Power distribution network voltage quality online evaluation method and system supporting new energy group access

Abstract

The invention provides a power distribution network voltage quality online evaluation method and system supporting new energy group access, and belongs to the field of power system automation. Comprising the following steps: generating a node dynamic feature vector through multi-source data fusion and feature extraction; constructing a dynamic causal association diagram, and synchronously outputting a voltage quality multi-dimensional evaluation index and disturbance source contribution degree analysis by using a gating space-time diffusion mechanism; and generating graded early warning based on evaluation and traceability results, and generating auxiliary decision-making information of recommended control measures through strategy matching, effect simulation and multi-target optimization. The system comprises a data acquisition and fusion module, a dynamic causal association diagram construction module, a causal time sequence diagram diffusion evaluation module, an endogenous traceability analysis module, a grading early warning and decision auxiliary module and the like. According to the invention, real-time evaluation, accurate traceability and intelligent decision support of the voltage quality problem of the new energy power distribution network are realized.

Description

Technical Field

[0001] This invention belongs to the field of power system operation and automation technology, specifically relating to an online voltage quality assessment method and system for distribution networks that supports the access of new energy clusters. Background Technology

[0002] With the advancement of dual-carbon goals, the penetration rate of distributed new energy sources, represented by photovoltaics and wind power, in distribution networks is continuously increasing. New energy power generation is characterized by intermittency, randomness, and volatility. Its large-scale grid connection has led to the transformation of distribution networks from traditional radial passive networks to complex active networks with bidirectional power flow. This presents severe challenges to the operation and control of distribution networks, especially voltage quality control. Traditional distribution network voltage regulation mainly relies on local control methods such as on-load tap-changing transformers and capacitor switching. Its design is based on the premise of relatively stable load and a single power flow direction, making it difficult to adapt to the frequent voltage fluctuations and limit-breaking problems caused by rapid changes in new energy output.

[0003] Currently, voltage quality assessment and control technologies for distribution networks containing renewable energy sources still have significant shortcomings. First, at the assessment level, existing methods largely rely on offline simulations or simple threshold exceedance judgments, making it difficult to quantify multi-dimensional indicators such as voltage deviation, fluctuations, and risk probabilities in real time and with high accuracy. In particular, they lack the ability to characterize the dynamic correlation between renewable energy fluctuations and voltage problems. Second, at the problem analysis level, when voltage anomalies occur, existing technologies can usually only indicate the location of the problem, but cannot quickly and accurately answer what caused the problem or the degree of impact of each disturbance source; that is, they lack effective disturbance tracing capabilities. Finally, at the decision support level, existing control strategies often rely on manual experience or simple matching based on fixed rules, lacking rapid simulation and multi-objective optimization comparison of the effects of control measures. This makes it difficult to generate scientific and efficient control commands in complex and ever-changing renewable energy scenarios. These deficiencies lead to a dilemma for operators when facing voltage problems: unclear status, inability to pinpoint the root cause, and indecisiveness, affecting the safe and stable operation of the power grid and the efficient absorption of renewable energy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide an online voltage quality assessment method and system for distribution networks supporting the integration of renewable energy clusters. This system enables real-time and accurate voltage quality assessment, rapid source tracing of disturbances, and intelligent early warning and decision support. The invention can simultaneously output multi-dimensional voltage indicators and disturbance source contribution levels, and generate tiered early warning and optimal control recommendations based on the results. This solves the problems of delayed assessment, unclear root causes, and reliance on experience in decision-making in existing technologies, thereby improving the grid's adaptability to renewable energy fluctuations and operational safety.

[0005] To achieve the above objectives, the present invention adopts the following technical solution; A method for online voltage quality assessment of distribution networks supporting the integration of renewable energy clusters includes: By collecting and preprocessing multi-source data from the distribution network, physical features characterizing the electrical correlation strength and operational status deviation of new energy sources are extracted and fused with the original measurement data to reduce dimensionality, generating dynamic fusion feature vectors for nodes. A dynamic causal relationship graph representing the causal influence relationship between nodes is constructed. The node state is updated using a gated spatiotemporal diffusion mechanism. Based on the final node state, the voltage quality assessment index of each node and the source tracing results and contribution of the upstream disturbance source node that caused its voltage quality problem are output synchronously. Based on voltage quality assessment indicators and source tracing results, a graded early warning signal containing root cause location information is generated; and based on the source tracing results, candidate control measures are matched, and the effects are simulated through a power grid response proxy model to generate auxiliary decision-making information containing recommended control measures and their expected effects.

[0006] As a further aspect of the present invention, the step of generating the dynamic fusion feature vector of the node includes: Real-time acquisition of measurement data and topology status data of multiple nodes in the distribution network, and time alignment, anomaly removal, missing value filling and standardization processing of the measurement data; Based on topology state data and measurement data, voltage correlation characteristics and operational residual characteristics are calculated for each node. Voltage correlation characteristics are used to characterize the electrical correlation strength between the node voltage and one or more preset new energy dominant nodes, while operational residual characteristics are used to characterize the degree of deviation between the node's current operating state and the expected or typical state. The preprocessed raw measurement data, voltage correlation features, and operational residual features are spliced ​​together to form a preliminary fusion feature vector. Then, a feature selection method based on correlation is used to reduce the dimensionality of the preliminary fusion feature vector to form a dynamic fusion feature vector.

[0007] As a further aspect of the present invention, the method for the voltage correlation feature includes: The electrical correlation strength is calculated based on the electrical distance, where the electrical distance is the magnitude of the equivalent impedance between two nodes, and the electrical correlation strength is taken as the reciprocal of the electrical distance; or, the voltage sensitivity coefficient is estimated based on the simplified power flow equation, and the sensitivity of the voltage of this node to the power injection of the new energy head node is calculated by obtaining or estimating the network admittance matrix online and combining it with the current operating point.

[0008] As a further aspect of the present invention, the calculation method for the running residual characteristics includes: The operational residual characteristics include power residual and voltage residual. The power residual is obtained by calculating the difference between the current active power value of the node and the historical average active power value for the same period, and dividing the difference between the current active power value of the node and the historical average active power value for the same period by its historical standard deviation. The voltage residual is obtained by calculating the difference between the current voltage amplitude of the node and the regional reference voltage or rated voltage, and dividing the difference between the current voltage amplitude of the node and the regional reference voltage or rated voltage by the historical voltage standard deviation or the preset allowable fluctuation range.

