A power distribution network operation risk control method based on bayesian optimization

Abstract

The application discloses a power distribution network operation risk control method based on Bayesian optimization, comprising the following steps: S1, a node-edge double-space coupled dynamic graph is established by using multi-source operation data of the power distribution network; S2, based on the dynamic graph, a space-time embedding vector is generated through a self-supervised task of multi-level space-time interaction of nodes and node neighborhoods; S3, a risk-sensitive network with multi-scale dynamic clustering is constructed to predict a risk propagation trajectory; S4, a cross-period continuous dynamic risk evolution link is formed by using the risk propagation trajectory, and a power distribution network risk control sub-domain is divided; S5, a joint optimization state space is constructed by evaluating the dynamic evolution of the sub-domain risk; S6, a distributed bidirectional information collaboration link between neighborhoods is constructed; S7, the convergence of real-time monitoring and decision-making is monitored, and path switching is performed. The application improves the real-time performance and accuracy of the power distribution network risk control, and ensures the safe and stable operation of the power distribution network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system automation, and in particular to a distribution network operation risk control method based on Bayesian optimization. BACKGROUND

[0002] With the continuous expansion of new energy access, the continuous increase of power load and the increasing complexity of distribution network structure, the operation risk management of distribution network is facing significant challenges. The traditional distribution network risk control method is mainly based on expert experience and static model, and the risk state is evaluated and controlled through preset rules or fixed parameters. Typical technical paths include fault tree analysis, risk matrix evaluation and fuzzy reasoning methods. However, the parameters of these methods usually need to rely on expert manual setting, and lack the ability of adaptive adjustment to real-time working conditions. In the face of complex operating environments such as rapid changes of power grid load, fluctuations of distributed power output, sudden extreme weather events, etc., the risk assessment accuracy and real-time performance are obviously insufficient, and it is difficult to respond and make quick decisions efficiently.

[0003] In recent years, with the development of artificial intelligence and machine learning technology, some scholars have proposed using deep learning algorithms such as recurrent neural networks and graph convolution networks to automatically extract time series and spatial features of distribution network, and to realize the identification and prediction of operation risk. However, due to the dependence of conventional supervised learning, these methods usually need a large amount of labeled data for training, and in actual application, it is difficult to obtain a large amount of labeled data, and the generalization ability of the trained model is insufficient, and the adaptability and generalization of new abnormal working conditions are obviously limited.

[0004] In addition, although the existing reinforcement learning algorithm can realize the dynamic optimization of decision strategy, the initial convergence speed is usually slow, and it cannot effectively deal with the rapidly evolving sudden risks in the distribution network. Especially in the context of complex risk multi-node and cross-regional collaborative control, the traditional reinforcement learning method cannot effectively realize the information interaction and strategy coordination between nodes or sub-domains, resulting in the difficulty of the overall control strategy of the system to quickly and effectively converge dynamically, and there are obvious deficiencies in the application.

[0005] Therefore, how to provide a distribution network operation risk control method based on Bayesian optimization is a problem that those skilled in the art need to solve. SUMMARY

[0006] One object of the present application is to provide a distribution network operation risk control method based on Bayesian optimization. The present application effectively improves the real-time performance and accuracy of the distribution network operation risk assessment, and enhances the adaptive ability of the complex dynamic operating conditions and the risk control robustness of the overall system.

[0007] According to the distribution network operation risk control method based on Bayesian optimization of the embodiment of the present application, the following steps are included:

[0008] S1, establish a node-edge double-space coupled dynamic graph using multi-source operation data of the power distribution network;

[0009] S2, based on the dynamic graph, through a self-supervised task of multi-level space-time interaction of nodes and their neighborhoods, dynamically adjust the associated feature weight of the node in different operation states, and generate a space-time embedding vector;

[0010] S3, based on the space-time embedding vector, construct a risk-sensitive network with multi-scale dynamic clustering, automatically extract the risk diffusion path in the operation state and predict the risk propagation trajectory;

[0011] S4, using the risk propagation trajectory, automatically form a cross-period continuous dynamic risk evolution link in the power distribution network, and dynamically divide the power distribution network risk control sub-domain according to the space-time embedding change trend on the link;

[0012] S5, for the dynamically formed risk control sub-domain, evaluate the operation risk dynamic evolution process in the sub-domain, construct a joint optimization state space of risk state-decision history-space-time feature, and update the risk control instruction sequence according to the evaluation result;

[0013] S6, during the dynamic execution of the risk control instruction sequence, construct a distributed bidirectional information collaboration link between neighborhoods, which is used for risk state synchronization update between risk control sub-domains and self-adjustment of cross-regional risk control instructions;

[0014] S7, monitor the decision dynamic convergence of each risk control sub-domain, and automatically start the pre-constructed redundant control path when the local risk convergence state is predicted to degrade, to realize seamless switching of the risk decision path.

[0015] Optionally, the S1 specifically comprises:

[0016] S11, collect the power flow data, voltage data and load data of all nodes of the power distribution network, calculate the data change rate and trend of each node, and construct a node feature dynamic hierarchical system using the node data change rate;

[0017] S12, determine the dynamic association relationship between nodes according to the node feature dynamic hierarchical system, calculate the data association strength and association direction between nodes, and form an updated node association strength matrix;

[0018] S13, according to the node association strength matrix, dynamically select the power distribution lines directly connected with each node, calculate the line load level and fluctuation trend of each line, and establish a dynamic association mapping between the nodes and the corresponding lines of the power distribution network;

[0019] S14, constructing a node-edge bidirectional state coupling network expressing the electrical connection relationship between nodes and the data dynamic change trend at the same time according to the dynamic association mapping, and updating the coupling network topology;

[0020] S15, detecting the topology structure change of the node-edge bidirectional state coupling network, and adaptively updating the dynamic coupling weight between the node and the corresponding line according to the topology structure change frequency, the number of changed nodes and the intensity of the coupling state;

[0021] S16, converting the node-edge bidirectional state coupling network into a dynamic graph according to the updated dynamic coupling weight and topology structure.

[0022] Optionally, the S2 specifically comprises:

[0023] S21, identifying the adjacent nodes with data interaction frequency higher than a preset threshold in the node neighborhood according to the topology structure of the dynamic graph and the node feature data change, and determining the direction of data interaction between nodes to form a node neighborhood interaction relationship graph;

[0024] S22, dividing the direct neighborhood, the secondary neighborhood and the edge neighborhood according to the node neighborhood interaction relationship graph to form a multi-level structure of the node neighborhood, and determining the feature interaction priority of different neighborhood levels;

[0025] S23, determining a feature sensitivity parameter based on the amplitude and frequency of the change of the node feature data, and screening the feature data in the different levels of neighborhood of the node according to the feature sensitivity parameter;

[0026] S24, calculating the data interaction sensitivity coefficient between the node and the nodes in each level of neighborhood according to the neighborhood feature data, and dynamically adjusting the feature association weight of the nodes in each level of neighborhood based on the sensitivity coefficient;

[0027] S25, determining the optimal feature aggregation mode according to the dynamically adjusted feature association weight, which is used to aggregate the screened feature information of the node itself and the neighborhood;

[0028] S26, generating a space-time embedding vector for each node by fusing the node feature and the multi-level neighborhood feature data using the feature aggregation mode.

