A method and system for diagnosing power system fault risk

Abstract

The present invention relates to the technical field of power system fault risk diagnosis, and specifically discloses a power system fault risk diagnosis method and system. By building a power system fault risk diagnosis system, a power system fault risk diagnosis method is implemented based on the power system fault risk diagnosis system. In the method, a spatiotemporal dynamic model based on time changes is established to obtain the state of each node and each line state in real time as input, and the spatiotemporal dynamic model is used to predict the state at the next moment, thereby performing a multi-dimensional analysis of the system fault. Finally, an intelligent decision-making and feedback adjustment unit is used to calculate the probability of power system fault risk, and to distinguish, accurately compare and divide the fault level. The problem of traditional risk diagnosis models, which lacks general mathematical modeling for different disaster scenarios and is difficult to fully reflect the multiple impacts of disasters on power systems, is solved.

Description

Technical Field

[0001] The present application relates to the technical field of power system fault risk diagnosis, and specifically discloses a power system fault risk diagnosis method and system. Background Art

[0002] As global climate change intensifies, natural disasters occur frequently, causing frequent power outages and serious economic losses to the power system. Diagnosing the failure risk of the power system under natural disasters, evaluating its disaster resistance in extreme events to reduce failure losses, and providing early warnings have become a common means to improve the power system's disaster resistance under extreme natural disasters and reduce the occurrence of power outages.

[0003] Through the development of artificial intelligence technology, deep learning and reinforcement learning have brought new breakthroughs in extreme risk modeling. Deep learning and reinforcement learning can provide an effective data-driven method for risk assessment of power systems with multiple highly nonlinear complex factors and lack of complete physical models.

[0004] However, the existing risk diagnosis models based on deep learning and reinforcement learning have limitations in their applicability in multiple scenarios. Especially in extreme weather scenarios, they usually only rely on historical data to fit the meteorological intensity, and cannot independently perform general and accurate mathematical modeling of different disaster scenarios, making it difficult to fully reflect the multiple impacts of disasters on the power system.

[0005] The present invention provides a method and system for diagnosing fault risks in an electric power system, so as to solve the above-mentioned problems. Summary of the invention

[0006] The purpose of the present invention is to solve the problem that the traditional risk diagnosis model lacks universal and accurate mathematical modeling for different disaster scenarios and is difficult to fully reflect the multiple impacts of disasters on the power system.

[0007] In order to achieve the above object, the basic scheme of the present invention provides a method for diagnosing fault risks in a power system, comprising the following steps:

[0008] Step S1: Real-time multi-source data is collected and pre-processed to obtain a disaster data information set during the period when the system failure occurs, including meteorological data information and equipment monitoring data, and a system network topology map before and after the disaster is constructed, as well as disaster data information during the period when the system failure occurs;

[0009] Step S2: Identify changes in node and line status based on the system power grid topology before and after the disaster, obtain the status of each node and each line, and annotate the status of each node and the line edge attributes to obtain an annotation result, obtain an annotation result set, and divide it into an annotation result training set and an annotation result test set;

[0010] Step S3: construct a spatiotemporal dynamic model, train the spatiotemporal dynamic model based on the labeled result training set, and test the trained spatiotemporal dynamic model based on the standard result test set, and save it as a new spatiotemporal dynamic model after passing the test;

[0011] Step S4: Based on the new spatiotemporal dynamic model obtained in step S3, the states of each node in the system operation and the states of each line at the current moment are taken as input, and the state vector of each node is used to perform a multi-dimensional analysis of the system failure at the next moment;

[0012] Step S41: Based on the disaster data information set of the system failure period collected in step S1, a node failure risk calculation model is constructed with the failure probability of each node as the result;

[0013] Step S42: Based on the node failure risk calculation model in step S41, a line failure risk calculation model is constructed with the failure probability of each line as the result;

[0014] Step S5: Based on the node failure risk calculation model established in step S41, calculate the sensitivity of the node to the disaster factor to obtain a comprehensive sensitivity factor;

[0015] Step S6: Establish a disaster warning sensitivity factor. When the size of the comprehensive sensitivity factor exceeds the disaster warning sensitivity factor, based on the node failure risk calculation model, the line failure risk calculation model and the meteorological information system's meteorological disaster forecast information for the next moment, the power system failure risk quantitative index is calculated to obtain the system failure probability, and risk feedback warning and failure size are provided to system management personnel based on the size of the system failure probability.

[0016] Further, the implementation of step S2 includes the following specific steps:

[0017] Step S21: extracting the disaster conditions of lines and nodes through the power grid topology diagram before and after the disaster, and identifying the changes in the states of nodes and lines to obtain the results of the states of each node and each line;

[0018] Step S211: Calculate the difference between the adjacency matrix before the disaster and the adjacency matrix after the disaster, identify the line state change, and analyze and extract its state;

[0019] Step S212: Calculate the node state vector before the disaster and the node state vector after the disaster, identify the node state change, and analyze and extract its state;

[0020] Step S213: Obtain the dependency strength between nodes by calculating the correlation of node state changes;

[0021] Step S22: labeling the node status and line edge attributes according to the obtained results to obtain labeling results;

[0022] Step S221: Marking the status of each node according to the node status analysis and judgment result;

[0023] Step S222: marking each line status according to the line status analysis and judgment result;

[0024] Step S23: Obtain a labeling result set based on the labeling results of step S221 and step S222, and divide it into a labeling result training set and a labeling result test set for storage.

