impact factor and the main contributing nodes and co-contributing nodes, and the similarity coefficient between the contribution impact factor and the co-contributing nodes.

[0169] Based on the obtained similarity coefficients, establish a similarity mapping between the contribution impact factor and the main contributing node and the co-contributing node, and construct a similarity consistency error.

[0170] 3) By analyzing similarity consistency error, training and optimizing the corresponding parameter relationships between the directed localization dynamic correction chain and the cross-type dynamic contribution flow network node forest, a positive comprehensive loss function is constructed and forward training is performed to obtain the trained localization correction path model. Based on the loss results of the forward training, a directed association graph model is trained online in real time.

[0171] Based on the directed positioning dynamic correction chain keyword matching algorithm, the positioning correction algorithm combines similarity consistency error, the error between the positioning key target result of the directed positioning dynamic correction chain and the reverse target result of the same target node obtained by the cross-type dynamic contribution flow network node forest back reasoning, the error between the target result after the synchronous change of the remaining key target result corresponding to each key target result in the directed positioning dynamic correction chain and the error between the reverse target result under the same target node as the remaining key target result obtained by the cross-type dynamic contribution flow network node forest back reasoning, and the positive comprehensive loss function constructed by the synchronous change delay of the remaining key target result. Forward training is performed from the first layer corresponding to the target node to the third layer corresponding to the collaborative contribution node set to obtain the trained positioning correction path model.

[0172] It should be further explained that one implementation of this embodiment is as follows:

[0173] Based on the directed localization dynamic correction chain, a semantic mapping between key target results and the cross-type dynamic contribution flow network node forest is established using a keyword matching algorithm (such as BM25) to obtain an initial matching feature vector. Based on this feature vector, a similarity consistency error is calculated using a contrastive loss function, forcing semantic alignment between the contribution influence factor and the main-co-node features to obtain a similarity loss value. For the key target results located by the directed localization dynamic correction chain, the reverse target results of the same target node are obtained through back-inference from the cross-type dynamic contribution flow network node forest. The smoothed L1 error between the two is calculated to obtain the reverse error value of the target node. Based on the target results after synchronous changes of the remaining key target results corresponding to each key target result in the directed localization dynamic correction chain, the reverse target results under the same target node are obtained through back-inference from the cross-type dynamic contribution flow network node forest. The Huber error between the two is calculated. The loss is calculated by obtaining the back-slip error value of the cooperating nodes; based on the time series of synchronous changes in the results of the remaining key targets, the actual propagation delay is estimated by a Hidden Markov Model, and the mean square error between the delay and the model prediction delay is calculated to obtain the variable delay loss value; based on the above similarity loss value, target node back-slip error value, cooperating node back-slip error value and variable delay loss value, a positive comprehensive loss function is constructed by linear combination of an adaptive weight matrix dynamically adjusted by gradient backpropagation; based on the positive comprehensive loss function, the localization correction path model is iteratively trained by forward propagation from the first layer corresponding to the target node to the third layer corresponding to the cooperating contribution node set, combined with random forest and simulated annealing algorithms, to obtain the trained localization correction path model.

[0174] It should be further explained that the similarity consistency error in this embodiment is based on the cosine similarity (similarity coefficient) between the contribution impact factor and the features of the main and co-contributing nodes, and is constructed by using contrastive loss to force similarity mapping. This is used to avoid feature shift caused by semantic gap and ensure semantic alignment between report indicators and node forest. Specifically, it is based on the mapping relationship formed by mapping each key target result in the directed positioning dynamic correction chain to the target nodes at the same time in the cross-type dynamic contribution flow network node forest, the similarity coefficient between the contribution impact factor and the main contributing node and the similarity coefficient between the contribution impact factor and the co-contributing node obtained by the similarity algorithm between the contribution impact factor corresponding to each key target result and the main contributing node and the co-contributing node set at the same time in the cross-type dynamic contribution flow network node forest, and the similarity mapping between the contribution impact factor and the main contributing node and the co-contributing node established based on the above similarity coefficients. The similarity consistency error is obtained by quantifying the deviation of the similarity coefficient in the similarity mapping and combining it with the contrastive loss function to calculate the similarity difference.