[0009] As a further aspect of the present invention, the step of dimensionality reduction using the correlation-based feature selection method includes: First, calculate the Pearson correlation coefficient between each feature dimension in the preliminary fused feature vector and the target voltage quality index. The Pearson correlation coefficient measures the degree of linear correlation between two variables, and the larger the absolute value, the stronger the correlation with the target. The correlation coefficient calculation formula involves the ratio of the covariance of the feature value and the target value to their respective standard deviations. Pearson correlation between feature dimensions and target metrics The expression is: ; This represents the number of training samples; For feature dimensions In the The values ​​that can be taken on each sample; For feature dimensions The mean across all samples; For the first The target voltage quality index value corresponding to each sample; The mean of the target indicator across all samples;

[0010] After calculating the correlation coefficients of all feature dimensions, all features are sorted from largest to smallest absolute value; the top Q feature dimensions with the largest absolute values ​​of correlation coefficients are selected to form the final dynamic fusion feature vector of each node. The dimensionality reduction process significantly reduces the feature dimensions and computational complexity while retaining the core information most relevant to voltage quality issues.

[0011] As a further aspect of the present invention, the synchronous output of voltage quality assessment indicators for each node, as well as the source tracing results and contribution of upstream disturbance source nodes causing voltage quality problems, includes: The power distribution network is abstracted as a directed graph; vertices in the directed graph correspond to power grid nodes, and the vertex attributes are dynamically fused feature vectors; edges in the directed graph represent the causal relationship between nodes, and the edge weights represent the intensity of the causal relationship. Causal diffusion and state update are based on a dynamic causal relationship graph. The process of causal information diffusion is performed, and the state representation of each node is iteratively updated according to the update rules. The current state of a node is determined by its previous state and the information from its upstream causal neighbors. The integrated output, based on the final node state obtained from the diffusion process, synchronously outputs voltage quality assessment indicators and the source tracing results and contribution through a parallel voltage quality assessment module and an endogenous source tracing module.

[0012] As a further aspect of the present invention, the edge construction step in the construction of the dynamic causal relationship graph includes: Initial physical association construction involves establishing directed edges between physically connected nodes and setting initial weights for these edges, which are the reciprocal of the electrical distance between the nodes. Dynamic causal strength learning employs an information-theoretic-based transfer entropy algorithm to calculate the dynamic causal influence strength between nodes. For any pair of nodes, the historical sequence of their dynamic features within a preset sliding time window is collected, and the transfer entropy from the source node to the target node is calculated. The transfer entropy is normalized to obtain the normalized causal influence strength value, which is used as the weight of the corresponding directed edge. The causal influence strength of all node pairs constitutes an asymmetric causal adjacency matrix, which is used to represent the weight of the edges in the directed graph.

[0013] As a further aspect of the present invention, the update rule is as follows: For each node, the weighted average state of all its causal upstream neighbors is calculated, with the weights being the normalized values ​​of the causal influence strength of the corresponding edge. Then, this weighted average state is combined with the node's previous state in a weighted manner, with the combination weight controlled by a preset information reception rate parameter. The initial state of the node is set as the node's dynamic fusion feature vector. The diffusion process stops after iterating for a preset number of steps. The final state of each node contains global information propagated from multi-hop neighbors along the causal path. For each target node, its position at the 1st... The state after step diffusion Calculated using the following formula: ; For the target node In the The state after the first diffusion step; For the target node The initial state; For information reception rate; For the target node The causal upstream neighbor node, that is, the source node; For the target node The causal upstream neighbor set; As the source node In the The state vector of the step; This is a temporary index variable for summation, used to normalize the denominator, and iterates through the set. , This indicates that for all target nodes Sum of causal strengths; This represents the normalized causal influence strength value.

[0014] As a further aspect of the present invention, the voltage quality assessment module includes a set of predefined assessment functions for mapping the final state of a node to specific voltage quality indicators. The assessment functions are designed based on the actual needs of voltage quality assessment. The voltage quality assessment module includes a voltage deviation index calculation unit, a voltage fluctuation intensity calculation unit, and a voltage limit exceedance probability calculation unit. The voltage deviation index reflects the degree to which the current voltage deviates from the rated value; Voltage fluctuation intensity is calculated based on the temporal variation characteristics implicit in the state, determining short-time voltage fluctuations. Voltage exceedance probability is calculated using a logic function or probability model to determine the risk of voltage exceeding a safe threshold.

[0015] As a further aspect of the present invention, the endogenous source tracing module calculates the contribution of each potential source node to the voltage problem of the target node based on the causal information flow accumulated during the diffusion process, thereby realizing the root cause location of the voltage quality problem. The endogenous source tracing module specifically includes the following calculation steps: Calculate the cumulative causal effect strength: calculate the sum of the multi-step causal effect strengths from each potential source node to the target node; From the source node To the target node The intensity of cumulative causal influence The formula is: ; To consider the maximum number of propagation steps, For the causal adjacency matrix Power of 1 The propagation step index represents the number of steps taken in the causal propagation path from the source node to the target node; As the source node To the target node go through The causal strength of step propagation; Calculate the anomaly degree of the source node: calculate the degree of deviation between the current state and the historical normal state of each potential source node. ; ; As the source node Dynamically fused feature vectors; As the source node The feature mean vector under normal operating conditions; Indicates Euclidean distance; This is the weighting coefficient, which can be set according to specific needs; Calculate the unnormalized contribution: Multiply the cumulative causal influence strength by the anomaly degree of the source node to obtain the unnormalized contribution of each potential source node; Calculate the contribution percentage: Normalize the unnormalized contribution of all potential source nodes to obtain the contribution percentage of each node, and output the main perturbation source nodes.

[0016] As a further aspect of the present invention, the generation of auxiliary decision-making information including recommended control measures and their expected effects includes: The root cause localization and hierarchical early warning generation is based on voltage quality assessment indicators and source tracing results. It triggers hierarchical early warnings according to preset multi-level threshold rules. The early warning information indicates the abnormal node, the type and degree of exceeding the limit, and lists the main disturbance source nodes and their contribution. Candidate control measures are matched by selecting candidate measures from a pre-set control strategy library based on the type of disturbance source and the nature of the voltage problem. Rapid effect simulation: Call the power grid response proxy model to simulate the effects of candidate measures and predict changes in voltage at key nodes; Generate auxiliary decision-making information, comprehensively compare the simulation effects, operating costs and potential risks of each candidate measure, and generate a recommended decision report through multi-objective evaluation.

[0017] Another aspect of this application provides an online voltage quality assessment system for distribution networks that supports the integration of new energy clusters, including: The data acquisition and fusion module is configured to collect and preprocess multi-source data of the distribution network in real time, extract physical features that characterize the electrical correlation strength and operational status deviation of new energy sources, and fuse them with the original measurement data to reduce dimensionality and generate dynamic fusion feature vectors for nodes. The causal diffusion assessment and endogenous source tracing module is configured to construct a dynamic causal relationship graph that characterizes the causal influence relationship between nodes. It uses a gated spatiotemporal diffusion mechanism to update the node state and, based on the final node state, synchronously outputs the voltage quality assessment index of each node as well as the source tracing results and contribution of the upstream disturbance source node that caused its voltage quality problem. The early warning and decision support module is configured to generate a graded early warning signal containing root cause location information based on voltage quality assessment indicators and source tracing results; and, based on the source tracing results, match candidate control measures, perform effect simulation through a power grid response proxy model, and generate auxiliary decision information containing recommended control measures and their expected effects.