[0025] Further, the implementation of step S3 includes the following specific steps:

[0026] Step S31: Build the network structure of SDM-GNN, generate model structure programming to obtain the spatiotemporal dynamic model;

[0027] Step S32: Establishing a state transfer function based on time sequence;

[0028] Step S33: performing spatial feature extraction to learn the node state change relationship in the topological structure;

[0029] Step S34: extracting time features and using a time series model to capture the state evolution of nodes over time;

[0030] Step S35: updating the fusion features to effectively combine the spatial features and the temporal features to capture the spatiotemporal dynamic information;

[0031] Step S36: Based on the labeled result set obtained in step S2, the spatiotemporal dynamic model is trained through the training set, the node state prediction error is predicted according to the node state error between the iteration result and the actual value, and the model weight is updated until the node state prediction error value is minimized. The training cycle ends, and the trained spatiotemporal dynamic model is tested based on the standard result test set. After passing the test, it is saved as a new spatiotemporal dynamic model.

[0032] Furthermore, during the implementation of step S4, after a multi-dimensional analysis of the system faults, single node faults, multi-node simultaneous faults, single line faults, multi-line faults, and node and line combination faults are diagnosed.

[0033] Further, in the implementation process of step S41, the process of calculating the probability of node damage based on the node failure risk calculation model includes the following steps:

[0034] Step S411: based on the meteorological data information and the disaster data information, characteristic parameters are obtained: disaster intensity, impact range and duration, and the characteristic parameters are standardized;

[0035] Step S412: Perform statistical analysis on characteristic parameters to construct typical characteristic distribution functions of various disasters;

[0036] Step S413: constructing a joint distribution function of disaster characteristics based on typical characteristic distribution functions of various disasters;

[0037] Step S414: Establish a comprehensive damage model of the time evolution of the fusion component based on the joint distribution function of the disaster characteristics, and calculate and obtain the comprehensive damage probability of the node.

[0038] Furthermore, during the implementation of step S414, when calculating the damage probability of the node, a time-dependent influencing factor is introduced into the damage probability to describe the impact of the evolution of node vulnerability over time on the damage probability, and a comprehensive damage model for the node is constructed in combination with the disaster intensity, impact range, duration, time and node aging to obtain the comprehensive damage probability of the node.

[0039] Furthermore, during the implementation of step S42, the damage probability of each line is calculated based on the line fault risk calculation model, including the calculation of the failure probability of a single line without chain propagation and the calculation of the failure probability of a multi-line chain propagation path.

[0040] Furthermore, during the implementation of step S6, the calculation of the quantitative index of power system fault risk includes the following steps:

[0041] Step S61: Calculate and obtain the single node failure probability, and calculate and obtain the multi-node simultaneous failure probability based on the single node failure probability;

[0042] Step S62: Calculate and obtain the single line fault rerouting, and calculate and obtain the multi-line fault probability based on the single line fault probability;

[0043] Step S63: Calculate and obtain the node and line combined failure probability based on the multi-node simultaneous failure probability and the multi-line failure probability;

[0044] Step S64: Based on the results calculated in steps S61, S62 and S63, the comprehensive risk value of the system is calculated to obtain the system failure probability.

[0045] Furthermore, during the implementation of step S6, two risk classification thresholds adjusted according to historical data and expert experience are established, namely, risk classification threshold 1 and risk classification threshold 2, wherein risk classification threshold 1 is greater than risk classification threshold 2.

[0046] Power system failure risks are divided into the following three categories:

[0047] High-risk failure: When the probability of system failure is greater than the risk classification threshold of one, it means that the component is very likely to fail in a disaster and needs to be handled with priority;

[0048] Medium risk failure: When the system failure probability is greater than or equal to the risk classification threshold 2 and less than or equal to the risk classification threshold 1, it means that the component has a failure risk, but it is controllable;

[0049] Low risk failure: When the probability of system failure is less than the risk classification threshold of two, it means that the component is safe under the current conditions.

[0050] The basic solution of the present invention also provides a power system fault risk diagnosis system, which is used to implement a power system fault risk diagnosis method as described in any one of the above items, including:

[0051] Data acquisition and processing unit, used for real-time acquisition of multi-source data and pre-processing;

[0052] The component and line analysis unit is used to identify the status of each node and each line according to the preprocessing results, mark the status of each node and the line edge attributes, and divide the marked results;

[0053] The component and line state prediction model building unit is used to construct a spatiotemporal dynamic model and train and test it according to the division results of the annotation results;

[0054] The system failure and disaster analysis unit is used to perform multi-dimensional analysis on the system failure at the next moment according to the tested spatiotemporal dynamic model, and calculate the sensitivity of the node to the disaster factor to obtain the comprehensive sensitivity factor;

[0055] Intelligent decision-making and feedback adjustment unit, through which a disaster warning sensitivity factor is established. When the size of the comprehensive sensitivity factor exceeds the disaster warning sensitivity factor, the system failure probability is calculated based on the multi-dimensional analysis results of the system failure at the next moment and the meteorological disaster forecast information at the next moment, and a risk feedback warning is carried out.