[0175] The target node back-inference error is constructed by the smoothed L1 loss of the target result of the directed positioning dynamic correction chain positioning and the node forest back-inference result. It is used to capture the cross-modal transmission deviation of energy data by combining the time consistency of dynamic time warping distance constraints. The cooperative node back-inference error is constructed by the path coefficient difference of the structural equation model. Huber loss is applied to the deviation between the target result after the change of the cooperative contributing node and the node forest inference result to balance the robustness and accuracy of outliers. The variable delay loss is constructed based on the mean square error of the state transition time of the hidden Markov model and the actual transmission delay. In this embodiment, the delay deviation of different time scales is dynamically weighted by introducing an attention mechanism to optimize the time-series response capability of multi-energy data linkage.

[0176] The positive integrated loss result obtained at each step in the training of the positioning correction path model is fed back into the training process (6) of the directed association graph model. Combined with the online training framework, synchronous reverse adjustment training is performed to obtain a directed association graph model trained in real time and synchronously online.

[0177] This process, through the construction of a bidirectional mapping mechanism between a directed localization dynamic correction chain and a cross-type dynamic contribution flow network node forest, has for the first time achieved accurate localization and self-correcting closed-loop of multi-source heterogeneous data in the energy field. In particular, based on a report parsing technology that integrates Faster R-CNN and OCR / NLP, the cell coordinates of physical reports are converted into topological space encoding. Combined with Granger causality tests and a linkage matrix generated by a Bayesian network, a dual binding between report location and energy semantics is established with pixel-level accuracy. This spatial-semantic joint localization mechanism effectively solves the disconnect between location information and business logic in traditional report analysis. Secondly, through a consortium blockchain architecture guaranteed by the PBFT consensus algorithm, it provides distributed evidence of causal correlation strength. Combined with the multi-version tracing capability of Merkle trees, it ensures the immutability of the transmission path. Specifically, the temporal delay features captured by the gated loop unit, in collaboration with the hierarchical optimization algorithm discovered by the Louvain community, construct a chain structure with spatiotemporal adaptive capabilities. When a sudden change in gas price triggers electricity price linkage, this chain can automatically locate the target area in the electricity report and complete the transmission path tracing in a short time, establishing a technical foundation for real-time intervention in emergencies. Thirdly, the cosine similarity calculation of contribution impact factors and cross-type dynamic contribution flow network node forest features, through comparative loss functions to force semantic alignment, solves the terminology gap across energy systems, such as the quantitative equivalence of "peak-valley difference" in electricity and "thermal inertia" in heat. This mapping not only eliminates feature offset but also verifies the equivalence of path coefficients through structural equation modeling, achieving a unified expression of report indicators and forest nodes at the mathematical logic level. Fourth, the smoothed L1 loss in the target node backpropagation error, combined with dynamic time warping, constrains the consistency of wind power output fluctuations in the power, heat, and gas systems. The variable delay loss, through hidden Markov model state transitions and attention weighting mechanisms, optimizes the cascading response speed of "carbon price changes, coal power costs, and industrial electricity prices." The Huber loss ensures the robustness of predicting collaborative contribution nodes under extreme scenarios. The resulting closed-loop training architecture reconstructs the energy data analysis paradigm. In particular, the localization correction path model, through the collaborative optimization of a four-dimensional comprehensive loss function (including similarity consistency, target matching, variable matching, and time delay constraints), establishes a dynamic correction mechanism for linked reports. Based on time-shift learning and attention weight adjustment using a bidirectional long short-term memory network, the response speed to multi-energy chain imbalances under sudden operating conditions is significantly improved.The closed-loop architecture of positive training and negative feedback enables the directed graph model to have continuous evolution capabilities and adapt to market fluctuations. Finally, this embodiment unifies the dynamic representation paradigm of energy data based on spatiotemporal contribution flow modeling, enabling heterogeneous systems to have computable semantic interoperability. The causal blockchain mechanism solves the verification dilemma of cross-type transmission paths, providing credible analysis for complex correlations such as carbon price, coal power and electricity cost cascade effects. The closed-loop correction system realizes the self-iterative optimization of analysis results, promoting the paradigm leap of energy decision-making from static reports to dynamic inference.

[0178] Embodiment 2 of the present invention provides a heterogeneous data processing system for energy big data, such as Figure 5 As shown, it includes:

[0179] The directed association module is used to acquire different types of energy data and input them into a pre-trained directed association graph model to obtain a cross-type dynamic contribution flow network node forest, thereby realizing database construction.

[0180] The causal analysis module is used to obtain cross-type linked reports and construct a basic directed positioning dynamic correction chain through report parsing, positioning, and causal analysis.