[0029] Optionally, the S3 specifically comprises:

[0030] S31, constructing a multi-dimensional dynamic perception network of feature interaction between nodes based on the space-time embedding vector of the node, and identifying the implicit feature interaction relationship between nodes for each node;

[0031] S32, determining a multi-scale clustering center according to the implicit feature interaction relationship, and generating a dynamic clustering topology between nodes by each scale clustering center;

[0032] S33, detecting a boundary region of feature change abnormality between the node sets according to the change trend of the node embedding vectors in the dynamic clustering topology, and forming a risk sensitive region with the boundary region as a core;

[0033] S34, tracking a node sequence with the most dramatic change and fluctuation of the node embedding vectors in the risk sensitive region, and extracting a dominant path of risk diffusion;

[0034] S35, analyzing the time sequence change direction and change strength of the node embedding vectors of each node along the dominant path of risk diffusion, and forming a continuous risk diffusion state chain;

[0035] S36, dynamically evaluating the state transition between each node in the risk diffusion state chain, and generating a multi-step continuous evolution mode of risk diffusion;

[0036] S37, deducing the risk diffusion development direction of the risk diffusion path based on the multi-step continuous evolution mode of risk diffusion, and predicting the risk propagation trajectory under the operation state of the power distribution network.

[0037] Optionally, the S31 specifically comprises:

[0038] S311, calculating the change rate and direction of the node feature vector according to the node spatio-temporal embedding vector, and generating a multi-dimensional feature space of the node feature change trend for each node;

[0039] S312, mapping each node feature in the multi-dimensional feature space, and determining the spatial position relationship between the node feature change trend and other node features for each node;

[0040] S313, identifying the feature conduction path between nodes according to the spatial position relationship of the node feature change trend, and labeling the conduction direction on the path;

[0041] S314, calculating the contribution coefficient of each node in the feature conduction path to the overall feature conduction strength of the path for each node, and determining the key interaction node with the contribution coefficient exceeding a preset threshold;

[0042] S315, monitoring the synchronicity of the feature change trend between the key interaction nodes, calculating a synchronicity index, and using the index to represent the implicit feature interaction strength between the nodes;

[0043] S316, dividing the implicit feature interaction relationship between the nodes into three levels of high strength, medium strength and low strength according to the implicit feature interaction strength, and forming an implicit feature interaction grading system.

[0044] Optionally, the S4 specifically comprises:

[0045] S41, automatically extract the risk transmission coupling path between nodes according to the change trend of the node space-time embedding vector in the risk transmission path, and calculate the feature change trend vector of each node in the path;

[0046] S42, calculate the change rate and acceleration of the feature change trend vector of each node in the risk transmission coupling path, and determine the node with acceleration exceeding the preset threshold as the risk state turning point node;

[0047] S43, divide the risk transmission coupling path into multiple independent risk transmission links according to the risk state turning point node, and determine the starting node, ending node and spatial boundary of each link;

[0048] S44, track the change of the space-time embedding vector of all nodes in each risk transmission link, establish a node space-time feature difference matrix for each link, and calculate the feature difference index in the matrix;

[0049] S45, analyze the convergence and difference between any two adjacent risk transmission links according to the feature difference index, and determine the correlation level between the links;

[0050] S46, adaptively aggregate the risk transmission links according to the correlation level between the links, and dynamically divide to form multiple risk control subdomains;

[0051] S47, monitor the feature change trend and risk transmission intensity of the nodes in the risk control subdomain boundary, and dynamically adjust the subdomain boundary for each subdomain.

[0052] Optionally, the S5 specifically includes:

[0053] S51, calculate the risk state index change trend of each node in the risk control subdomain, identify the node set with consistent risk state index change trend for each node, and determine the spatial distribution characteristics of the risk evolution of the nodes in the subdomain;

[0054] S52, track the historical risk control instruction sequence of each node in the spatial distribution characteristics, determine the front and rear dependency relationship between the historical risk control instructions for each node, and form a historical decision interaction network between nodes;

[0055] S53, calculate the dynamic influence factor of the historical risk control instruction on the current node risk state for each node according to the historical decision interaction network between nodes, and determine the specific action degree of the historical decision on the current risk state;

[0056] S54, extract the node risk state index, historical decision dynamic influence factor and node space-time embedding vector, build a risk state-decision history-space-time feature joint state data set for each node, and fuse to generate a three-dimensional joint optimization state space;

[0057] S55, according to the three-dimensional joint optimization state space, the sensitivity coefficient of the risk state change of each node to the risk control instruction update demand is calculated node by node, and the priority of the risk control instruction update of each node is dynamically determined;

[0058] S56, according to the node update priority, the update interval time of the risk control instruction sequence in the sub-domain is determined, and the execution order of each node risk control instruction is dynamically adjusted;

[0059] S57, the fluctuation amplitude and frequency of the overall risk state of the risk control sub-domain are monitored, and the updated risk control instruction sequence is automatically triggered to execute the risk control instruction node by node.

[0060] Optionally, the S6 specifically comprises:

[0061] S61, when the risk control instruction sequence is executed in the risk control sub-domain, the sub-domain boundary node monitors the change amount of the spatio-temporal embedding vector and the risk score after the relative last instruction execution, and when any threshold change amount reaches the preset asynchronous trigger condition, the boundary node automatically generates a boundary node feature snapshot containing the node spatio-temporal embedding vector, the risk quantization value, the time stamp and the node unique identification;

[0062] S62, based on the self-organizing grid topology, the boundary node feature snapshot is packaged as a broadcast data packet with time slot identification, and is periodically broadcast to the sub-domain directly physically adjacent to the sub-domain through the adjacent link in a multi-path parallel manner;

[0063] S63, after receiving the boundary node feature snapshot broadcast by the adjacent sub-domain, each sub-domain parses the spatio-temporal embedding vector and the risk quantization value of the boundary node in the snapshot, and sequentially sorts the time stamp to construct a cross-period boundary coupling graph containing the boundary node information of the current period and the previous period, wherein the nodes of the boundary coupling graph represent the boundary nodes, and the edges represent the risk association between the nodes in the adjacent period;

[0064] S64, for each edge in the cross-period boundary coupling graph, the difference degree of the spatio-temporal embedding vector and the fluctuation rate of the risk quantization value of the boundary node in the adjacent period are calculated edge by edge, and the comprehensive index of the difference degree and the fluctuation rate is compared with the preset mutation threshold, and the risk fluctuation mutation edge meeting the condition higher than the threshold is automatically identified, and a dynamic negotiation trigger signal is generated taking the mutation edge as the trigger source;

[0065] S65, the dynamic negotiation trigger signal is broadcast to all boundary nodes in the sub-domain through a plurality of backup communication channels in a load balancing manner, and the boundary nodes perform weighted fusion calculation according to the local spatio-temporal embedding vector and the neighborhood snapshot information after receiving the trigger signal, and adjust the risk quantization value calibration factor of the boundary node;

[0066] S66, each sub-domain generates a micro-transition risk control instruction mapping sequence containing micro-transition incremental values and time sequence order based on the calibrated boundary node risk rating value and multi-sub-domain interaction feedback information, and each micro-transition incremental value in the mapping sequence is combined with the risk gradient difference of the adjacent sub-domain to drive the evolutionary self-coordination of the cross-sub-domain risk control instruction;

[0067] S67, each sub-domain issues the updated risk control instruction to each node in the sub-domain according to the mapping sequence, and continuously monitors the changes of the spatio-temporal embedding vector of the boundary node and the calibrated risk quantization value, and when the spatio-temporal embedding vector change rate or risk quantization value fluctuation rate of any boundary node again meets the asynchronous triggering condition, a new boundary node feature snapshot is automatically generated, forming an adaptive boundary suppression closed loop and iteratively updating the state of the two-way information collaboration link.