[0018] As a further aspect of the present invention, the voltage quality assessment module in the causal diffusion assessment and endogenous tracing module includes a set of predefined assessment functions for mapping the final state of a node to a specific voltage quality index. The assessment functions include at least a voltage deviation index calculation function, a voltage fluctuation intensity calculation function, and a voltage limit exceedance probability calculation function.

[0019] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are as follows: Compared with existing technologies, the online voltage quality assessment method and system for distribution networks supporting the access of new energy clusters provided by this invention, through the organic integration of a series of technical steps such as multi-source data fusion, dynamic causal graph modeling, gated spatiotemporal diffusion, and intelligent decision generation, produces the following comprehensive beneficial effects: First, this invention significantly improves the real-time performance, accuracy, and comprehensiveness of voltage quality assessment. Step S100 performs online preprocessing of multi-source data and extracts key physical features such as the correlation strength of new energy electrical systems and the deviation of operating status. This is then dimensionality-reduced to obtain a high-information-density dynamic feature vector, providing high-quality input for the assessment. Furthermore, in step S200, utilizing a constructed dynamic causal relationship graph and a gated diffusion mechanism, node states are continuously updated and integrated with global information during the simulation of disturbance propagation. Ultimately, this allows for the simultaneous output of precisely quantified multi-dimensional indicators such as voltage deviation index, fluctuation intensity, and over-limit probability. This integrated process overcomes the shortcomings of existing methods, such as assessment lag and single-dimensionality limitations, achieving real-time, panoramic, and accurate perception of the grid voltage status.

[0020] Secondly, this invention innovatively achieves endogenous source tracing and contribution quantification of voltage quality problems, solving the problem of unclear root causes. Within the same causal diffusion framework of S200, the system not only outputs evaluation indicators but also automatically calculates the cumulative causal influence strength from each potential source node to the target node based on the causal information flow accumulated during the diffusion process and the exponentiation of the causal adjacency matrix. Combined with the node's own anomaly degree, it ultimately outputs the main disturbance source node and its specific contribution percentage. This beneficial effect directly addresses and solves the deficiency of existing methods mentioned in the background art, which lack effective source tracing capabilities. It enables operators to quickly locate the responsible source point causing voltage anomalies and understand the magnitude of its impact, laying a solid foundation for subsequent precise control.

[0021] Third, this invention forms a complete auxiliary closed loop from state assessment to intelligent decision-making, greatly improving the scientific nature and efficiency of operation control. Based on the assessment and tracing results output by S200, step S300 first generates a hierarchical early warning that integrates root cause location information, achieving accurate alarms. Subsequently, by matching candidate measures from a pre-set strategy library, using a power grid response proxy model for rapid effect simulation, and employing a multi-objective scoring function for optimization and comparison, an auxiliary decision-making report containing recommended measures and expected effects is finally generated. The beneficial effect of this series of steps is that it seamlessly connects the traditionally dispersed assessment, analysis, and decision-making processes, automating them into a coherent intelligent process. This directly solves the problems of decision support relying on manual intervention and lacking optimization in the background technology, effectively assisting operators in quickly formulating scientific and economical control schemes, and improving the power grid's autonomous adaptation and security defense capabilities against new energy fluctuations. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the online voltage quality assessment method for distribution networks supporting the integration of renewable energy clusters; Figure 2 S100 flowchart for online voltage quality assessment method of distribution network to support the access of new energy clusters; Figure 3 A flowchart of the online voltage quality assessment method S200 for distribution networks supporting the access of new energy clusters; Figure 4 S300 flowchart for online voltage quality assessment method of distribution network to support the access of new energy clusters; Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0025] The core objective of this invention is: The core objective of this invention is to construct a complete and efficient online voltage quality assessment and intelligent decision support system for distribution networks that supports the integration of renewable energy clusters. This core objective is specifically manifested in the following three aspects: The first goal is to achieve real-time and accurate assessment of voltage quality in distribution networks with high penetration rates of renewable energy. The method needs to overcome the shortcomings of traditional assessment techniques in adapting to the randomness and volatility of renewable energy output such as wind and solar power. It should be able to accurately and online assess the degree of voltage deviation, short-term fluctuation intensity, and probability of exceeding limits in dynamic scenarios where the proportion of renewable energy integration is constantly increasing, providing operators with a clear and accurate profile of the power grid status.

[0026] Secondly, it enables rapid source tracing and impact quantification of voltage quality issues. This requires not only the ability to determine if voltage is abnormal, but also the capacity to investigate the root causes. Through an endogenous source tracing mechanism, it automatically locates upstream disturbance sources that trigger voltage problems, such as specific photovoltaic power plants or wind farm clusters, and accurately calculates the percentage contribution of each disturbance source to the voltage problem at the target node. This deepens the assessment from simply identifying what happened to understanding why it happened and whose influence is greater.

[0027] Thirdly, it aims to achieve intelligent closed-loop support from state perception to control decision-making. Based on accurate assessment and root cause analysis results, the system needs to automatically trigger tiered early warning signals containing root cause location information, and intelligently match candidate control measures from a pre-set strategy library based on the root cause analysis conclusions. Furthermore, a lightweight power grid response proxy model is used to perform rapid effect simulation and multi-objective optimization comparison of candidate measures, ultimately generating an auxiliary decision-making report containing recommended measures, expected effects, execution parameters, and risk warnings, providing dispatching and operation personnel with complete decision support from early warning to response.

[0028] The underlying logic and technical approach of this method: This method follows a clear technical route from data to knowledge to decision-making. Its internal logic is the organic connection of three main stages: data fusion and feature extraction, causal modeling and integrated assessment, and accurate early warning and intelligent decision support.

[0029] The first stage is data fusion and feature extraction, corresponding to step S100 in the instruction manual. The logical starting point for this stage is the multi-source heterogeneous data of the distribution network, including node voltage, active and reactive power measurement data, and network topology connection data. First, this data undergoes cleaning, alignment, and standardization preprocessing to eliminate dimensions and ensure data quality. Then, two types of core physical features are extracted online: one is voltage correlation features, used to quantify the electrical correlation strength between the current node's voltage and the preset renewable energy dominant nodes, specifically achieved by calculating the reciprocal of the electrical distance or a voltage sensitivity coefficient based on simplified power flow equations; the other is operational residual features, used to characterize the deviation of the node's current operating state from its historical normal operating conditions, achieved by calculating the standardized difference between the current power value and the historical average for the same period, and the standardized residual between the current voltage and the regional reference voltage. Finally, the preprocessed raw measurement data is concatenated with these two types of physical features to form a high-dimensional preliminary fusion feature vector. Then, a feature selection method based on the Pearson correlation coefficient is used for dimensionality reduction, selecting the feature dimensions most relevant to the voltage quality target, generating a low-dimensional, high-information dynamic fusion feature vector for each node.