[0056] The principle and effect of this scheme are:

[0057] 1. Compared with the prior art, the present invention implements a power system fault risk diagnosis method based on the power system fault risk diagnosis system by building a power system fault risk diagnosis system. In the method, a disaster data information set of the system fault occurrence period is obtained by a data acquisition and processing unit, and the node state and line state changes are identified based on the collected information set in combination with the component and line analysis unit, and each node state and line edge attribute is labeled and divided into a labeled result training set and a labeled result test set. The component and line state prediction model construction unit constructs a spatiotemporal dynamic model and trains based on the labeled result training set, updates the iterative model, and performs multi-dimensional analysis of the system fault through the system fault and disaster analysis unit based on the new spatiotemporal dynamic model completed by iteration, obtains a comprehensive sensitivity factor by calculating the sensitivity of the component to the disaster factor, and makes a judgment based on the intelligent decision-making unit and the feedback adjustment unit, and establishes a double-layer early warning judgment benchmark: when the obtained comprehensive sensitivity factor is greater than the established disaster early warning sensitivity factor, the system failure probability is calculated and a secondary judgment is made, and the system failure is accurately graded.

[0058] 2. Compared with the prior art, the present invention establishes a spatiotemporal dynamic model based on time changes, which obtains the status of each node and each line in real time as input, and realizes real-time prediction of its status at the next moment through the spatiotemporal dynamic model, and then conducts multi-dimensional analysis of system failures. Finally, the intelligent decision-making and feedback adjustment unit calculates the probability of power system failure risk, and distinguishes and accurately compares and divides the fault level, which solves the problem of traditional risk diagnosis models that lack universal and accurate mathematical modeling for different disaster scenarios and are difficult to fully reflect the multiple impacts of disasters on the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 A flow chart of a power system fault risk diagnosis method proposed in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0061] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0062] A method for diagnosing power system fault risks, for example Figure 1 As shown:

[0063] Including the construction of the power system fault risk diagnosis system, including:

[0064] A data acquisition and processing unit, the input end of which is connected to the equipment monitoring system, sensor network, meteorological information system, and energy management system of the power grid, respectively, for real-time acquisition of multi-source data and pre-processing;

[0065] The component and line analysis unit is composed of a component and line state determination module, a component and line state labeling module connected to the component and line state determination module, and a labeling result integration and processing module connected to the component and line state determination module, and is used to identify each node state and each line state according to the preprocessing result, label each node state and line edge attribute, and divide the labeling result;

[0066] The component and line state prediction model building unit is composed of a comprehensive state processing module, a time feature extraction module connected to the comprehensive state processing module, and a space feature extraction module connected to the comprehensive state processing module, and is used to construct a spatiotemporal dynamic model, and train and test it according to the division results of the annotation results;

[0067] The system fault and disaster analysis unit is composed of a system fault analysis module and a component fault and disaster analysis module connected to the system fault analysis module, and is used to perform a multi-dimensional analysis of the system fault at the next moment according to the tested spatiotemporal dynamic model, and calculate the sensitivity of the node to the disaster factor to obtain a comprehensive sensitivity factor;

[0068] The intelligent decision-making and feedback adjustment unit is composed of a risk warning module, a risk calculation module connected to the risk warning module, and a feedback information module connected to the risk warning module. A disaster warning sensitivity factor is established. When the size of the comprehensive sensitivity factor exceeds the disaster warning sensitivity factor, the system failure probability is calculated based on the multi-dimensional analysis results of the system failure at the next moment and the meteorological disaster prediction information at the next moment, and a risk feedback warning is performed.

[0069] The power system fault risk diagnosis method provided in this embodiment is implemented based on the power system fault risk diagnosis system constructed above, and includes the following steps:

[0070] Step S1: The data acquisition and processing unit collects multi-source data in real time and performs preprocessing to obtain a disaster data information set during the period when the system failure occurs, including meteorological data information and equipment monitoring data, and constructs a system network topology diagram before and after the disaster, as well as obtains disaster data information during the period when the system failure occurs.

[0071] Specifically, the disaster data information set collected during the period of system failure is mainly divided into two parts: power grid equipment operation status information and disaster information.

[0072] In the process of collecting and preprocessing the operating status information of power grid equipment, the collected operating status information of power grid equipment mainly includes:

[0073] The power grid topology before and after the disaster, the disaster conditions of the lines and nodes (such as faults, damage degree, and repair status), and the pre-processed power grid topology before and after the disaster are all represented in the form of a graph structure.

[0074] Topological map before the disaster G pre : Contains status information of nodes and lines.

[0075] G pre =(V,E,A pre );

[0076] ;

[0077] Topological map after the disaster post : Contains status information of nodes and lines.

[0078] G post =(V,E,A post );

[0079] ;

[0080] In which, the power grid components are taken as nodes, V is the node set, representing the n component nodes in the risk diagnosis system, V={v1,v2,...,v n}, and 1≤i≤n, 1≤j≤n, i≠j, E is a set of lines, representing the dependency relationship between components (directed edges), E={e ij}, W represents the edge weight, which represents the strength of the dependency relationship, W = {w ij}, the value range is w ij ∈[0,1];

[0081] A pre is the adjacency matrix before the disaster, A post is the adjacency matrix after the disaster.