[0068] Optionally, the S7 specifically comprises:

[0069] S71, for each risk control sub-domain, asynchronously capture the spatio-temporal embedding vector of all nodes in the sub-domain and the corresponding risk control instruction execution feedback state, and generate a distributed adaptive error spectrum by accumulating the state transition of the nodes, wherein the distributed adaptive error spectrum represents the error distribution characteristics of the node risk state in the time domain and the space domain;

[0070] S72, parallel comparison of the distributed adaptive error spectrum and the historical multi-scale error spectrum is performed, and the spectrum resonance detection mechanism is used to compare the spectrum energy level distribution, when the cross-period error energy level aggregation appears in the error spectrum and the spectrum resonance coefficient exceeds the preset stable threshold, a high-energy level error node group is identified, so as to determine that the local risk convergence state has degenerated;

[0071] S73, after determining that the local risk convergence state has degenerated, trigger the internal path regeneration mechanism of the sub-domain, and autonomously query a pre-stored redundant path candidate set corresponding to the high-energy level error node group in the redundant control path library, wherein the redundant control path library has been pre-stored according to the node spatio-temporal embedding matching degree, path redundancy, and fault tolerance level;

[0072] S74, for the candidate redundant path set, the spatio-temporal embedding vector spectrum compatibility index of each node in the path, the link oscillation amplitude index of the path relay node, and the node load jump tolerance index are calculated in turn, and a parallel adaptive priority calculation strategy is comprehensively adopted to select the redundant control path with the highest adaptive priority as the switching target;

[0073] S75, based on the start node, relay node and termination node information of the selected redundant control path, a token type command sequence is generated according to the self-organizing token ring protocol, the token type command sequence contains the control instruction of each node in the redundant path and the token passing rule, and the current control time slot is seamlessly switched to the token type command sequence;

[0074] S76, after switching to the redundant control path, the asynchronous hedging mechanism is started in parallel in the sub-domain, the convergence rate difference and risk control resonance strength of each node on the redundant path are collected, when the resonance strength is lower than the preset resonance stability threshold, the fast rollback mechanism is triggered and the step S73 is re-entered to retrieve and switch to the next redundant control path;

[0075] S77, when the switched redundant control path reaches the convergence and risk resonance balance state, the adaptive priority weight of the corresponding path in the redundant control path library is updated through the dynamic memory interaction mechanism, and the asynchronous capture and monitoring of the conventional distributed adaptive error spectrum in the sub-domain are restored.

[0076] The beneficial effects of the present application are:

[0077] (1) The present application realizes high-precision real-time mapping between nodes and lines of the power distribution network by constructing a node-edge double-space coupled dynamic graph and generating a multi-dimensional feature space by real-time calculation of node feature change trend, effectively improving the accuracy and dynamic response speed of risk assessment, and enhancing the adaptive ability to complex operating conditions.

[0078] (2) The present application automatically identifies the risk diffusion dominant path through dynamic clustering and implicit feature interaction perception network, and divides the risk control sub-domain in real time, significantly improving the accuracy of risk state prediction and the efficiency of risk propagation path identification, and showing better real-time adaptability in the risk burst and propagation scene of the power distribution network.

[0079] (3) In the aspect of cross-regional risk collaborative control, the present application effectively solves the problem of lack of real-time self-coordination ability of traditional methods by constructing inter-neighborhood distributed bidirectional information collaboration link and asynchronous event triggering mechanism, breaks through the bottleneck of difficult fast synchronization update of risk control instructions in complex node interaction scene of the prior art, realizes efficient collaboration and adaptive optimization of risk control instructions, and thus effectively improves the overall risk control stability of the power distribution network.

[0080] (4) The present application establishes a redundant control path library and a self-organizing token ring mechanism, monitors and quickly determines the degradation or failure of risk convergence state in real time, automatically starts the redundant path to realize seamless switching, significantly reduces the delay and error of risk decision path switching, and effectively guarantees the real-time and robustness of the risk decision of the power distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0081] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0082] Figure 1A node-edge dual-space coupled dynamic graph structure schematic diagram of a power distribution network operation risk control method based on Bayesian optimization is provided for the present application.

[0083] Figure 2 A risk propagation trajectory and risk control subdomain division schematic diagram of a power distribution network operation risk control method based on Bayesian optimization is provided for the present application.

[0084] Figure 3 A neighborhood inter-distributed bidirectional information collaboration link structure schematic diagram of a power distribution network operation risk control method based on Bayesian optimization is provided for the present application. DETAILED DESCRIPTION

[0085] The present application will now be further described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams and only illustrate the basic structure of the present application in a schematic manner, and thus only show the components related to the present application.

[0086] REFERENCE Figures 1-3 A power distribution network operation risk control method based on Bayesian optimization, comprising the following steps:

[0087] S1, a node-edge dual-space coupled dynamic graph is established using power distribution network multi-source operation data;

[0088] S2, based on the dynamic graph, through the self-supervised task of node and node neighborhood multi-level space-time interaction, the associated feature weight of the node in different operation states is dynamically adjusted node by node to generate a space-time embedding vector;

[0089] S3, based on the space-time embedding vector, a risk-sensitive network with multi-scale dynamic clustering is constructed, the risk diffusion path in the operation state is automatically extracted, and the risk propagation trajectory is predicted;

[0090] S4, using the risk propagation trajectory, a cross-period continuous dynamic risk evolution link is automatically formed in the power distribution network, and the power distribution network risk control subdomain is dynamically divided according to the space-time embedding change trend on the link;

[0091] S5, for the dynamically formed risk control subdomain, the operation risk dynamic evolution process in the subdomain is evaluated, a risk state-decision history-space-time feature joint optimization state space is constructed, and the risk control instruction sequence is updated according to the evaluation result;

[0092] S6, during the dynamic execution process of the risk control instruction sequence, a neighborhood inter-distributed bidirectional information collaboration link is constructed, which is used for risk state synchronous updating between risk control subdomains and self-coordinated adjustment of cross-regional risk control instructions;

[0093] S7, monitor the decision dynamic convergence of each risk control sub-domain, and automatically start the pre-constructed redundant control path when the local risk convergence state is predicted to degrade, to realize seamless switching of the risk decision path.

[0094] By constructing a node-edge two-space coupled dynamic graph, fine dynamic modeling of the multi-source data of the power distribution network is realized, spatiotemporal embedding vectors are generated by using multi-level self-supervised tasks, and the accuracy of risk state assessment is improved; the risk propagation path is accurately predicted by combining a multi-scale dynamic clustering risk-sensitive network, and the risk control sub-domain is dynamically divided, thereby improving the real-time and pertinence of risk management; by constructing a risk state-decision history-spatiotemporal feature joint optimization state space and a distributed two-way information collaboration link between neighborhoods, real-time self-coordinated adjustment of cross-regional risk control instructions is effectively realized; at the same time, a redundant control path is established to ensure seamless switching of the risk decision path, thereby improving the robustness and reliability of the overall system.

[0095] In the embodiment, S1 specifically includes:

[0096] S11, collecting power flow data, voltage data and load data of all nodes of the power distribution network, calculating data change rate and change trend for each node, and constructing a node feature dynamic hierarchical system by using the node data change rate;

[0097] S12, determining the dynamic correlation relationship between nodes according to the node feature dynamic hierarchical system, calculating the data correlation strength and correlation direction between nodes, and forming an updated node correlation strength matrix;

[0098] S13, dynamically selecting the power distribution lines directly connected to each node according to the node correlation strength matrix, calculating the line load level and fluctuation trend for each line, and establishing a dynamic correlation mapping between the nodes and the corresponding lines of the power distribution network;

[0099] S14, constructing a node-edge two-way state coupled network that simultaneously expresses the electrical connection relationship between nodes and the data dynamic change trend according to the dynamic correlation mapping, and updating the coupled network topology;

[0100] S15, detecting the topology structure change of the node-edge two-way state coupled network, and adaptively updating the dynamic coupling weight between the nodes and the corresponding lines according to the topology structure change frequency, the number of changed nodes and the severity of the coupled state;

[0101] S16, converting the node-edge two-way state coupled network into a dynamic graph according to the updated dynamic coupling weight and topology structure.