[0030] The second stage is causal diffusion assessment and endogenous source tracing, corresponding to step S200 in the instruction manual. The core logic of this stage is to construct a dynamic causal relationship graph and utilize a gated diffusion mechanism to achieve integrated output for assessment and source tracing. First, the system abstracts the distribution network as a directed graph, where the node attributes are the dynamically fused feature vectors generated in the previous stage. The construction of the edge set includes two sub-steps: first, initial physical association construction, establishing directed edges between physically directly connected nodes, with weights set to the reciprocal of the electrical distance; second, dynamic causal strength learning, employing an information-theoretic-based transfer entropy algorithm to calculate the dynamic causal influence strength between any pair of nodes, and performing normalization processing to ultimately form an asymmetric adjacency matrix characterizing the directed causal propagation path. Based on this, the causal information diffusion process is executed, iteratively updating the state representation of each node through a gated update formula, ensuring it not only contains its own information but also incorporates multi-hop neighbor information propagating along the causal chain. Based on the final node state after diffusion convergence, the results are synchronously output through the parallel voltage quality assessment module and endogenous source tracing module. The evaluation module maps node states to specific indicators such as voltage deviation index, voltage fluctuation intensity, and voltage limit exceedance probability. The source tracing module calculates the multi-step cumulative causal influence intensity from each potential source node to the target node, and combines this with the state anomaly degree of the source nodes themselves to calculate the percentage contribution of each source node to the voltage problem of the target node. The innovation of this stage lies in the fact that, within the same causal diffusion framework, the evaluation and source tracing outputs are naturally derived, ensuring the consistency of their physical mechanisms.

[0031] The third stage is precise early warning and intelligent decision support, corresponding to step S300 in the instruction manual. The logical endpoint of this stage is to generate decision information that can guide operation. First, based on the voltage quality assessment indicators output by S200, red, yellow, or green early warnings are triggered according to preset multi-level threshold rules. The main disturbance source nodes and their contributions obtained from S200 are deeply integrated into the early warning information to achieve precise root cause alarm. Subsequently, the decision engine matches several candidate control measures from a pre-set control strategy library based on the type of disturbance source and the nature of the voltage problem. For each candidate measure, a power grid response proxy model built based on simplified principles such as sensitivity analysis is invoked to quickly simulate and predict the changes in voltage at key nodes after the implementation of the measure. Finally, a multi-objective scoring function is constructed to comprehensively weigh multiple factors such as the voltage improvement magnitude of the measure, the number of new over-limit nodes that may be triggered, and control costs, to score and compare each candidate measure, generating an auxiliary decision report that includes the preferred recommended measure, expected effects, execution parameters, and risk warnings.

[0032] A method for online voltage quality assessment of distribution networks supporting the integration of renewable energy clusters includes: S100. Extract physical features characterizing the electrical correlation strength and operational status deviation of new energy sources, and fuse them with the original measurement data to reduce dimensionality and generate node dynamic feature vectors for voltage quality assessment. S200. Utilizing a pre-trained causal temporal graph diffusion network, and through a gated spatiotemporal diffusion mechanism, it achieves integrated output of multi-index evaluation of voltage quality and endogenous source attribution of disturbance sources. The S300 system drives root cause localization and hierarchical early warning, and utilizes a reinforcement learning strategy library and a power grid response agent model to perform rapid countermeasure simulation and optimization decision recommendation.

[0033] like Figure 1 As shown, this application illustrates an exemplary method for online voltage quality assessment of a distribution network supporting the access of new energy clusters, specifically including the following steps: Please refer to Figure 2 The diagram illustrates a flowchart of an exemplary online voltage quality assessment method S100 for a distribution network supporting the access of new energy clusters, according to this application. In the method for online voltage quality assessment of distribution networks supporting the access of new energy clusters, the purpose of S100 is to extract physical features that characterize the electrical correlation strength and operational status deviation of new energy by collecting and preprocessing multi-source data of the distribution network, and to fuse and reduce the dimensionality with the original measurement data to generate node dynamic feature vectors for voltage quality assessment, providing high-quality, low-dimensional feature input for subsequent causal diffusion assessment and source tracing.

[0034] The specific steps include: S110. Data Acquisition and Preprocessing: Real-time acquisition of voltage, active power, and reactive power measurement data of multiple nodes in the distribution network, as well as topology status data reflecting network connectivity; topology status data includes information such as node connectivity, branch switch status, and transformer tap position. The measurement data is time-aligned, outliers are removed, and missing values ​​are filled. It is also standardized to eliminate dimensional differences. S120. Online extraction of physical features: Based on topology state data and measurement data, voltage correlation features and operational residual features are calculated for each node. Voltage correlation features are used to characterize the electrical correlation strength between the voltage of the node and one or more preset new energy dominant nodes, and operational residual features are used to characterize the degree of deviation between the current operating state of the node and the expected or typical state. In one possible implementation, voltage correlation characteristics are used to quantify the electrical correlation strength between the voltage of this node and one or more preset high-penetration renewable energy nodes. Specific implementation methods include: The first method is to calculate the electrical association strength based on the electrical distance. The electrical distance is defined as the magnitude of the equivalent impedance between two nodes, and the electrical association strength is taken as the reciprocal of the electrical distance. The closer the distance, the stronger the association. Electrical association strength based on electrical distance The calculation formula is: ; in, For nodes The equivalent impedance amplitude between the new energy-dominant node; The second method is to estimate the voltage sensitivity coefficient based on the simplified power flow equation. By obtaining or estimating the network admittance matrix online, and combining it with the current operating point, the sensitivity of the voltage of this node to the power injection of the new energy head node is calculated. The greater the sensitivity, the greater the impact of the new energy fluctuation on the voltage of this node. Voltage sensitivity coefficient The calculation formula is: ; in, For nodes The voltage amplitude; The voltage amplitude of the new energy-dominant node.