[0181] The positioning correction module is used to input the cross-type dynamic contribution flow network node forest and the basic directed positioning dynamic correction chain into the pre-trained positioning correction path model, and to perform real-time positioning error correction adjustment on the cross-type linkage report to obtain the updated and adjusted cross-type linkage report.

[0182] Embodiment 3 of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions;

[0183] The processor is configured to operate according to the instructions to execute the steps of the method.

[0184] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0185] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0186] This invention constructs a directed correlation graph model and a positioning correction path model to form a cross-type dynamic contribution flow network node forest, realizing the deep fusion and dynamic correlation analysis of multi-source heterogeneous energy data. It can eliminate data silos, establish a reliable correlation mapping of cross-type energy data, accurately capture the nonlinear time-varying law of multi-energy coupling, and realize cross-type transmission analysis of the dynamic contribution of the main-co-indicators.

[0187] This invention enhances the model's ability to learn the correlation patterns of energy data by integrating training functions and back reasoning training. Combined with the positioning and correction path model, it enables real-time error correction of cross-type linkage reports and dynamically corrects multi-energy chain imbalances.

[0188] This invention utilizes an online synchronous training mechanism to adapt to dynamic changes in the energy system, providing accurate analysis results for energy forecasting with interconnected self-correcting capabilities, thereby improving the accuracy and timeliness of energy decision-making.

[0189] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0190] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0191] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0192] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A heterogeneous data processing method for energy big data, characterized in that, The method includes: Different types of energy data are acquired and input into a pre-trained directed graph model to obtain a cross-type dynamic contribution flow network node forest; Obtain cross-type linked reports and construct a basic directed positioning dynamic correction chain through report parsing, positioning, and causal analysis; The cross-type dynamic contribution flow network node forest and the basic directed positioning dynamic correction chain are input into the pre-trained positioning correction path model to perform real-time positioning error correction adjustment on the cross-type linkage report, and obtain the updated and adjusted cross-type linkage report. The training process of the directed graph model includes: Historical information on various types of energy is obtained and key variables are extracted to obtain a time series set of energy parameters and a set of associated change attributes for each type of energy. Multi-stage time-series factor analysis was performed on the energy parameter time series set to obtain the main indicator variable and the corresponding co-indicator variable sequence, the instantaneous main contribution of the main indicator variable to the target result, the instantaneous co-contribution of the co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables. Repeat the process of obtaining the instantaneous main contribution and instantaneous collaborative contribution to obtain the change path of the instantaneous main contribution and the change path of the instantaneous collaborative contribution. Combine the covariate correlation coefficients between collaborative indicator variables to construct a dynamic contribution flow network node tree corresponding to a single type of energy data. A causal correlation analysis was performed on the aforementioned set of related change attributes to obtain the joint contribution factors of the main indicator variables to the target outcome, including: Based on the set of associated change attributes of all types of energy parameters at each time point, the causal association analysis algorithm is used to obtain the main joint contribution of all main indicator variables corresponding to each target result to the same target result within a single time period, the association influence relationship and directional association influence direction among all main indicator variables corresponding to each target result, and the directional causal association coefficient between the sequences of synergistic indicator variables corresponding to different main indicator variables. The joint contribution factor of each principal indicator variable to the same target result at a single time point is obtained by dividing the instantaneous principal contribution of a single principal indicator variable to the target result by the principal joint contribution of all principal indicator variables corresponding to each target result to the same target result within a single time period. Based on the joint contribution impact factor and the dynamic contribution flow network node tree corresponding to all types of energy data, a cross-type dynamic contribution flow network node forest is constructed, including: Based on the joint contribution influence factor of each main indicator variable to the same target result at a single time point, the main contribution relationship connection between each main contribution node and the corresponding target node is constructed. Based on the correlation and influence relationship and the direction of the directed correlation among all main indicator variables, a directed correlation connection between the main contributing nodes is constructed through a blockchain algorithm. At the same time, a cross-tree covariate correlation connection is constructed based on the directed causal correlation coefficient between the sequences of co-indicator variables corresponding to different main indicator variables. The dynamic contribution flow network node trees corresponding to all types of energy data are connected to the main contribution relationship, directed association connection and cross-tree covariate association connection at continuous time points. After alignment, the topology space algorithm is combined to construct a cross-type dynamic contribution flow network node forest. By analyzing the acquisition process of the cross-type dynamic contribution flow network node forest, a reverse synthesis training function is constructed and reverse inference training is performed to obtain a trained directed association graph model.