[0102] By collecting real-time power flow, voltage and load data, calculating the data change rate and trend of nodes, establishing a dynamic hierarchical system of node characteristics, determining the data correlation strength and direction between nodes in real time, a dynamic correlation strength matrix and a node-edge bidirectional state coupling network are constructed to effectively realize real-time and refined mapping of node and line states; By dynamically detecting the change of network topology structure and adaptively updating the coupling weight between nodes and lines, the real-time and accuracy of the dynamic graph model of the distribution network are improved, and the response ability and accuracy of the distribution network operation risk assessment and control are further optimized.

[0103] In the embodiment, the S2 specifically comprises:

[0104] S21, according to the topological structure of the dynamic graph and the change of the node characteristic data, identifying the adjacent nodes in the node neighborhood whose data interaction frequency is higher than a preset threshold, and determining the direction of data interaction between nodes to form a node neighborhood interaction relationship graph;

[0105] S22, according to the node neighborhood interaction relationship graph, dividing the direct neighborhood, the secondary neighborhood and the edge neighborhood to form a multi-level structure of the node neighborhood, and determining the characteristic interaction priority of different neighborhood levels;

[0106] S23, determining the characteristic sensitivity parameter based on the amplitude and frequency of the change of the node characteristic data, and filtering the characteristic data in the node neighborhood of different levels according to the characteristic sensitivity parameter;

[0107] S24, calculating the data interaction sensitivity coefficient between the node and the nodes in each level of neighborhood according to the neighborhood characteristic data, and dynamically adjusting the characteristic correlation weight of the nodes in each level of neighborhood based on the sensitivity coefficient;

[0108] S25, determining the optimal characteristic aggregation mode according to the dynamically adjusted characteristic correlation weight, which is used to aggregate the filtered characteristic information of the node itself and the neighborhood;

[0109] S26, using the characteristic aggregation mode to fuse the node characteristic and the multi-level neighborhood characteristic data to generate a spatio-temporal embedding vector for each node.

[0110] By identifying the adjacent nodes with high data interaction frequency in the node neighborhood in real time, forming a node neighborhood interaction relationship graph, and determining the real-time optimal characteristic aggregation mode according to the data interaction sensitivity parameter and the characteristic correlation weight between the node and the neighborhood, and then fusing the node characteristic and the neighborhood characteristic data to generate a spatio-temporal embedding vector, the adaptive dynamic adjustment of the node characteristic is realized, the accuracy and real-time of the distribution network risk state assessment are effectively improved, and the adaptability of the system to complex dynamic operation environment is enhanced.

[0111] In the embodiment, the S3 specifically comprises:

[0112] S31, constructing a multi-dimensional dynamic perception network of feature interaction between nodes based on the spatio-temporal embedding vectors of the nodes, and identifying the implicit feature interaction relationship between nodes node by node;

[0113] S32, determining multi-scale clustering centers according to the implicit feature interaction relationship, and generating a dynamic clustering topology between nodes from the clustering centers of each scale;

[0114] S33, detecting a boundary region of feature change anomaly between node sets according to the change trend of the node embedding vector in the dynamic clustering topology, and forming a risk sensitive region with the boundary region as the core;

[0115] S34, tracking a node sequence with the most dramatic change fluctuation of the node embedding vector in the risk sensitive region, and extracting a dominant path of risk diffusion;

[0116] S35, analyzing the time sequence change direction and change strength of the node embedding vector node by node along the dominant path of risk diffusion, and forming a continuous risk diffusion state chain;

[0117] S36, dynamically evaluating the state transition between nodes in the risk diffusion state chain, and generating a multi-step continuous evolution mode of risk diffusion;

[0118] S37, based on the multi-step continuous evolution mode of risk diffusion, deducing the development direction of the risk diffusion path of risk diffusion, and predicting the risk propagation trajectory under the operation state of the power distribution network.

[0119] By constructing a multi-dimensional dynamic perception network of feature interaction between nodes, the real-time identification of the implicit feature interaction relationship between nodes is realized, and the multi-scale clustering topology structure is determined, which further accurately divides the risk sensitive region. By tracking the node sequence with dramatic change fluctuation of the node embedding vector to extract the dominant path of risk diffusion, and analyzing the node state transition features in real time to generate a multi-step continuous evolution mode of risk diffusion, the subsequent development direction of the risk diffusion path can be accurately deduced, the accuracy and real-time performance of the risk propagation trajectory prediction of the power distribution network are effectively improved, and the perception ability of the dynamic evolution of the risk is enhanced.

[0120] In the embodiment, the S31 specifically comprises:

[0121] S311, calculating the change rate and direction of the node feature vector according to the spatio-temporal embedding vector of the node, and generating a multi-dimensional feature space of the node feature change trend node by node;

[0122] S312, mapping each node feature in the multi-dimensional feature space, and determining the spatial position relationship between the node feature and the change trend of other node features node by node;

[0123] S313, identify the feature conduction path between nodes according to the spatial position relationship of the feature change trend of the nodes, and mark the conduction direction on the path;

[0124] S314, calculate the contribution coefficient of each node in the feature conduction path to the overall feature conduction strength of the path, and determine the key interaction node whose contribution coefficient exceeds the preset threshold;

[0125] S315, monitor the synchronization of the feature change trend between the key interaction nodes, calculate the synchronization index, and use the index to represent the implicit feature interaction strength between the nodes;

[0126] S316, according to the implicit feature interaction strength, divide the implicit feature interaction relationship between the nodes into three levels of high strength, medium strength and low strength, and form an implicit feature interaction classification system.

[0127] By calculating the change rate and direction of the node feature vector, a multi-dimensional feature space is generated, and the spatial position relationship of the feature change trend between the nodes is determined in real time, the feature conduction path between the nodes is identified and marked. Further, the contribution coefficient of each node to the overall feature conduction strength of the path is calculated, the key interaction node is determined, and the synchronization of the feature change trend thereof is monitored, so as to quantitatively determine the implicit feature interaction strength between the nodes in real time, and establish an explicit implicit feature interaction classification system. Therefore, the implicit feature interaction relationship between the nodes can be more accurately described, and the fineness and reliability of the risk feature propagation analysis are effectively improved.

[0128] In this embodiment, the S4 specifically includes:

[0129] S41, according to the change trend of the node space-time embedding vector in the risk propagation trajectory, automatically extract the continuous risk conduction coupling path between the nodes across the time period, and calculate the feature change trend vector of each node in the path;

[0130] S42, calculate the change rate and change acceleration of the feature change trend vector of each node in the risk conduction coupling path, and determine the node whose acceleration exceeds the preset threshold as the risk state turning point node;

[0131] S43, divide the risk conduction coupling path into a plurality of independent risk propagation links according to the risk state turning point node, and determine the starting node, the terminal node and the spatial boundary of each link;

[0132] S44, track the space-time embedding vector change of all nodes in each risk propagation link, establish a node space-time feature difference matrix for each link, and calculate the feature difference index in the matrix;

[0133] S45, analyze the convergence and difference between any two adjacent risk propagation links according to the feature difference index, and determine the correlation level between the links;

[0134] S46, adaptively aggregate the risk propagation links according to the correlation levels between the links, and dynamically divide to form a plurality of risk control sub-domains;

[0135] S47, monitor the feature change trend and risk propagation intensity of the nodes within the risk control sub-domain boundary, and dynamically adjust the sub-domain boundary by sub-domain.