[0035] In one possible implementation, the operational residual feature is used to characterize the degree of deviation of the node's current operating state from its historical normal or expected operating state. Specific calculation methods include: The first method is to calculate the difference between the current active power value of the node and the historical average active power value for the same period, and then divide the difference in active power value by its historical standard deviation to obtain the standardized power residual. Power residual Calculation formula: ; in, This is the historical active power for the same period; This is the historical average active power for the same period; The historical standard deviation; The second method is to calculate the difference between the current voltage amplitude of the node and the regional reference voltage or rated voltage, and then divide the difference by the historical voltage standard deviation or the preset allowable fluctuation range to obtain the standardized voltage residual. Standardized voltage residual Calculation formula: ; in, For nodes The voltage amplitude; This is the regional reference voltage; The historical voltage standard deviation; S130. Feature splicing and dimensionality reduction: The preprocessed original measurement data, voltage correlation features and operating residual features are spliced ​​together to form a wide-dimensional preliminary fusion feature vector. The preliminary fusion feature vector integrates three types of information: direct measurement, topological correlation and state deviation. Since the initial fused feature vectors have a high dimensionality, direct use will affect the efficiency of subsequent calculations and may lead to overfitting. Therefore, dimensionality reduction is required. This invention uses a feature selection method based on correlation for dimensionality reduction. In one possible implementation, the specific steps of dimensionality reduction using the correlation-based feature selection method include: First, calculate the Pearson correlation coefficient between each feature dimension in the preliminary fused feature vector and the target voltage quality index. The Pearson correlation coefficient measures the degree of linear correlation between two variables, and the larger the absolute value, the stronger the correlation with the target. The correlation coefficient calculation formula involves the ratio of the covariance of the feature value and the target value to their respective standard deviations. Pearson correlation between feature dimensions and target metrics The expression is: ; This represents the number of training samples; For feature dimensions In the The values ​​that can be taken on each sample; For feature dimensions The mean across all samples; For the first The target voltage quality index value corresponding to each sample; The mean of the target indicator across all samples; After calculating the correlation coefficients of all feature dimensions, all features are sorted from largest to smallest absolute value; the top Q feature dimensions with the largest absolute values ​​of correlation coefficients are selected to form the final dynamic fusion feature vector of each node. The dimensionality reduction process significantly reduces the feature dimensions and computational complexity while retaining the core information most relevant to voltage quality issues.

[0036] Please refer to Figure 3 It shows a flowchart of S200 in an exemplary online voltage quality assessment method for distribution networks supporting the access of new energy clusters in this application; In the online voltage quality assessment method for distribution networks supporting the access of new energy clusters, the purpose of S200 is to utilize a pre-trained causal time-series graph diffusion network and a gated spatiotemporal diffusion mechanism to achieve integrated output of multi-index voltage quality assessment and endogenous source tracing of disturbance source contribution, thereby simultaneously completing state assessment and root cause tracing under a unified framework.

[0037] The specific steps include: S210. Construction of dynamic causal relationship graph: The distribution network is abstracted as a directed graph; the vertices in the directed graph correspond to all power grid nodes, and the vertex attributes are the dynamic fusion feature vectors of the power grid nodes; the edges in the directed graph represent the causal relationship between nodes, and the edge weights represent the intensity of the causal relationship. Edge construction involves two sub-steps: S2110. Initial physical association construction: Establish directed edges between physically connected nodes, and set the initial weights of the edges. Set to the reciprocal of the electrical distance, where the electrical distance is the magnitude of the equivalent impedance between nodes; For the source node To the target node The initial physical weight of a directed edge is calculated using the following formula: ; As the source node With the target node The equivalent impedance magnitude between; S2120. Dynamic Causal Strength Learning: This method employs an information-theoretic-based transfer entropy algorithm to calculate the dynamic causal influence strength between nodes; specifically, it includes: For any pair of nodes, collect the historical sequence of their dynamic features within a preset sliding time window, and calculate the transfer entropy from the source node to the target node. The formula for calculating transfer entropy is: ; The joint probability or conditional probability is estimated based on historical sequence data within a sliding time window and is used to quantify the likelihood of each state sequence occurring. The target node is The observed value at time; The time lag order of the target node's historical sequence; The time lag order of the historical sequence of the source node; target node exist Time and before A sequence of observations at a historical moment; For the source node Time and before A sequence of observations at a historical moment; Transmission entropy is based on the concept of information theory. It measures the additional contribution of the source node's past information to the prediction of the target node's current value, given that the target node's own past information is known. The greater the contribution, the stronger the causal influence. To obtain a normalized causal strength value, the propagation entropy value is divided by the sum of the propagation entropies from all possible source nodes to the target node, resulting in the normalized causal influence strength value. The formula is expressed as: ; The summation index variable represents the possible summation for the target node. Any influential source node; This represents the total number of nodes. The propagation entropy from the source node to the target node; The sum of the propagation entropy of nodes that affect all possible target nodes; The causal influence strength of all node pairs constitutes an asymmetric causal adjacency matrix. ,in As a structural representation of a dynamic causal relationship graph, each element in the causal adjacency matrix represents the causal influence strength from one node to another, which can characterize the directed causal propagation path. S220. Causal diffusion and state update: Based on the constructed dynamic causal relationship graph, the causal information diffusion process is executed to update the state representation of each node according to the update rules; the causal information diffusion process simulates the propagation of disturbances in the power grid, and the current state of a node is jointly determined by its previous state and information from its causal upstream neighbors. In one possible implementation, the update rule is: For each node, the weighted average state of all its causal upstream neighbors is calculated, with the weights being the normalized values ​​of the causal influence strength of the corresponding edge. Then, this weighted average state is combined with the node's previous state in a weighted manner, with the combination weight controlled by a preset information reception rate parameter. The initial state of the node is set as the dynamic fusion feature vector of that node. The diffusion process stops after iterating for a preset number of steps, and the final state of each node contains global information propagated from multi-hop neighbors along the causal path. For each target node, its position at the 1st... The state after step diffusion Calculated using the following formula: ; For the target node In the The state after the first diffusion step; For the target node The initial state; For information reception rate; For the target node The causal upstream neighbor node, that is, the source node; For the target node The causal upstream neighbor set; As the source node In the The state vector of the step; This is a temporary index variable for summation, used to normalize the denominator, and iterates through the set. , This indicates that for all target nodes Sum of causal strengths; This represents the normalized causal influence strength value.

[0038] S230. Integrated output: Based on the final node state obtained from the diffusion process, the evaluation and tracing results are synchronously output through parallel computing modules; the parallel computing modules include: The voltage quality assessment module contains a set of predefined assessment functions to map the final state of a node to specific voltage quality indicators. The assessment functions are designed based on the actual needs of voltage quality assessment. The voltage quality assessment module includes a voltage deviation index calculation unit, a voltage fluctuation intensity calculation unit, and a voltage limit exceedance probability calculation unit. The voltage deviation index reflects the degree to which the current voltage deviates from the rated value; Voltage fluctuation intensity is calculated based on the temporal variation characteristics implicit in the state, determining short-time voltage fluctuations. Voltage exceedance probability is calculated using a logic function or probability model to determine the risk of voltage exceeding the safe threshold. The endogenous source tracing module calculates the contribution of each potential source node to the voltage problem of the target node based on the causal information flow accumulated during the diffusion process, thereby realizing the root cause location of the voltage quality problem. In one possible implementation, the endogenous tracing module specifically includes the following calculation steps: Calculate the cumulative causal effect strength: calculate the sum of the multi-step causal effect strengths from each potential source node to the target node; From the source node To the target node The intensity of cumulative causal influence The formula is: ; To consider the maximum number of propagation steps, For the causal adjacency matrix Power of 1 The propagation step index represents the number of steps taken in the causal propagation path from the source node to the target node; As the source node To the target node go through The causal strength of step propagation; Calculate the anomaly degree of the source node: calculate the degree of deviation between the current state and the historical normal state of each potential source node. ; ; As the source node Dynamically fused feature vectors; As the source node The feature mean vector under normal operating conditions; Indicates Euclidean distance; This is the weighting coefficient, which can be set according to specific needs; Calculate the unnormalized contribution: Multiply the cumulative causal influence strength by the anomaly degree of the source node to obtain the unnormalized contribution of each potential source node; Calculate contribution percentage: Normalize the unnormalized contribution of all potential source nodes to obtain the contribution percentage of each node, and output the main disturbance source nodes. By using two parallel computing modules, voltage quality assessment results and disturbance source contribution analysis are generated synchronously under the same computing framework, which ensures both computational efficiency and the inherent consistency between the assessment and source tracing results.