[0082] Disaster information is defined as a set: D = {D1, D2, D3, ..., D n}, the disaster information collected mainly includes: meteorological disasters D1 = {typhoon, lightning, freezing}, geological disasters D2 = {earthquake, landslide}, hydrological disasters D3 = {flood, rainstorm}, and the remaining subsets D n Other types of disasters (such as biological or man-made) may be included.

[0083] During the preprocessing process, characteristic parameters are extracted for each disaster, and the characteristics of each disaster are expressed in parameterized form:

[0084] Duration T: represents the duration of the disaster event, represented by the random variable T, usually in hours, and the expression is as follows:

[0085] T=t end -t start ,T≧0;

[0086] Among them, t end is the end time of disaster propagation, t start is the starting time of disaster propagation, and T is in hours.

[0087] Impact range A: represents the geographical scope of the disaster impact, represented by the random variable A, the unit is usually square kilometers, and A∈[0,∞).

[0088] Disaster intensity I: It indicates the physical intensity of the disaster, represented by the random variable I, and the unit depends on the type of disaster (such as wind speed m / s, earthquake magnitude, or water level m, etc.).

[0089] Step S2: Through the component and line state identification module in the component and line analysis unit, the system power grid topology map before and after the disaster obtained in step S1 is used as input, the disaster conditions of the lines and nodes (such as faults, damage degree, and repair status) are extracted, and the changes in the node and line states are identified to obtain the results of each node state and each line state. The obtained results are then used as the input of the component and line state labeling module, and the node state and line edge attributes are labeled to obtain the labeling results. Through the labeling result integration and processing module, based on the labeling results obtained by the component and line state labeling module, a labeling result set is obtained, and it is divided into a labeling result training set and a labeling result test set in a ratio of 7:3 and saved.

[0090] Specifically, the steps include:

[0091] Step S21: The power grid topology map before and after the disaster in the data acquisition unit is used as input, the disaster conditions of the lines and nodes (such as faults, damage degree, and repair status) are extracted, and the changes in the node and line status are identified to obtain the results of the status of each node and each line.

[0092] Step S211: Calculate the difference between the adjacency matrix before the disaster and the adjacency matrix after the disaster, identify line status changes, and analyze and extract their status.

[0093] The difference between the adjacency matrix before the disaster and the adjacency matrix after the disaster is calculated as follows:

[0094] ΔA=A post -A pre ;

[0095] Among them, when ΔA[i,j]=-1, it means that the component node v i and component node v j Damage or interruption of the lines between them;

[0096] When ΔA[i,j]=1, it means that the component node v i and component node v j New or repaired lines between them;

[0097] When ΔA[i,j]=0, it means that the line status has not changed.

[0098] Step S212: Calculate the node state vector before the disaster and the node state vector after the disaster, identify the node state change, and analyze and extract its state.

[0099] It includes the calculation of the node state vector before the disaster and the node state vector after the disaster.

[0100] Among them, the calculation formula of the node state vector before the disaster is as follows:

[0101] ;

[0102] Where S pre Represents node v i The state before the disaster.

[0103] The calculation formula of the node state vector after the disaster is as follows:

[0104] ;

[0105] Where S post Represents node v i The state before the disaster.

[0106] Each node v before and after the disaster i The state change is defined as:

[0107] ;

[0108] When ΔS i >0: indicates node v i Improvement of status, e.g., from impaired to normal;

[0109] When ΔS i =0: indicates node v i The state remains unchanged;

[0110] When ΔSi <0: indicates node v i A state deterioration, such as changing from normal to faulty.

[0111] Step S213: Obtain the dependency strength w between nodes by calculating the correlation of node state changes ij :

[0112] ;

[0113] Among them, Cov(ΔS i ,ΔS j ) represents node v i and node v j Covariance of state changes, Var(ΔS i )、Var(ΔS j ) represents node v i and node v j The variance of the state change, w ij ∈[-1, 1), take a positive value w ij =max(0,w ij ) indicates positive dependence.

[0114] Step S22: using the obtained result as the input of the component and line state labeling module, labeling the state of each node and the line edge attribute to obtain the labeling result.

[0115] Step S221: Mark the status of each node according to the status analysis result.

[0116] Specifically, the node attributes are marked as :

[0117] ;

[0118] Among them, Type(v i ) means: node v i Component type (such as power station, load), Status (v i ) means: node v i Status (normal, faulty).

[0119] And each node v i The state is defined as:

[0120] ;

[0121] Step S222: Mark the status of each line according to the status analysis result.

[0122] Specifically, the line edge attributes are marked as :

[0123] ;

[0124] in, Indicates the operating status (normal, faulty), Indicates the degree of damage, a scalar value .

[0125] Step S23: Based on the annotation results of step S221 and step S222, a annotation result set is obtained through a standard result integration and processing module, and the annotation result set is divided into an annotation result training set and an annotation result test set and saved.

[0126] Specifically, load the labeled result set and randomly shuffle it, use the stratified sampling method to divide the training set and the test set in a ratio of 7:3, where the training set is used for model training and the test set is used for performance evaluation. Save the division results as needed.