[0136] By calculating the node feature change trend and change acceleration on the risk transmission path in real time, the risk state turning point node is automatically identified and the risk propagation link is accurately divided; the feature difference matrix is established by using the node space-time embedding vector, and the difference index is calculated to determine the correlation level between the links, and the risk propagation link is dynamically aggregated or separated, and a plurality of risk control sub-domains are formed in real time; by dynamically monitoring the feature trend and risk propagation intensity of the nodes within the sub-domain boundary, the sub-domain boundary is dynamically adjusted in real time, thereby improving the accuracy and flexibility of the risk control region division, and effectively improving the overall performance of the power distribution network risk management.

[0137] In the embodiment, the S5 specifically includes:

[0138] S51, calculate the node risk state index change trend for each node within the risk control sub-domain, identify the node set with consistent risk state index change trend by node, and determine the spatial distribution characteristics of the node risk evolution within the sub-domain;

[0139] S52, track the historical risk control instruction sequence executed by each node within the spatial distribution characteristics, determine the front and rear dependency relationship between the historical risk control instructions by node, and form the historical decision interaction network between nodes;

[0140] S53, calculate the dynamic influence factor of the historical risk control instruction on the current node risk state by node according to the historical decision interaction network between nodes, and determine the specific action degree of the historical decision on the current risk state;

[0141] S54, extract the node risk state index, the dynamic influence factor of the historical decision, and the node space-time embedding vector, construct a risk state-decision history-space-time feature joint state data set by node, and fuse to generate a three-dimensional joint optimization state space;

[0142] S55, calculate the sensitivity coefficient of the node risk state change to the risk control instruction update demand according to the three-dimensional joint optimization state space by node, and dynamically determine the priority of the node risk control instruction update;

[0143] S56, determine the update interval time of the risk control instruction sequence within the sub-domain according to the node update priority, and dynamically adjust the execution order of the risk control instruction of each node;

[0144] S57, monitor the fluctuation amplitude and frequency of the overall risk state of the risk control sub-domain, automatically trigger the updated risk control instruction sequence, and execute the risk control instruction node by node.

[0145] By calculating the trend of the node risk state index in real time and identifying the evolution characteristics of the node risk, a real-time interaction network of historical decision-making of the node is formed, the dynamic influence degree of the historical risk decision on the current risk state is quantitatively analyzed, a three-dimensional joint optimization state space of risk state-decision history-spatiotemporal characteristics is constructed, and the sensitivity coefficient and priority of the risk control instruction update are determined in real time to dynamically adjust the execution order and update frequency of the risk control instruction, thereby improving the pertinence and real-time performance of the risk control strategy and significantly enhancing the dynamic adaptability and response speed of the risk decision of the power distribution network.

[0146] In the embodiment, the S6 specifically includes:

[0147] S61, when the risk control instruction sequence is executed in the risk control sub-domain, the sub-domain boundary node monitors the change amount of the spatiotemporal embedding vector and the risk score after the previous instruction execution, and when any threshold change amount reaches a preset asynchronous trigger condition, the boundary node automatically generates a boundary node feature snapshot containing the spatiotemporal embedding vector, the risk quantization value, the time stamp and the node unique identifier;

[0148] S62, based on the self-organizing grid topology, the boundary node feature snapshot is packaged as a broadcast data packet with a time slot identifier, and is periodically broadcast to the sub-domain directly physically adjacent to the sub-domain through the adjacent link in a multi-path parallel manner;

[0149] S63, after receiving the boundary node feature snapshot broadcast by the adjacent sub-domain, each sub-domain parses the spatiotemporal embedding vector and the risk quantization value of the boundary node in the snapshot, and sequentially sorts them by time stamp to construct a cross-period boundary coupling graph containing the boundary node information of the current period and the previous period, wherein the nodes of the boundary coupling graph represent the boundary nodes, and the edges represent the risk association between the nodes in the adjacent period;

[0150] S64, for each edge in the cross-period boundary coupling graph, the difference degree of the spatiotemporal embedding vector and the fluctuation rate of the risk quantization value of the boundary node in the adjacent period are calculated edge by edge, and the comprehensive index of the difference degree and the fluctuation rate is compared with the preset mutation threshold, and the risk fluctuation mutation edge meeting the condition higher than the threshold is automatically identified, and a dynamic negotiation trigger signal is generated taking the mutation edge as the trigger source;

[0151] S65, the dynamic negotiation trigger signal is broadcast to all boundary nodes in the sub-domain through multiple backup communication channels in a load balancing manner, and the boundary nodes perform weighted fusion calculation based on the local spatiotemporal embedding vector and the neighborhood snapshot information after receiving the trigger signal to adjust the risk quantization value calibration factor of the boundary node;

[0152] S66, each sub-domain generates a micro-transition risk control instruction mapping sequence containing micro-transition incremental values and time sequence order based on the calibrated boundary node risk rating value and multi-sub-domain interaction feedback information, and each micro-transition incremental value in the mapping sequence is combined with the risk gradient difference of the adjacent sub-domain to drive the evolutionary self-coordination of the cross-sub-domain risk control instruction;

[0153] S67, each sub-domain issues the updated risk control instruction to each node in the sub-domain according to the mapping sequence, and continuously monitors the change of the spatio-temporal embedding vector of the boundary node and the calibrated risk quantization value, and when the change rate of the spatio-temporal embedding vector or the fluctuation rate of the risk quantization value of any boundary node again meets the asynchronous triggering condition, a new boundary node feature snapshot is automatically generated, an adaptive boundary suppression closed loop is formed, and the state of the two-way information cooperation link is iteratively updated.

[0154] The asynchronous triggering mechanism of the boundary node is used to monitor the change of the node feature in real time, generate a boundary node feature snapshot, and use the self-organizing grid topology to perform multi-path parallel broadcasting, so as to realize the construction of the cross-time period boundary coupling graph; the spatio-temporal embedding difference and the risk quantization fluctuation rate of the boundary node are automatically identified, the negotiation triggering signal and the micro-transition risk control instruction mapping sequence are dynamically generated, the real-time cooperation of the information between the risk control sub-domains and the dynamic self-coordination adjustment of the risk control instruction are realized, and the real-time response capability and the cross-region cooperation precision of the power distribution network risk control are effectively improved, and the overall robustness of the risk control strategy is enhanced.