[0039] Please refer to Figure 4It shows a flowchart of S300 in an exemplary online voltage quality assessment method for distribution networks supporting the access of new energy clusters in this application; In the online voltage quality assessment method for distribution networks supporting the access of new energy clusters, the S300 aims to drive root cause localization and hierarchical early warning, and utilize a reinforcement learning strategy library and a power grid response agent model to perform rapid countermeasure simulation and optimization decision recommendation, thereby providing operators with accurate alarms and intelligent auxiliary decision support.

[0040] The specific steps include: S310. Root cause localization and hierarchical early warning generation: The early warning engine triggers early warnings based on evaluation indicators and source tracing results, according to preset multi-level threshold rules; the early warning information is deeply integrated with the source tracing results, indicating abnormal nodes, the type and degree of exceeding limits, and listing the main disturbance source nodes and their contribution, so as to achieve accurate root cause alarm. The threshold rule is as follows: A red alert is triggered when the voltage deviation index is greater than 10% or the voltage over-limit probability is greater than 0.95. A yellow warning is triggered when 7% < voltage deviation index ≤ 10% or 0.8 < voltage over-limit probability ≤ 0.95. When the voltage deviation index is less than 7% and the voltage over-limit probability is ≤0.8, no warning will be triggered or a green normal status prompt will be triggered. S320. Candidate control measure matching: The decision engine matches candidate measures from a pre-set control strategy library based on the type of disturbance source and the nature of the voltage problem. The strategy library is a set of scenario-action-effect triplets, which can be constructed through historical data mining or simulation analysis. S330. Rapid effect simulation: For matched candidate measures, the grid response proxy model is called to perform effect simulation. The grid response proxy model is a linear approximation model or a simplified physical model based on sensitivity analysis. The current state and control instructions are input to quickly predict the changes in voltage at key nodes. The simulation formula is as follows: ; This is the sensitivity matrix of node voltage to power injection; The power adjustment vector corresponding to the control measures; This is the predicted voltage change vector; S340. Generate auxiliary decision-making information, comprehensively compare the simulation effects, operating costs and potential risks of each candidate measure, and generate a recommended decision report through multi-objective evaluation, including the preferred measure, expected effects, execution parameters and risk warnings, to support operators in making quick and accurate decisions.

[0041] In one possible implementation, the multi-objective scoring calculation formula is as follows: ; For multi-objective scoring; To add the number of nodes that exceed the limit, To control the cost of measures, , , These are the weighting coefficients; The target voltage change; Rated voltage; This represents the total number of nodes in the distribution network. Maximum allowable cost.

[0042] In the process of generating recommendation decision reports through multi-objective evaluation: After the decision engine matches several candidate control measures from the pre-set control strategy library, it calls the power grid response proxy model to perform rapid effect simulation on each measure and obtains the results. , , As input for multi-objective evaluation; Substitute the above three data points into the preset multi-objective scoring formula to quantify and score each candidate measure. In the multi-objective scoring formula: To evaluate the direct effect of voltage improvement, the greater the improvement, the higher the score; To assess the side effects of a measure, the fewer new problems it causes, the higher the score; To assess the economic viability of a measure, the lower the cost, the higher the score; Calculate the overall score of all candidate measures and sort them from highest to lowest score. The measure with the highest score is automatically identified as the recommended measure; a report is automatically generated, which includes at least the preferred measure, expected results, implementation parameters, and risk warnings.

[0043] Example 2 An online voltage quality assessment system for distribution networks supporting the integration of renewable energy clusters includes: The data acquisition and fusion module is configured to collect and preprocess multi-source data from the distribution network in real time, extract physical features characterizing the electrical correlation strength and operational status deviation of new energy sources, and fuse them with the original measurement data for dimensionality reduction to generate dynamic fusion feature vectors for the nodes; specifically: Real-time acquisition of voltage, active power, and reactive power measurement data from multiple nodes in the distribution network, as well as topology status data reflecting network connectivity. The measurement data is processed by time alignment, outlier removal, missing value imputation, and standardization. Based on topology status data and measurement data, voltage correlation features and operational residual features are extracted online for each node. The voltage correlation features are used to characterize the electrical correlation strength between the voltage of the node and the preset new energy dominant node. The operational residual features are used to characterize the degree of deviation of the current operating state of the node from its historical normal operating conditions. The preprocessed raw measurement data, voltage correlation features, and operational residual features are concatenated, and a correlation-based feature selection method is used for dimensionality reduction to generate a dynamic fusion feature vector for each node.

[0044] The causal diffusion assessment and endogenous source tracing module is configured to construct a dynamic causal relationship graph characterizing the causal influence relationships between nodes. It updates node states using a gated spatiotemporal diffusion mechanism and, based on the final node states, synchronously outputs the voltage quality assessment index for each node, as well as the source tracing results and contribution of the upstream disturbance source nodes causing its voltage quality problems. Specifically: The power distribution network is abstracted as a directed graph, where vertices correspond to power grid nodes, and the vertex attributes are the dynamically fused feature vectors; edges represent the causal relationships between nodes, and edge weights represent the intensity of causal influence. The process of causal information diffusion is executed, and the state representation of each node is iteratively updated according to the preset update rules; Based on the final node state obtained from the diffusion process, the voltage quality assessment index, source tracing results, and contribution of each node are output synchronously through the parallel voltage quality assessment module and endogenous source tracing module. The voltage quality assessment module includes a set of predefined assessment functions to map the final state of a node to a specific voltage quality index. The assessment functions include at least a voltage deviation index calculation function, a voltage fluctuation intensity calculation function, and a voltage limit exceedance probability calculation function. The endogenous source tracing module calculates the contribution of each potential source node to the voltage problem of the target node based on the causal information flow accumulated during the diffusion process, thereby realizing the root cause location of the voltage quality problem.

[0045] The early warning and decision support module is configured to generate a graded early warning signal containing root cause location information based on the voltage quality assessment indicators and the source tracing results; and, based on the source tracing results, match candidate control measures, perform effect simulation via a power grid response proxy model, and generate auxiliary decision information containing recommended control measures and their expected effects; specifically: Based on voltage quality assessment indicators and source tracing contribution results, hierarchical early warning signals containing root cause location information are automatically triggered according to preset multi-level threshold rules. Based on the type of disturbance source and the nature of the voltage problem, candidate control measures are matched from the preset control strategy library; The power grid response proxy model is invoked to perform rapid effect simulation on the matched candidate measures and predict the changes in voltage at key nodes; Based on multi-objective evaluation, the simulation effects, operating costs and potential risks of each candidate measure are comprehensively compared to generate an auxiliary decision-making report that includes recommended measures, expected effects, implementation parameters and risk warnings.