2. The heterogeneous data processing method for energy big data according to claim 1, characterized in that: The process of acquiring historical information on various types of energy and extracting key variables to obtain a time series set of energy parameters and a set of associated change attributes for each type of energy includes: The system acquires historical demand change information, cost change information, and related change information for each type of energy, extracts key variables, and obtains the time series set of energy parameters and the set of related change attributes for each type of energy.

3. The heterogeneous data processing method for energy big data according to claim 1, characterized in that: The multi-stage time-series factor analysis of the energy parameter time series set yields the main indicator variable and its corresponding co-indicator variable sequences, the instantaneous main contribution of the main indicator variable to the target result, the instantaneous co-contribution of the co-indicator variable sequences to the corresponding main indicator variables, and the covariate correlation coefficients among the co-indicator variables, including: Based on the time series set of energy parameters for each type of energy, the parameter loading matrix corresponding to continuous time points of each type of energy parameter is obtained through a multi-stage time series factor analysis model. Statistical analysis is performed on the parameter loading matrix corresponding to each time point to obtain the sequence of each type of main indicator variable and its corresponding co-indicator variable, the instantaneous main contribution of each main indicator variable to the target result, the instantaneous co-contribution of each co-indicator variable sequence to the corresponding main indicator variable, and the covariate correlation coefficient between the co-indicator variables.

4. The heterogeneous data processing method for energy big data according to claim 1, characterized in that: The process of repeatedly obtaining the instantaneous primary contribution and instantaneous collaborative contribution is used to obtain the change paths of the instantaneous primary contribution and instantaneous collaborative contribution. Combined with the covariate correlation coefficients between collaborative indicator variables, a dynamic contribution flow network node tree corresponding to a single type of energy data is constructed, including: Repeat the process of obtaining instantaneous main contribution and instantaneous collaborative contribution to obtain the change path of instantaneous main contribution of each main indicator variable to the target result and the change path of instantaneous collaborative contribution of each collaborative indicator variable to the corresponding main indicator variable at continuous time points. The target result is used as the target node of the first layer, the main indicator variable is used as the main contribution node of the second layer, and the sequence of collaborative indicator variables corresponding to each main indicator variable is used as the set of collaborative contribution nodes of the third layer. Based on the main contributing node and the instantaneous main contribution change path to the target node corresponding to each type of energy data, the instantaneous collaborative contribution change path of each collaborative contributing node to the corresponding main indicator variable, and the covariate correlation coefficients between the same collaborative contributing node set, a dynamic contribution flow network node tree corresponding to a single type of energy data at continuous time points is constructed by combining a graph neural network with a graph database.

5. The heterogeneous data processing method for energy big data according to claim 4, characterized in that: The process of constructing a reverse synthesis training function and performing reverse inference training by analyzing the acquisition process of the cross-type dynamic contribution flow network node forest to obtain the trained directed association graph model includes: Based on the instantaneous contribution mean square error, instantaneous main contribution change path loss, main index correlation cross-entropy loss, cross-tree covariate causal relationship loss, continuous timestamp alignment loss, complexity penalty for the construction of the number of main contributing nodes and corresponding collaborative contributing nodes, and constraint penalty function for the corresponding correlation strength in the cross-type dynamic contribution flow network node forest being less than the preset correlation coefficient, a reverse comprehensive training function is constructed. Based on the inverse synthesis training function combined with the cross-type dynamic contribution flow network node forest, and through random forest combined with simulation algorithm, the trained directed association graph model is obtained by inverse reasoning training from the third layer collaborative contribution node set to the first layer target node.

6. The heterogeneous data processing method for energy big data according to claim 1, characterized in that: The training process of the localization correction path model includes: By acquiring historical cross-type linkage report information and locating the key target results in the table through report parsing, we can obtain the sequence of key target results in the table and the relationship of synchronous changes in key target results, the position of key target results in the table and their contribution factors. Through causal analysis, we can obtain key target results with related changes and the corresponding causal relationship strength. Based on the position of key target results in the table, key target results with related changes and the corresponding causal correlation strength, a training optimization directed localization dynamic correction chain is constructed, and it is mapped and similarly analyzed with the cross-type dynamic contribution flow network node forest to obtain the similarity consistency error. By analyzing similarity consistency error, training and optimizing the corresponding parameter relationships between the directed localization dynamic correction chain and the cross-type dynamic contribution flow network node forest, a positive comprehensive loss function is constructed and forward training is performed to obtain the trained localization correction path model. Based on the loss results of the forward training, the directed association graph model is trained online in real time.