[0155] In the embodiment, the S7 specifically comprises:

[0156] S71, for each risk control sub-domain, the spatio-temporal embedding vector and the corresponding risk control instruction execution feedback state of all nodes in the sub-domain are asynchronously captured, and a distributed adaptive error spectrum is generated by accumulating the state transition of the nodes, wherein the distributed adaptive error spectrum represents the error distribution characteristics of the node risk state in the time domain and the space domain;

[0157] S72, the distributed adaptive error spectrum is compared with a historical multi-scale error spectrum in parallel, and the spectrum resonance detection mechanism is used to compare the spectrum energy level distribution, when the cross-time period error energy level aggregation appears in the error spectrum and the spectrum resonance coefficient exceeds a preset stable threshold, a high-energy level error node group is identified, so that it is determined that the local risk convergence state is degraded;

[0158] S73, after it is determined that the local risk convergence state is degraded, an internal path regeneration mechanism of the sub-domain is triggered, a plurality of redundant path candidate sets corresponding to the high-energy level error node group are autonomously queried from a pre-stored redundant control path library, wherein the redundant control path library has been pre-stored according to the node spatio-temporal embedding matching degree, the path redundancy, and the fault tolerance level;

[0159] S74, for the candidate redundant path set, the spatiotemporal embedding vector spectrum compatibility index of each path node is calculated in turn, the link oscillation amplitude index of the path relay node and the node load jump tolerance index are calculated, and the parallel adaptive priority calculation strategy is adopted to select the redundant control path with the highest adaptive priority as the switching target;

[0160] S75, based on the starting node, relay node and terminating node information of the selected redundant control path, a token type command sequence is generated according to the self-organizing token ring protocol, which contains the control instructions and token passing rules of each node in the redundant path, and is seamlessly switched to the token type command sequence in the current control time slot;

[0161] S76, after switching to the redundant control path, the asynchronous hedging mechanism is started in parallel in the sub-domain, the convergence rate difference and risk control resonance strength of each node on the redundant path are collected, when the resonance strength is lower than the preset resonance stability threshold, the fast rollback mechanism is triggered and the step S73 is re-entered to retrieve and switch to the next redundant control path;

[0162] S77, when the switched redundant control path reaches the convergence and risk resonance balance state, the adaptive priority weight of the corresponding path in the redundant control path library is updated through the dynamic memory interaction mechanism, and the asynchronous capture and monitoring of the regular distributed adaptive error spectrum in the sub-domain are restored.

[0163] By asynchronously capturing node state transition data, a distributed adaptive error spectrum is constructed and a spectrum resonance mechanism is used to identify the local risk convergence degradation state; the redundant path regeneration mechanism is triggered, the pre-stored path library is queried autonomously, and the path is selected based on the spatiotemporal embedding compatibility, link oscillation amplitude and fault tolerance index; the self-organizing token ring protocol is used to realize the seamless switching of the risk control path, the convergence rate and the risk resonance strength are monitored in real time, and the path priority is updated through the dynamic memory mechanism, thereby effectively improving the rapid identification and response ability of the risk convergence degradation state, and significantly improving the real-time and reliability of the risk decision path switching.

[0164] Embodiment 1:

[0165] In order to verify the feasibility of the application in implementation, the application is applied to the real-time control and management scene of the operation risk of a regional power grid of a certain power grid company, which specifically relates to the risk state monitoring, risk propagation path prediction and control strategy optimization tasks of multiple substations and distributed new energy power generation nodes in the regional power grid. The new energy ratio in the regional distribution network is high, the load changes quickly, and the interaction between nodes is complex. Traditional methods rely on artificial experience to preset static risk models and parameters, and cannot accurately assess and effectively control risks in real time, especially in the case of sudden abnormal load fluctuations and extreme weather conditions. The traditional model has poor prediction accuracy and obvious response lag, making it difficult to quickly develop effective risk control strategies, which poses a great challenge to the safe and stable operation of the power grid.

[0166] In actual application, first, the power flow data, voltage data and load data of all nodes in the distribution network are collected in real time, the data change rate and trend of each node are calculated in real time, the dynamic feature space of the node is generated, and the real-time dynamic correlation between nodes is determined. Then, according to the dynamic correlation of the node, the lines directly connected to the node are selected, the real-time load level and fluctuation trend of each line are calculated, and a real-time node-line state association mapping diagram is established. On this basis, a node-edge double-space dynamic coupling network is constructed, the network topology is detected in real time and the network weight is updated, and a dynamic graph model of the distribution network is formed.

[0167] Next, the system performs a self-supervised learning task on the dynamic graph of the distribution network, dynamically adjusts the node feature correlation weight through a multi-level node neighborhood feature interaction mechanism, and obtains a node spatio-temporal embedding vector. Then, a multi-scale dynamic clustering risk sensitive network is constructed based on the node embedding vector, the risk diffusion dominant path is automatically identified in real time, the risk propagation trajectory is formed and the risk diffusion trend is predicted. According to the real-time risk propagation trajectory, the distribution network is divided into multiple risk control subdomains, and the risk propagation is accurately monitored and managed.

[0168] For each risk control subdomain divided, the system evaluates the risk dynamic evolution process, real-time fusion of node risk state indicators, historical risk control decisions and node spatio-temporal embedding features, and establishes a three-dimensional joint optimization state space of risk state-decision history-spatio-temporal features to automatically update the risk control instruction sequence of the nodes in the subdomain. At the same time, by constructing a distributed bidirectional information collaboration link between neighborhoods, the risk state information of the boundary nodes between each subdomain is shared in real time, and the cross-regional risk control strategy is self-coordinated and optimized.

[0169] In the risk control instruction execution process, the system monitors the boundary node space-time embedding vector and the risk score change in real time. Once the node characteristics change reaches the preset asynchronous trigger condition, the boundary node generates a feature snapshot and broadcasts it to the adjacent sub-domain, realizing real-time synchronization update of the risk state between sub-domains. In addition, by constructing an asynchronous monitoring mechanism and a distributed error spectrum resonance analysis method, the system can capture and determine the risk convergence degradation of the sub-domain in real time. Once the risk convergence state is detected to be significantly degraded, the preset redundant path switching strategy is triggered.

[0170] To verify the actual effect of the application, 30 key nodes in the regional power grid are selected for risk state prediction and control effect evaluation. The performance of the traditional artificial static risk assessment method and the method of the application in predicting risk state indicators, risk propagation path and risk convergence effect is compared. In the specific experiment process, the system uses the Bayesian optimization algorithm for automatic optimization and adaptive adjustment of model parameters. After 50 rounds of Bayesian optimization iteration, the risk feature embedding dimension is finally selected as 64, the risk clustering scale is 4, the risk sensitive network dynamic clustering radius threshold is 0.05, the risk control instruction update sensitivity coefficient threshold is 0.08, and the boundary node asynchronous trigger threshold is 0.03.

[0171] The following table shows the risk state prediction results and measured value comparison data of some key nodes:

[0172] Table 1 Comparison of node risk state prediction and measured performance

[0173] Node number Risk state measured value Risk state predicted value Error absolute value A-01 0.58 0.56 0.02 A-05 0.73 0.71 0.02 A-12 0.65 0.66 0.01 B-07 0.41 0.40 0.01 B-15 0.85 0.83 0.02

[0174] As can be clearly observed from the above Table 1 data, the absolute error of the application in node risk state prediction is within 0.02, showing high precision and reliability of risk state prediction. Taking node A-05 as an example, the prediction error of the traditional static risk model is about 0.11, while the prediction error of the application method is only 0.02, which is reduced by 81.8% compared with the traditional method, indicating that the risk assessment model of the application has significantly improved in accuracy.

[0175] Further analysis of the risk propagation path prediction performance, randomly selected 5 risk propagation dominant paths and subsequent measured paths are compared, the results are shown in the following table:

[0176] Table 2 Comparison of risk propagation path prediction and measured path

[0177] Propagation path number Measured path node sequence Predicted path node sequence Path accuracy rate P-1 A-01→A-05→A-12 A-01→A-05→A-12 100％ P-2 B-03→B-07→B-15 B-03→B-07→B-15 100％ P-3 C-08→C-11→C-14 C-08→C-10→C-14 66.7％ P-4 D-02→D-05→D-09 D-02→D-05→D-09 100％ P-5 E-01→E-04→E-08 E-01→E-04→E-07 66.7％

[0178] As shown in Table 2, the prediction accuracy of the risk propagation path of the application is generally high, and the accuracy of most paths reaches 100%, which is significantly better than the accuracy of 60% of the traditional static model, and reflects the accuracy and reliability of the application in capturing the risk diffusion trend.