[0127] Step S3: In the component and line state prediction model construction unit, the comprehensive state processing module constructs a spatiotemporal dynamic model (SDM) and uses a graph neural network (GNN) to learn the relationship between fault diffusion and timing state. Combining the graph structure and timing dependency, it captures the state changes of nodes and edges in the power grid and the law of fault propagation, predicts the node state and line state, and trains the spatiotemporal dynamic model based on the labeled result training set. The node state prediction error value is obtained according to the node state error between the predicted node state and the actual node state, and the model weight is updated until the node state prediction error value is minimized. The training is terminated and saved as a new spatiotemporal dynamic model.

[0128] Specifically, the steps include:

[0129] Step S31: The network structure of SDM-GNN is constructed by the comprehensive state processing module in the component and line state prediction model construction unit, and the model structure is generated by programming to obtain a spatiotemporal dynamic model. In the established spatiotemporal dynamic model, the power grid topology is represented as a graph G=(V,E).

[0130] Where V is a node set, representing n component nodes in the risk diagnosis system, V={v1,v2,...,v n}, E is a set of lines, representing the dependency relationship between components (directed edges), E={e ij}, and each node v i ∈V has an initial state , each edge e ij ∈E has weight w ij .

[0131] And at time step t, node v i The state of the vector Indicates that the current operating status is When, the historical state set is expressed as: .

[0132] Step S32: Since fault propagation is determined by edge weights and node states, a state transfer function f(·) based on time series is established, and ,in, Represents node v i The state at time t reflects the characteristics of the node in time series. Represents node v i The state at time t-1, Represents node v j The state at time t-1, w ij Represents the edge weight of edge ij, representing node v i and v j The strength of dependencies between nodes.

[0133] Step S33: Perform spatial feature Extraction: Learn the node state change relationships in the topology.

[0134] ;

[0135] in, Represents the current node v i The spatial features at time t, N(i) represents the node v i The set of neighbor nodes, w ij Represents the edge weight of edge ij, representing node v i and v j The strength of the dependency between nodes. W1 is a learnable weight matrix used to linearly transform the states of neighboring nodes. Represents node v j The state at time t, σ is the activation function, and b1 is the bias term used to adjust the output result.

[0136] Step S34: Perform time feature Extraction: Use the time series model to capture the state evolution of the node over time. Update the GRU with spatial features to obtain a new time series state.

[0137] ;

[0138] in, Represents node v i The time series feature state at time t is the state vector of the node, which contains the feature information of the node changing over time. Represents node v iThe spatial features at time t usually include the topological information of the node, the relationship with neighboring nodes, physical properties, etc. Represents node v i The time series state at time t-1 is the node state of the previous time step. GRU combines the state of the previous moment and the spatial characteristics of the current moment , thereby updating the state of the node at the current moment.

[0139] Step S35: Update the fusion features: effectively combine the spatial features and the temporal features to capture the spatiotemporal dynamic information and obtain the node v predicted by the model. i The state at time t :

[0140] ;

[0141] in, Node v representing the model prediction i The final updated state at time t combines the results of spatial and temporal domain information to capture the comprehensive characteristics of the node at the current moment. Represents spatial features (features extracted from node neighbors and edge weight information, capturing the impact of neighbor nodes on the current node).

[0142] ;

[0143] in, Represents time features (represents the dynamic change characteristics of the node in the time series dimension and captures the dynamic changes of the node itself).

[0144] ;

[0145] Among them, concat(·) means to convert and They are concatenated into a new feature vector, W2 is a learnable weight matrix used to adjust the influence of the concatenated features on the final node state, and b2 is a bias term used to adjust the overall value of the node state.

[0146] Step S36: Perform spatiotemporal dynamic model training based on the labeled result training set obtained in step S2, and predict L according to the node state error between the iteration result and the actual value. state , update the model weights. Loop until the node state prediction error L is reached state When the value is the smallest, the loop training ends, and the trained spatiotemporal dynamic model is tested based on the labeled result test set obtained in step S3. After passing the test, it is saved as a new spatiotemporal dynamic model.

[0147] Node state prediction error Lstate :

[0148] ;

[0149] Where N represents the total number of nodes in the graph, Node v representing the model prediction i The state at time t is Represents the actual node v i The state at time t.

[0150] Step S4: Based on the new spatiotemporal dynamic model obtained in step S3, the system fault analysis module in the system fault and disaster analysis unit converts the states of each node and each line at time t in the component and line analysis unit into a state vector of each node As input, we get the state vector of each node at time t+1 , thus performing a multi-dimensional analysis of the system failure at time t+1.

[0151] Specifically, in the implementation process of step S4, the multi-dimensional analysis of the system failure at time t+1 is mainly based on the node state change ΔS according to the node state vectors before and after the prediction. i <0 and line state change ΔA[i,j]=-1, and after analysis, they are classified into the following conditions:

[0152] Single node failure: ΔS i <0 and there is only one i, and there is no ΔA[i,j]=-1;

[0153] Multiple nodes fail simultaneously: , where N is the total number of nodes, and there is no ΔA[i,j]=-1;

[0154] Single line fault: ΔA[i,j]=-1, and there is only one [i,j];

[0155] Multiple line failure:∑ i,j |ΔA[i,j]|>1;

[0156] Combined node and line failure: When a node and a line are affected by a disaster at the same time, it can be expressed by a joint analysis of the node and line state changes: ΔS i ≠0, and ΔA[i,j]≠0;

[0157] Step S41: Based on the disaster data information set of the system failure period collected in step S1, through the component failure and disaster analysis module in the system failure and disaster analysis unit, the failure probability of each component node is described by fusing the damage probability model of the component evolving over time with the joint distribution function based on the disaster characteristics, and a node failure risk calculation model is constructed.