7. The heterogeneous data processing method for energy big data according to claim 6, characterized in that: The process involves acquiring historical cross-type linked report information and locating key target results in the table through report parsing, obtaining the sequence of key target results and the synchronous changes of key target results, the position of key target results in the table and their contributing factors, and through causal analysis, obtaining key target results with correlated changes and the corresponding causal correlation strength, including: Obtain historical cross-type linkage report information, and through the preset report parsing and positioning model, obtain the sequence of key target results in the cross-type linkage report, the relationship between the synchronous change of the remaining key target results when each key target result changes, the position of each key target result in the cross-type linkage report, and the contribution impact factor corresponding to each key target result. Based on the sequence of key target results and the relationship between the synchronous changes of the remaining key target results when each key target result changes, as well as the contribution and influence factors corresponding to each key target result, a causal analysis algorithm is used to obtain the key target results with correlated changes and the corresponding causal correlation strength.

8. The heterogeneous data processing method for energy big data according to claim 6, characterized in that: The training and optimization of a directed localization dynamic correction chain is constructed based on the position of key target results in the table, the key target results with correlation and change relationships, and the corresponding causal correlation strength. This chain is then mapped and similarly analyzed with the cross-type dynamic contribution flow network node forest to obtain the similarity consistency error, including: Based on the key target results with related changes and the corresponding causal relationship strength and the position of each key target result in the report, a training and optimization directed positioning dynamic correction chain is constructed by combining blockchain algorithms. Each key objective result in the training optimization directed localization dynamic correction chain is mapped one-to-one with the target nodes in the cross-type dynamic contribution flow network node forest at the same time. At the same time, based on the contribution impact factor corresponding to each key objective result and the set of main contributing nodes and co-contributing nodes in the cross-type dynamic contribution flow network node forest at the same time point, a similarity algorithm is used to obtain the similarity coefficient between the contribution impact factor and the main contributing nodes and co-contributing nodes, and the similarity coefficient between the contribution impact factor and the co-contributing nodes. Based on the obtained similarity coefficients, establish a similarity mapping between the contribution impact factor and the main contributing node and the co-contributing node, and construct a similarity consistency error.

9. The heterogeneous data processing method for energy big data according to claim 6, characterized in that: The process involves analyzing similarity consistency errors, training and optimizing the corresponding parameter relationships between the directed localization dynamic correction chain and the cross-type dynamic contribution flow network node forest to construct a forward comprehensive loss function and perform forward training to obtain a trained localization correction path model. Based on the loss results of the forward training, a directed association graph model is trained online in real time, including: Combining similarity consistency error, the error between the location key target result of the training and optimization directed positioning dynamic correction chain and the reverse target result of the same target node obtained by the cross-type dynamic contribution flow network node forest back reasoning, the error between the target result after the synchronous change of the remaining key target result corresponding to each key target result in the training and optimization directed positioning dynamic correction chain and the error between the reverse target result under the same target node as the remaining key target result obtained by the cross-type dynamic contribution flow network node forest back reasoning, and the positive comprehensive loss function constructed by the synchronous change delay of the remaining key target result, positive training is performed from the first layer corresponding to the target node to the third layer corresponding to the collaborative contribution node set to obtain the trained positioning correction path model; The positive integrated loss result obtained at each step of the localization correction path model training is fed back to the directed relation graph model for synchronous reverse adjustment training, thereby obtaining a directed relation graph model trained in real time and synchronously online.

10. A heterogeneous data processing system for energy big data, comprising the method described in any one of claims 1-9, characterized in that, The system includes: The directed association module is used to acquire different types of energy data and input them into a pre-trained directed association graph model to obtain a cross-type dynamic contribution flow network node forest. The causal analysis module is used to obtain cross-type linked reports and construct a basic directed positioning dynamic correction chain through report parsing, positioning, and causal analysis. The positioning correction module is used to input the cross-type dynamic contribution flow network node forest and the basic directed positioning dynamic correction chain into the pre-trained positioning correction path model, and to perform real-time positioning error correction adjustment on the cross-type linkage report to obtain the updated and adjusted cross-type linkage report.

11. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-9.

Images