[0046] Example 3 I. Overview of the Verification Scenario: This invention underwent a six-month field verification in a 10kV distribution network in an economic and technological development zone of a coastal province. This area is a typical scenario with high penetration of new energy sources, including 12 distributed photovoltaic power stations (total capacity 28MW) and 2 small wind farms (total capacity 4MW), with a peak penetration rate of 37.7% for new energy sources. The power grid covers a 15-square-kilometer mixed industrial and commercial area and has 35 key monitoring nodes.

[0047] Implementation Configuration Key algorithm parameter settings:

[0048] At 13:30 on August 15, 2025, the voltage of node #23 exceeded the limit (10.78kV > the upper limit of 10.5kV), and the system triggered a red alert.

[0049] Results of disturbance source contribution analysis:

[0050] Verification results

[0051] The system accurately identified the main sources of disturbance as photovoltaic power plants #5 and #8, with a combined contribution of 74%. Based on the source tracing results, the decision-making module recommended prioritizing the adjustment of the inverter power factor of photovoltaic power plant #5, with an expected voltage drop of 0.15kV. After actual execution, the voltage recovered to 10.52kV, verifying the accuracy of the source tracing and the effectiveness of the decision-making.

[0052] Compared with traditional threshold over-limit judgment methods, this invention achieves significant improvements in multiple dimensions: the voltage over-limit detection rate increases from 78.3% to 96.7%, the false alarm rate decreases from 12.5% ​​to 3.2%, and the assessment delay is shortened from 5-10 minutes to 30 seconds. Dispatcher decision-making time is reduced by an average of 86.7%, and the frequency of manual intervention decreases by 75%.

[0053] After implementation, the grid voltage qualification rate increased from 92.3% to 98.7%, the average number of over-limit incidents decreased by 72.4% per month, and the average recovery time was shortened by 72.2%. The renewable energy curtailment rate decreased from 8.7% to 3.2%, improving the renewable energy absorption capacity. The system demonstrated good adaptability and reliability in scenarios with frequent renewable energy fluctuations, providing an effective solution for voltage quality control of distribution networks with a high proportion of renewable energy integration.

[0054] The sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0055] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for online voltage quality assessment of distribution networks supporting the access of renewable energy clusters, characterized in that, include: By collecting and preprocessing multi-source data from the distribution network, physical features characterizing the electrical correlation strength and operational status deviation of new energy sources are extracted and fused with the original measurement data to reduce dimensionality, generating dynamic fusion feature vectors for nodes. A dynamic causal relationship graph representing the causal influence relationship between nodes is constructed. The node state is updated using a gated spatiotemporal diffusion mechanism. Based on the final node state, the voltage quality assessment index of each node and the source tracing results and contribution of the upstream disturbance source node that caused its voltage quality problem are output synchronously. Based on voltage quality assessment indicators and source tracing results, a graded early warning signal containing root cause location information is generated; and based on the source tracing results, candidate control measures are matched, and the effects are simulated through a power grid response proxy model to generate auxiliary decision-making information containing recommended control measures and their expected effects.

2. The method for online voltage quality assessment of distribution networks supporting the access of new energy clusters according to claim 1, characterized in that, The steps for generating the dynamic fusion feature vector of the nodes include: Real-time acquisition of measurement data and topology status data of multiple nodes in the distribution network, and time alignment, anomaly removal, missing value filling and standardization processing of the measurement data; Based on topology state data and measurement data, voltage correlation characteristics and operational residual characteristics are calculated for each node. Voltage correlation characteristics are used to characterize the electrical correlation strength between the node voltage and one or more preset new energy dominant nodes, while operational residual characteristics are used to characterize the degree of deviation between the node's current operating state and the expected or typical state. The preprocessed raw measurement data, voltage correlation features, and operational residual features are spliced ​​together to form a preliminary fusion feature vector. Then, a feature selection method based on correlation is used to reduce the dimensionality of the preliminary fusion feature vector to form a dynamic fusion feature vector.

3. The method for online voltage quality assessment of distribution networks supporting the access of new energy clusters according to claim 2, characterized in that, The method for describing the voltage correlation features includes: The electrical correlation strength is calculated based on the electrical distance, where the electrical distance is the magnitude of the equivalent impedance between two nodes, and the electrical correlation strength is taken as the reciprocal of the electrical distance; or, the voltage sensitivity coefficient is estimated based on the simplified power flow equation, and the sensitivity of the voltage of this node to the power injection of the new energy head node is calculated by obtaining or estimating the network admittance matrix online and combining it with the current operating point.

4. The method for online voltage quality assessment of distribution networks supporting the access of new energy clusters according to claim 2, characterized in that, The calculation method for the operational residual characteristics includes: The operational residual characteristics include power residual and voltage residual. The power residual is obtained by calculating the difference between the current active power value of the node and the historical average active power value for the same period, and dividing the difference between the current active power value of the node and the historical average active power value for the same period by its historical standard deviation. The voltage residual is obtained by calculating the difference between the current voltage amplitude of the node and the regional reference voltage or rated voltage, and dividing the difference between the current voltage amplitude of the node and the regional reference voltage or rated voltage by the historical voltage standard deviation or the preset allowable fluctuation range.

5. The method for online voltage quality assessment of distribution networks supporting the access of new energy clusters according to claim 2, characterized in that, The steps for dimensionality reduction using the correlation-based feature selection method include: First, calculate the Pearson correlation coefficient between each feature dimension in the preliminary fused feature vector and the target voltage quality index. The Pearson correlation coefficient measures the degree of linear correlation between two variables, and the larger the absolute value, the stronger the correlation with the target. The correlation coefficient calculation formula involves the ratio of the covariance of the feature value and the target value to their respective standard deviations. Pearson correlation between feature dimensions and target metrics The expression is: ； This represents the number of training samples; For feature dimensions In the The values ​​that can be taken on each sample; For feature dimensions The mean across all samples; For the first The target voltage quality index value corresponding to each sample; The mean of the target indicator across all samples; After calculating the correlation coefficients of all feature dimensions, all features are sorted from largest to smallest absolute value; the top Q feature dimensions with the largest absolute values ​​of correlation coefficients are selected to form the final dynamic fusion feature vector of each node. The dimensionality reduction process significantly reduces the feature dimensions and computational complexity while retaining the core information most relevant to voltage quality issues.