[0158] Specifically, the calculation of the node damage probability based on the node failure risk calculation model is as follows:

[0159] Step S411: Based on the meteorological data information and disaster data information in the disaster data information set during the period when the system failure occurs, characteristic parameters are obtained, and characteristic parameters: duration T, impact range A, and disaster intensity I are standardized. The standardization process is as follows:

[0160] ;

[0161] Among them, μ T , μ A , μ I are the means of T, A, and I, respectively, T , σ A , σ I is the standard deviation of T, A, and I.

[0162] Step S412: Perform statistical analysis on T, A, and I to construct typical characteristic distribution functions of various disasters. The typical characteristic distribution functions are as follows:

[0163] F D ={f T (T),f A (A),f I (I)};

[0164] Among them, f T (T), f A (A), f I (I) represents the probability density function of disaster duration, impact range and intensity, forming a typical characteristic distribution, which is used to describe the impact of different types of disasters on the power grid.

[0165] Step S413: Based on the typical characteristic distribution functions of various disasters, a joint distribution function of disaster characteristics is constructed.

[0166] Specifically, the joint distribution function of disaster characteristics (I, A, T) is F I,A,T (x,y,z), the corresponding probability density function is f I,A,T (x,y,z). The joint distribution describes the comprehensive probability of disaster intensity I, impact range A, and duration T. And:

[0167] ;

[0168] Step S414: Establish a joint distribution function based on disaster characteristics to fuse the damage probability model of the components evolving over time, and calculate and obtain the damage probability of the node.

[0169] In the specific implementation process, the probability of damage to the grid component nodes depends on the disaster intensity I, the impact range A, and the duration T. Its conditional probability can be defined as:

[0170] P c (x,y,z)=P(damaged node c|I=x,A=y,T=z);

[0171] Considering the joint distribution of all disaster characteristics, the overall probability of component damage is:

[0172] ;

[0173] Furthermore, in addition to disaster factors, component nodes also have their inherent characteristics. Combining time factors, component vulnerability and common causal effects can further describe the damage probability model of power grid components that evolves over time under the influence of disasters. The specific implementation steps are as follows:

[0174] First, the time impact factor g(t) is introduced into the damage probability to describe the impact of the evolution of component vulnerability over time t on the damage probability, which is in the form of:

[0175] ;

[0176] in, is the aging coefficient, which indicates the rate at which component performance decays over time.

[0177] After that, the initial damage probability is calculated. The initial damage probability is expressed as the direct impact of the initial event described by ordinary causality. The calculation formula is:

[0178] ;

[0179] in, Representation Node The probability of damage at the initial moment, Indicates the component vulnerability coefficient, reflecting the node inherent characteristics (such as disaster resistance), Representation Node The initial state of the device, usually 0 or 1 (normal or faulty).

[0180] Finally, the comprehensive damage probability model combining disaster intensity I, impact range A, duration T, time t and component aging influence is calculated as follows:

[0181] ;

[0182] Among them, P c Indicates the disaster intensity I, impact range A, and duration T on the component node v i The damage probability, g(t) represents the aging impact factor, which represents the weight of the component performance decay over time, and 1-g(t) represents the weight of the disaster impact ratio changing with time.

[0183] Step S42: Based on the node failure risk calculation model in step S41, through the component failure and disaster analysis module in the system failure and disaster analysis unit, a line failure risk calculation model is constructed with the failure probability of each line as the result.

[0184] Specifically, the failure probability of a single line without chain propagation is calculated as follows:

[0185] ;

[0186] Among them, w ij is from node v i to v j The dependency weight of For node v i The overall probability of damage, For node v j The overall probability of damage.

[0187] The failure probability of a multi-line chain propagation path is calculated as follows:

[0188] The chain propagation path is a set of directed paths P that propagate from the initial damaged node to the target node:

[0189] ;

[0190] Among them, P k Represents a specific directed path, the nodes are arranged in chronological order, 1≤k, k is an integer, represents the initial damaged node, represents the target node that is ultimately affected. Since chain propagation requires at least two chain lines, it is set to be 3≤m. k The cumulative risk is the product of the probability of damage along the path and the weight. The cumulative risk for each path is calculated as follows:

[0191] ;

[0192] in, represents the dependency weight of adjacent nodes in the path, represents the damage probability of each node in the path at time t.

[0193] Step S5: Based on the node failure risk calculation model established in step S41, the sensitivity of the node to the disaster factor is calculated through the component failure and disaster analysis module to obtain a comprehensive sensitivity factor.

[0194] The impact of power system component failure is the comprehensive impact of multiple disaster factors on the component, which is manifested as the sensitivity of the component to the disaster factor. The disaster factor includes the disaster intensity I, the impact range A, and the duration T. The probability of damage is , the sensitivity is defined as the partial derivative of these hazard factors with respect to the probability of damage:

[0195] ;

[0196] The sensitivity coefficient is then normalized:

[0197] ;

[0198] Then calculate the comprehensive sensitivity factor, which can be defined as a weighted sum: ;

[0199] Among them, W I , W A , W T is the weight of each disaster factor, satisfying W I +W A +W T =1.