[0179] In terms of risk convergence effect evaluation, the sub-domain risk decision convergence period is used as an evaluation index. The average convergence time of the traditional method is 18 minutes, while the average convergence time of the method of the application is reduced to 4.2 minutes after asynchronous monitoring and redundant path switching optimization, and the convergence speed is increased by about 76.7%, which greatly improves the efficiency of power distribution network risk management.

[0180] In summary, the embodiment fully verifies that the application effectively realizes node-line dynamic coupling modeling, accurate risk state prediction and efficient risk propagation path identification, real-time risk decision dynamic coordination and redundant path rapid switching optimization through the deep fusion of self-supervised spatio-temporal graph transformer and meta-reinforcement learning optimization algorithm, significantly improves the accuracy, real-time performance and robustness of power distribution network risk prediction and control in practical applications, and can provide reliable technical support for real-time risk management in complex dynamic power distribution network environment, and has wide application prospect.

[0181] The above is only a preferred specific embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical scheme and inventive concept of the application within the technical range disclosed by the application, which should be covered within the protection scope of the application.

Claims

1. A method for controlling the operation risk of a distribution network based on Bayesian optimization, characterized in that, Includes the following steps: S1. Establish a node-edge dual-space coupling dynamic diagram using multi-source operation data of the distribution network; S2. Based on dynamic graphs, through a self-supervised task of multi-level spatiotemporal interaction between nodes and their neighborhoods, the associated feature weights of nodes under different operating states are dynamically adjusted node by node to generate spatiotemporal embedding vectors. S3. Based on spatiotemporal embedding vectors, a risk-sensitive network with multi-scale dynamic clustering is constructed to automatically extract risk diffusion paths under operating conditions and predict risk propagation trajectories. S4. Utilize risk propagation trajectories to automatically form continuous and dynamic risk evolution links across time periods within the distribution network, and dynamically divide the distribution network risk control subdomains according to the spatiotemporal embedding change trends on the links. S5. For dynamically formed risk control subdomains, evaluate the dynamic evolution process of operational risks within the subdomains, construct a joint optimization state space of risk status, decision history, and spatiotemporal characteristics, and update the risk control instruction sequence based on the evaluation results. S6. During the dynamic execution of risk control instruction sequences, construct distributed bidirectional information collaboration links between neighboring domains for synchronous updates of risk status between risk control subdomains and self-coordinated adjustments of cross-regional risk control instructions. S7. Monitor the dynamic convergence of decisions in each risk control subdomain. When a degradation of the local risk convergence state is predicted, automatically start the pre-built redundant control path to achieve seamless switching of risk decision paths. S6 specifically includes: S61. When a risk control instruction sequence is executed within the risk control subdomain, the subdomain boundary node monitors the change in its spatiotemporal embedding vector and risk score relative to the last instruction execution. When any threshold change reaches the preset asynchronous trigger condition, the boundary node automatically generates a boundary node feature snapshot containing the node's spatiotemporal embedding vector, risk quantification value, timestamp, and unique node identifier. S62. Based on the self-organizing mesh topology, the feature snapshots of boundary nodes are encapsulated into broadcast data packets with time slot identifiers, and periodically broadcast to subdomains that are directly physically adjacent to this subdomain in a multi-path parallel manner through adjacency links. S63. After receiving the boundary node feature snapshots broadcast by adjacent subdomains, each subdomain parses the spatiotemporal embedding vector and risk quantification value of the boundary nodes in the snapshot, and sorts them sequentially by timestamp to construct a cross-time period boundary coupling graph containing boundary node information of the current time period and the previous time period. The nodes in the boundary coupling graph represent boundary nodes, and the edges represent the risk association between nodes in adjacent time periods. S64. For each edge in the cross-time period boundary coupling graph, calculate the spatiotemporal embedding vector difference and risk quantification value volatility of the boundary node in adjacent time periods. Based on the comprehensive index of difference and volatility, compare it with the preset mutation threshold, automatically identify the risk volatility mutation edge that meets the condition of exceeding the threshold, and generate a dynamic negotiation trigger signal with the mutation edge as the trigger source. S65. The dynamic negotiation trigger signal is broadcast to all boundary nodes through multiple backup communication channels within the subdomain in a load-balanced manner. After receiving the trigger signal, the boundary nodes perform weighted fusion calculation based on the local spatiotemporal embedding vector and neighborhood snapshot information to adjust the risk quantification value calibration factor of the boundary nodes. S66. Based on the calibrated boundary node risk rating value and multi-subdomain interactive feedback information, each subdomain self-organizes to generate a micro-transition risk control instruction mapping sequence containing micro-transition increment values ​​and temporal order. Each micro-transition increment value in the mapping sequence is combined with the risk gradient difference of neighboring subdomains to drive the evolutionary self-coordination of cross-subdomain risk control instructions. S67. Each subdomain issues updated risk control instructions to each node within the subdomain according to the mapping sequence, and continuously monitors the changes in the spatiotemporal embedding vector of the boundary node and the calibrated risk quantification value. When the rate of change of the spatiotemporal embedding vector of any boundary node or the volatility of the risk quantification value meets the asynchronous triggering condition again, a new boundary node feature snapshot is automatically generated, forming an adaptive boundary suppression closed loop and iteratively updating the bidirectional information cooperation link status.

2. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 1, characterized in that, S1 specifically includes: S11. Collect power flow data, voltage data, and load data of all nodes in the distribution network, calculate the data change rate and trend for each node, and construct a dynamic hierarchical system of node characteristics using the node data change rate. S12. Based on the dynamic hierarchical system of node characteristics, determine the dynamic association relationship between nodes, calculate the data association strength and association direction between nodes, and form an updated node association strength matrix. S13. Based on the node association strength matrix, dynamically select the distribution lines directly connected to each node, and calculate the line load level and fluctuation trend for each line to establish a dynamic association mapping between distribution network nodes and corresponding lines. S14. Based on the dynamic association mapping, construct a node-edge bidirectional state coupling network that simultaneously expresses the electrical connection relationship between nodes and the dynamic change trend of data, and update the topology of the coupling network. S15. Detect the topology changes of the node-edge bidirectional state coupling network, and adaptively update the dynamic coupling weights between nodes and corresponding lines based on the frequency of topology changes, the number of changing nodes, and the severity of coupling states. S16. Based on the updated dynamic coupling weights and topology, transform the node-edge bidirectional state-coupled network into a dynamic graph.

3. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 1, characterized in that, S2 specifically includes: S21. Based on the topology of the dynamic graph and the changes in node characteristic data, identify neighboring nodes whose data interaction frequency in the neighborhood of a node is higher than a preset threshold, determine the direction of data interaction between nodes, and form a node neighborhood interaction relationship graph. S22. Based on the node neighborhood interaction relationship graph, divide the neighborhood into direct neighborhood, secondary neighborhood and edge neighborhood to form a multi-level structure of node neighborhood, and determine the feature interaction priority of different neighborhood levels. S23. Based on the amplitude and frequency of changes in the node's own characteristic data, determine the feature sensitivity parameters, and filter the feature data in the neighborhood of different levels of the node according to the feature sensitivity parameters. S24. Based on the neighborhood feature data, calculate the data interaction sensitivity coefficient between the node and the neighborhood nodes at each level, and dynamically adjust the feature association weight of the nodes in each neighborhood level based on the sensitivity coefficient. S25. Based on the dynamically adjusted feature association weights, determine the optimal feature aggregation mode to aggregate the filtered feature information of the node itself and its neighborhood. S26. Utilize the feature aggregation mode to fuse the node's own features with multi-level neighborhood feature data, and generate a spatiotemporal embedding vector for each node.

4. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 1, characterized in that, S3 specifically includes: S31. Based on the spatiotemporal embedding vector of nodes, construct a multi-dimensional dynamic perception network for feature interaction between nodes, and identify the implicit feature interaction relationship between nodes on a node-by-node basis. S32. Determine multi-scale cluster centers based on implicit feature interaction relationships, and generate dynamic clustering topology between nodes using cluster centers at each scale; S33. Based on the changing trend of node embedding vectors in dynamic clustering topology, detect boundary regions with abnormal feature changes between node sets, and form risk-sensitive regions with the boundary regions as the core. S34. Track the node sequence with the most drastic changes in node embedding vectors within the risk-sensitive area and extract the dominant path of risk diffusion; S35. Analyze the temporal change direction and intensity of the node embedding vector along the dominant risk diffusion path to form a continuous risk diffusion state chain. S36. Dynamically evaluate the state transitions between nodes in the risk diffusion state chain to generate a multi-step continuous evolution model for risk diffusion. S37. Based on the multi-step continuous evolution model of risk diffusion, deduce the risk diffusion development direction of the risk diffusion path and predict the risk propagation trajectory under the operation state of the distribution network.

5. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 4, characterized in that, S31 specifically includes: S311. Calculate the rate and direction of change of node feature vectors based on node spatiotemporal embedding vectors, and generate a multi-dimensional feature space of node feature change trends for each node. S312. Map the features of each node in the multi-dimensional feature space, and determine the spatial positional relationship between each node and the feature change trends of other nodes. S313. Based on the spatial relationship of the changing trends of node characteristics, identify the characteristic transmission path between nodes and mark the transmission direction on the path; S314. Calculate the contribution coefficient of each node in the feature transmission path to the overall feature transmission intensity of the path node by node, and determine the key interactive nodes whose contribution coefficient exceeds the preset threshold. S315. Monitor the synchronicity of feature change trends between key interaction nodes, calculate the synchronicity index and use the index to characterize the implicit feature interaction strength between nodes. S316. Based on the implicit feature interaction strength, the implicit feature interaction relationships between nodes are divided into three levels: high intensity, medium intensity, and low intensity, forming an implicit feature interaction hierarchical system.

6. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 1, characterized in that, S4 specifically includes: S41. Based on the spatiotemporal embedding vector change trend of nodes within the risk propagation trajectory, automatically extract the continuous risk transmission coupling path between nodes across time periods, and calculate the characteristic change trend vector of each node in the path. S42. Calculate the rate of change and acceleration of change of the characteristic change trend vector of each node on the risk transmission coupling path, and determine the node whose acceleration exceeds the preset threshold as the risk state turning point node. S43. Based on the risk state inflection point, the risk transmission coupling path is divided into multiple independent risk propagation links, and the starting node, ending node and spatial boundary of each link are determined. S44. Track the spatiotemporal embedding vector changes of all nodes in each risk propagation link, establish a spatiotemporal feature difference matrix for each node in each link, and calculate the feature difference index within the matrix. S45. Analyze the convergence and differences between any two adjacent risk propagation links based on the characteristic difference index to determine the association level between the links; S46. Based on the association level between links, the risk propagation links are adaptively aggregated and dynamically divided into multiple risk control subdomains. S47. Monitor the characteristic change trend and risk propagation intensity of nodes within the risk control subdomain boundary, and dynamically adjust the subdomain boundary on a subdomain-by-subdomain basis.

7. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 1, characterized in that, S5 specifically includes: S51. Calculate the changing trend of node risk status indicators for each node in the risk control subdomain, identify the set of nodes with consistent changing trends of risk status indicators for each node, and determine the spatial distribution characteristics of node risk evolution in the subdomain. S52. Track the sequence of risk control instructions executed by each node in the spatial distribution characteristics, determine the dependencies between historical risk control instructions for each node, and form a historical decision-making interaction network between nodes. S53. Based on the historical decision-making interaction network between nodes, calculate the dynamic impact factor of historical risk control instructions on the current node risk status for each node, and determine the specific degree of effect of historical decisions on the current risk status. S54. Extract node risk status indicators, historical decision dynamic influencing factors and node spatiotemporal embedding vectors, construct a joint state data set of risk status-decision history-spatiotemporal features for each node, and fuse them to generate a three-dimensional joint optimization state space. S55. Calculate the sensitivity coefficient of node risk state changes to risk control instruction update requirements on a node-by-node basis according to the three-dimensional joint optimization state space, and dynamically determine the priority of node risk control instruction updates. S56. Determine the update interval of the risk control instruction sequence within the subdomain based on the node update priority, and dynamically adjust the execution order of risk control instructions for each node. S57. Monitor the fluctuation amplitude and frequency of the overall risk status of the risk control subdomain, automatically trigger the updated risk control instruction sequence, and execute the risk control instructions node by node.

8. The method for controlling distribution network operation risks based on Bayesian optimization according to claim 1, characterized in that, Specifically, S7 includes: S71. For each risk control subdomain, asynchronously capture the spatiotemporal embedding vectors of all nodes in the subdomain and the corresponding risk control instruction execution feedback states. Generate a distributed adaptive error spectrum through node state transitions, where the distributed adaptive error spectrum represents the error distribution characteristics of node risk states in the time and spatial domains. S72. The distributed adaptive error spectrum is compared with the historical multi-scale error spectrum in parallel. The spectral resonance detection mechanism is used to compare the spectral energy level distribution. When the error spectrum shows cross-time period error energy level aggregation and the spectral resonance coefficient exceeds the preset stability threshold, the high-energy error node group is identified, thereby determining that the local risk convergence state has degraded. S73. After determining that the local risk convergence state has degraded, the subdomain internal path regeneration mechanism is triggered to autonomously query the redundant control path library for multiple candidate sets of redundant paths corresponding to the high-energy error node group. The redundant control path library has been pre-classified and stored according to the node spatiotemporal embedding matching degree, path redundancy and fault tolerance level. S74. For the candidate redundant path set, calculate the spatiotemporal embedding vector spectrum compatibility index of the nodes in each path, the link oscillation amplitude index of the relay nodes in the path, and the node load jump tolerance index in turn. Then, adopt a parallel adaptation priority calculation strategy to select the redundant control path with the highest adaptation priority as the switching target. S75. Based on the information of the starting node, relay node and ending node of the selected redundant control path, generate a token command sequence according to the self-organizing token ring protocol. The token command sequence contains the control instructions and token passing rules of each node in the redundant path, and seamlessly switch to the token command sequence in the current control time slot. S76. After switching to the redundant control path, the asynchronous hedging mechanism is started in parallel within the subdomain. The convergence rate difference and risk control resonance intensity of each node on the redundant path are collected. When the resonance intensity is lower than the preset resonance stability threshold, the fast rollback mechanism is triggered and the process re-enters step S73 for cyclic retrieval and switches to the next redundant control path. S77. When the switched redundant control path reaches the convergence and risk resonance equilibrium state, the adaptive priority weight of the corresponding path in the redundant control path library is updated through the dynamic memory interaction mechanism, and the asynchronous capture and monitoring of the conventional distributed adaptive error spectrum in the subdomain is restored.

Images