6. The method for online voltage quality assessment of distribution networks supporting the access of new energy clusters according to claim 1, characterized in that, The synchronous output of voltage quality assessment indicators for each node, as well as the source tracing results and contribution of upstream disturbance source nodes causing voltage quality problems, includes: The power distribution network is abstracted as a directed graph; vertices in the directed graph correspond to power grid nodes, and the vertex attributes are dynamically fused feature vectors; edges in the directed graph represent the causal relationship between nodes, and the edge weights represent the intensity of the causal relationship. Causal diffusion and state update are based on a dynamic causal relationship graph. The process of causal information diffusion is performed, and the state representation of each node is iteratively updated according to the update rules. The current state of a node is determined by its previous state and the information from its upstream causal neighbors. The integrated output, based on the final node state obtained from the diffusion process, synchronously outputs voltage quality assessment indicators and the source tracing results and contribution through a parallel voltage quality assessment module and an endogenous source tracing module.

7. The method for online voltage quality assessment of a distribution network supporting the access of new energy clusters according to claim 6, characterized in that, In the construction of the dynamic causal relationship graph, the edge construction steps include: Initial physical association construction involves establishing directed edges between physically connected nodes and setting initial weights for these edges, which are the reciprocal of the electrical distance between the nodes. Dynamic causal strength learning employs an information-theoretic-based transfer entropy algorithm to calculate the dynamic causal influence strength between nodes. For any pair of nodes, the historical sequence of their dynamic features within a preset sliding time window is collected, and the transfer entropy from the source node to the target node is calculated. The transfer entropy is normalized to obtain the normalized causal influence strength value, which is used as the weight of the corresponding directed edge. The causal influence strength of all node pairs constitutes an asymmetric causal adjacency matrix, which is used to represent the weight of the edges in the directed graph.

8. The method for online voltage quality assessment of a distribution network supporting the access of new energy clusters according to claim 6, characterized in that, The update rule is as follows: For each node, calculate the weighted average state of all its causal upstream neighbors, with the weights being the normalized values ​​of the causal influence strength of the corresponding edges; then, combine this weighted average state with the node's previous state in a weighted manner, with the combination weights controlled by a preset information reception rate parameter; the initial state of the node is set as the node's dynamic fusion feature vector. The diffusion process stops after iterating for a preset number of steps. The final state of each node contains global information that has been propagated from its multi-hop neighbors along causal paths. For each target node, its position at the 1st... The state after step diffusion Calculated using the following formula: ； For the target node In the The state after the first diffusion step; For the target node The initial state; For information reception rate; For the target node The causal upstream neighbor node, that is, the source node; For the target node The causal upstream neighbor set; As the source node In the The state vector of the step; This is a temporary index variable for summation, used to normalize the denominator, and iterates through the set. , This indicates that for all target nodes Sum of causal strengths; This represents the normalized causal influence strength value.

9. The method for online voltage quality assessment of a distribution network supporting the access of new energy clusters according to claim 6, characterized in that, The voltage quality assessment module includes a set of predefined assessment functions to map the final state of a node to a specific voltage quality index. The assessment functions are designed based on the actual needs of voltage quality assessment. The voltage quality assessment module includes a voltage deviation index calculation unit, a voltage fluctuation intensity calculation unit, and a voltage limit exceedance probability calculation unit. The voltage deviation index reflects the degree to which the current voltage deviates from the rated value; Voltage fluctuation intensity is calculated based on the temporal variation characteristics implicit in the state, determining short-time voltage fluctuations. Voltage exceedance probability is calculated using a logic function or probability model to determine the risk of voltage exceeding a safe threshold.

10. The method for online voltage quality assessment of a distribution network supporting the access of new energy clusters according to claim 6, characterized in that, The endogenous source tracing module calculates the contribution of each potential source node to the voltage problem of the target node based on the causal information flow accumulated during the diffusion process, thereby realizing the root cause location of the voltage quality problem. The endogenous source tracing module specifically includes the following calculation steps: Calculate the cumulative causal effect strength: calculate the sum of the multi-step causal effect strengths from each potential source node to the target node; From the source node To the target node The intensity of cumulative causal influence The formula is: ； To consider the maximum number of propagation steps, For the causal adjacency matrix Power of 1 The propagation step index represents the number of steps taken in the causal propagation path from the source node to the target node; As the source node To the target node go through The causal strength of step propagation; Calculate the anomaly degree of the source node: calculate the degree of deviation between the current state and the historical normal state of each potential source node. ; ； As the source node Dynamically fused feature vectors; As the source node The feature mean vector under normal operating conditions; Indicates Euclidean distance; This is the weighting coefficient, which can be set according to specific needs; Calculate the unnormalized contribution: Multiply the cumulative causal influence strength by the anomaly degree of the source node to obtain the unnormalized contribution of each potential source node; Calculate the contribution percentage: Normalize the unnormalized contribution of all potential source nodes to obtain the contribution percentage of each node, and output the main perturbation source nodes.

11. The method for online voltage quality assessment of a distribution network supporting the access of new energy clusters according to claim 1, characterized in that, The generation of auxiliary decision-making information, which includes recommended control measures and their expected effects, includes: The root cause localization and hierarchical early warning generation is based on voltage quality assessment indicators and source tracing results. It triggers hierarchical early warnings according to preset multi-level threshold rules. The early warning information indicates the abnormal node, the type and degree of exceeding the limit, and lists the main disturbance source nodes and their contribution. Candidate control measures are matched by selecting candidate measures from a pre-set control strategy library based on the type of disturbance source and the nature of the voltage problem. Rapid effect simulation: Call the power grid response proxy model to simulate the effects of candidate measures and predict changes in voltage at key nodes; Generate auxiliary decision-making information, comprehensively compare the simulation effects, operating costs and potential risks of each candidate measure, and generate a recommended decision report through multi-objective evaluation.

12. An online voltage quality assessment system for distribution networks supporting the access of new energy clusters, characterized in that, include: The data acquisition and fusion module is configured to collect and preprocess multi-source data of the distribution network in real time, extract physical features that characterize the electrical correlation strength and operational status deviation of new energy sources, and fuse them with the original measurement data to reduce dimensionality and generate dynamic fusion feature vectors for nodes. The causal diffusion assessment and endogenous source tracing module is configured to construct a dynamic causal relationship graph that characterizes the causal influence relationship between nodes. It uses a gated spatiotemporal diffusion mechanism to update the node state and, based on the final node state, synchronously outputs the voltage quality assessment index of each node as well as the source tracing results and contribution of the upstream disturbance source node that caused its voltage quality problem. The early warning and decision support module is configured to generate a graded early warning signal containing root cause location information based on voltage quality assessment indicators and source tracing results; and, based on the source tracing results, match candidate control measures, perform effect simulation through a power grid response proxy model, and generate auxiliary decision information containing recommended control measures and their expected effects.

13. The online voltage quality assessment system for distribution networks supporting the access of new energy clusters according to claim 12, characterized in that, The voltage quality assessment module in the causal diffusion assessment and endogenous source tracing module contains a set of predefined assessment functions to map the final state of a node to a specific voltage quality index. The assessment functions include a voltage deviation index calculation function, a voltage fluctuation intensity calculation function, and a voltage limit exceedance probability calculation function.

Images