[0200] Step S6: Set the disaster warning sensitivity factor S through the risk warning module in the intelligent decision-making and feedback adjustment unit. F (v i ) as the warning threshold, when the disaster factor S total (v i ) exceeds the disaster warning sensitivity factor S F (v i ), based on the multi-dimensional analysis results of the system fault analysis module at time t+1 in step S5 and the meteorological disaster prediction information at time t+1 obtained based on the meteorological information system, the data include disaster intensity I, impact range A, and duration T. The risk of power system failure is calculated by the risk calculation module in the intelligent decision-making and feedback adjustment unit, and the feedback information module provides risk feedback warning and failure risk category to system management personnel according to the system failure probability.

[0201] Specifically, the disaster warning sensitivity factor S F (v i ) is the average value of the historical comprehensive sensitivity factor of the system.

[0202] ;

[0203] ;

[0204] Among them, S avg (t) represents the average value of the comprehensive sensitivity factors of all nodes in the system at time t, It represents the comprehensive sensitivity factor of the node at time t, N is the number of component nodes in the system, and M is the total duration of the historical period.

[0205] Specifically, the calculation of the quantitative index of power system failure risk includes:

[0206] Step S61: Single node failure probability ;

[0207] ;

[0208] Probability of multiple nodes failing simultaneously ;

[0209] ;

[0210] in, is the set of damaged nodes.

[0211] Step S62: Single line failure probability :

[0212] ;

[0213] Multi-line failure probability :

[0214] ;

[0215] in, is the set of fault lines.

[0216] Step S63: Node and line combination failure probability :

[0217] ;

[0218] in, is the set of damaged nodes, is the set of fault lines, P k Represents a specific directed path.

[0219] Step S64: The comprehensive risk value of the system (system failure probability) is expressed as: P E_total ;

[0220] ;

[0221] in, Indicates the occurrence of a single node failure, multiple node failures, a single line failure, multiple line failures, or a combination of node and line failures. The value is 0 or 1.

[0222] In this embodiment, by setting thresholds, the power system failure risks are divided into the following three categories:

[0223] High risk failure: When , it indicates that the component is very likely to fail in a disaster and needs to be handled as a priority.

[0224] Medium risk failure: , it indicates that there is a risk of component failure, but it is controllable.

[0225] Low risk failure: When , it indicates that the component is safe under the current conditions.

[0226] in, and Represents risk classification threshold one and risk classification threshold two, which are adjusted based on historical data and expert experience.

[0227] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for diagnosing power system fault risks, characterized in that: The steps include: Step S1: Real-time multi-source data is collected and pre-processed to obtain a disaster data information set during the period when the system failure occurs, including meteorological data information and equipment monitoring data, and a system network topology map before and after the disaster is constructed, as well as disaster data information during the period when the system failure occurs; Step S2: Identify changes in node and line status based on the system power grid topology before and after the disaster, obtain the status of each node and each line, and annotate the status of each node and the line edge attributes to obtain an annotation result, obtain an annotation result set, and divide it into an annotation result training set and an annotation result test set; Step S3: construct a spatiotemporal dynamic model, train the spatiotemporal dynamic model based on the labeled result training set, and test the trained spatiotemporal dynamic model based on the standard result test set, and save it as a new spatiotemporal dynamic model after passing the test; Step S4: Based on the new spatiotemporal dynamic model obtained in step S3, the states of each node in the system operation and the states of each line at the current moment are taken as input, and the state vector of each node is used to perform a multi-dimensional analysis of the system failure at the next moment; Step S41: Based on the disaster data information set of the system failure period collected in step S1, a node failure risk calculation model is constructed with the failure probability of each node as the result; Step S42: Based on the node failure risk calculation model in step S41, a line failure risk calculation model is constructed with the failure probability of each line as the result; Step S5: Based on the node failure risk calculation model established in step S41, calculate the sensitivity of the node to the disaster factor to obtain a comprehensive sensitivity factor; Step S6: Establish a disaster warning sensitivity factor. When the size of the comprehensive sensitivity factor exceeds the disaster warning sensitivity factor, the power system failure risk quantitative index is calculated based on the node failure risk calculation model, the line failure risk calculation model and the meteorological information system's meteorological disaster forecast information for the next moment to obtain the system failure probability, and the system management personnel are given risk feedback warning and failure size according to the size of the system failure probability.

2. A method for diagnosing power system fault risks according to claim 1, characterized in that: The implementation of step S2 includes the following specific steps: Step S21: extracting the disaster conditions of lines and nodes through the power grid topology diagram before and after the disaster, and identifying the changes in the states of nodes and lines to obtain the results of the states of each node and each line; Step S211: Calculate the difference between the adjacency matrix before the disaster and the adjacency matrix after the disaster, identify the line state change, and analyze and extract its state; Step S212: Calculate the node state vector before the disaster and the node state vector after the disaster, identify the node state change, and analyze and extract its state; Step S213: Obtain the dependency strength between nodes by calculating the correlation of node state changes; Step S22: labeling the node status and line edge attributes according to the obtained results to obtain labeling results; Step S221: marking the status of each node according to the node status analysis and judgment result; Step S222: marking each line status according to the line status analysis and judgment result; Step S23: Obtain a labeling result set based on the labeling results of step S221 and step S222, and divide it into a labeling result training set and a labeling result test set for storage.

3. A method for diagnosing power system fault risks according to claim 2, characterized in that: The implementation of step S3 includes the following specific steps: Step S31: Build the network structure of SDM-GNN, generate model structure programming to obtain the spatiotemporal dynamic model; Step S32: Establishing a state transfer function based on time sequence; Step S33: performing spatial feature extraction to learn the node state change relationship in the topological structure; Step S34: extracting time features and using a time series model to capture the state evolution of nodes over time; Step S35: updating the fusion features to effectively combine the spatial features and the temporal features to capture the spatiotemporal dynamic information; Step S36: Based on the labeled result set obtained in step S2, the spatiotemporal dynamic model is trained through the training set, the node state prediction error is predicted according to the node state error between the iteration result and the actual value, and the model weight is updated until the node state prediction error value is minimized. The training cycle ends, and the trained spatiotemporal dynamic model is tested based on the standard result test set. After passing the test, it is saved as a new spatiotemporal dynamic model.

4. A method for diagnosing power system fault risks according to claim 1, characterized in that: During the implementation of step S4, after a multi-dimensional analysis of the system faults, single node faults, multi-node simultaneous faults, single line faults, multi-line faults and node and line combination faults are diagnosed.

5. A method for diagnosing power system fault risks according to claim 4, characterized in that: In the implementation process of step S41, the process of calculating the probability of node damage based on the node failure risk calculation model includes the following steps: Step S411: based on the meteorological data information and the disaster data information, characteristic parameters are obtained: disaster intensity, impact range and duration, and the characteristic parameters are standardized; Step S412: Perform statistical analysis on characteristic parameters to construct typical characteristic distribution functions of various disasters; Step S413: constructing a joint distribution function of disaster characteristics based on typical characteristic distribution functions of various disasters; Step S414: Establish a comprehensive damage model of the joint distribution function fusion components evolving over time based on disaster characteristics, and calculate and obtain the comprehensive damage probability of the node.

6. A method for diagnosing power system fault risks according to claim 5, characterized in that: During the implementation of step S414, when calculating the damage probability of the node, a time-dependent influencing factor is introduced into the damage probability to describe the impact of the evolution of node vulnerability over time on the damage probability, and a comprehensive damage model for the node is constructed by combining the disaster intensity, impact range, duration, time and node aging to obtain the comprehensive damage probability of the node.

7. A method for diagnosing power system fault risks according to claim 6, characterized in that: During the implementation of step S42, the damage probability of each line is calculated based on the line fault risk calculation model, including the calculation of the failure probability of a single line without chain propagation and the calculation of the failure probability of a multi-line chain propagation path.

8. A method for diagnosing power system fault risks according to claim 7, characterized in that: In the process of implementing step S6, the calculation of the quantitative index of power system fault risk includes the following steps: Step S61: Calculate and obtain the single node failure probability, and calculate and obtain the probability of multiple nodes failing simultaneously based on the single node failure probability; Step S62: Calculate and obtain the single line failure probability, and calculate and obtain the multi-line failure probability based on the single line failure probability; Step S63: Calculate and obtain the node and line combined failure probability based on the multi-node simultaneous failure probability and the multi-line failure probability; Step S64: Based on the results calculated in steps S61, S62 and S63, the comprehensive risk value of the system is calculated to obtain the system failure probability.

9. A method for diagnosing power system fault risks according to claim 8, characterized in that: In the process of implementing step S6, two risk classification thresholds adjusted according to historical data and expert experience are established, namely, risk classification threshold 1 and risk classification threshold 2, wherein risk classification threshold 1 is greater than risk classification threshold 2; Power system failure risks are divided into the following three categories: High-risk failure: When the probability of system failure is greater than the risk classification threshold of one, it means that the component is very likely to fail in a disaster and needs to be handled with priority; Medium risk failure: When the system failure probability is greater than or equal to the risk classification threshold 2 and less than or equal to the risk classification threshold 1, it means that the component has a failure risk, but it is controllable; Low risk failure: When the probability of system failure is less than the risk classification threshold of two, it means that the component is safe under the current conditions.

10. A power system fault risk diagnosis system, used to implement a power system fault risk diagnosis method as claimed in any one of claims 1 to 9, characterized in that: include: Data acquisition and processing unit, used for real-time acquisition of multi-source data and pre-processing; The component and line analysis unit is used to identify the status of each node and each line according to the preprocessing results, mark the status of each node and the line edge attributes, and divide the marked results; The component and line state prediction model building unit is used to construct a spatiotemporal dynamic model and train and test it according to the division results of the annotation results; The system failure and disaster analysis unit is used to perform multi-dimensional analysis on the system failure at the next moment according to the tested spatiotemporal dynamic model, and calculate the sensitivity of the node to the disaster factor to obtain the comprehensive sensitivity factor; Intelligent decision-making and feedback adjustment unit, through which a disaster warning sensitivity factor is established. When the size of the comprehensive sensitivity factor exceeds the disaster warning sensitivity factor, the system failure probability is calculated based on the multi-dimensional analysis results of the system failure at the next moment and the meteorological disaster forecast information at the next moment, and a risk feedback warning is carried out.

Images