Search

CN-121998416-A - Smart grid state monitoring method and system based on digital twinning

CN121998416ACN 121998416 ACN121998416 ACN 121998416ACN-121998416-A

Abstract

The invention discloses a smart grid state monitoring method and a smart grid state monitoring system based on digital twinning, which are characterized in that firstly, a scene tag is generated based on a semi-supervised learning model according to collected power grid data, and a model template matched with the scene tag is generated according to the scene tag and a semi-supervised learning model template library; then constructing a power grid digital twin model based on the model template, generating an abnormal path set based on the power grid digital twin model, calculating independent risk values of single power grid equipment based on the abnormal path set, and finally generating a smart grid comprehensive risk value according to the independent risk values of the power grid equipment. The invention ensures the seamless transition from equipment-level risk to system-level risk, and provides a visual and reliable decision support for intelligent power grid state monitoring.

Inventors

  • JIANG YUE
  • XIONG WEI

Assignees

  • 湖北工业大学

Dates

Publication Date
20260508
Application Date
20260114

Claims (10)

  1. 1. The intelligent power grid state monitoring method based on digital twinning is characterized by comprising the following steps of: step 1, generating a scene tag based on a semi-supervised learning model according to acquired power grid data, and generating a model template matched with the scene tag according to the scene tag and a semi-supervised learning model template library; Step 2, constructing a power grid digital twin model based on the model template, and generating an abnormal path set based on the power grid digital twin model; step 3, calculating independent risk values of single power grid equipment based on the abnormal path set; and 4, generating a comprehensive risk value of the intelligent power grid according to the independent risk value of each power grid device.
  2. 2. The method for monitoring the state of the intelligent power grid based on the digital twin system according to claim 1, wherein in the step 1, the semi-supervised learning model is composed of an input processing module, a feature extraction module, a dynamic threshold self-training module, a prototype consistency constraint module and an output layer; the input processing module is in charge of receiving the scene feature vectors and generating diversified input views by applying standardized and data enhancement technologies; the feature extraction module is used for mapping input data to a high-dimensional feature space based on a deep neural network architecture (such as a convolutional neural network) and extracting semantic features, and generally comprises a plurality of convolutional layers and pooled layers for capturing local and global modes; the dynamic threshold self-training module integrates STDCT an algorithm and comprises a confidence coefficient calculator and a threshold value regulator, wherein the confidence coefficient calculator evaluates the prediction certainty of the unlabeled sample, and the threshold value regulator dynamically updates the confidence coefficient threshold value to adapt to the current training state; the prototype consistency constraint module applies consistency constraint by calculating the similarity between the unlabeled sample and the prototype, so as to ensure the prediction reliability; The output layer generates probability distribution of each category by using the full connection layer and the Softmax activation function, and obtains scene labels through argmax operation Wherein S is an output scene tag; The model predictive feature vector fs is the probability of belonging to the category K, and K is the preset category number.
  3. 3. The intelligent power grid state monitoring method based on digital twinning according to claim 1, wherein in step 1, the scene label is compared with a label set in a semi-supervised learning model template library to judge whether a template matched with the scene label exists or not; if yes, calling a corresponding template from the semi-supervised learning model template library to serve as a model template of the scene tag; if not, mapping the scene feature vector into a high-dimensional coding representation based on the semi-supervised learning model, calculating the similarity between the high-dimensional coding representation and each template prototype vector in the semi-supervised learning model template library, selecting a model template with the maximum similarity as a transition template, generating a temporary template based on the transition template, and taking the temporary template as a model template matched with the scene label.
  4. 4. The intelligent power grid state monitoring method based on digital twin of claim 1 is characterized by comprising the steps of constructing a state vector set based on the power grid digital twin model in step 2, generating causal influence intensity between equipment nodes according to the state vector set and a preset causal scoring function, generating a directed causal graph, identifying abnormal nodes, reversely tracking influence paths from each abnormal node in the directed causal graph, calculating the abnormal influence intensity of each influence path, and generating an abnormal path set according to the abnormal influence intensity.
  5. 5. The method for monitoring the state of a smart grid based on digital twinning as set forth in claim 4, wherein the causal scoring function is: ; Wherein, the Representing nodes For a pair of Is a causal influence on intensity; is a real-time state variable And Mutual information between the two is used for measuring nonlinear dependency relationship and is obtained through multi-time state sequence estimation in a history sliding window; is the ratio of the current equipment temperature rise state to the upper operating limit thereof, Is a node The allowable highest temperature rise value of the equipment specification or the historical safety threshold value; L 1 is the distance of the state difference and represents the degree of the deviation of the node state from the stable coupling relation; Is a node Is statistically obtained from the long-term operation data; Is an extremely small constant for avoiding zero denominator; 、 、 Is a weight factor, and controls the balance of mutual information intensity, temperature rise risk sensitivity and state deviation penalty.
  6. 6. The intelligent power grid state monitoring method based on digital twin according to claim 1, wherein in step 3, for each power grid device, screening all single abnormal paths containing the power grid device from the abnormal path set; the abnormal path set includes a path list, device nodes to which the paths relate, and an abnormal influence strength eta (p) value for each path, Wherein, the method comprises the steps of, Representing the product of all causal edge weights in path p; Is causal edge weight, which represents the influence intensity of the node u on the node v, j is the index of the end point node; is a path end point node Reading the current state value of the data from the real-time monitoring data; And Respectively, are path end nodes The historical state mean value and standard deviation of the model (1) are obtained by statistics from long-term operation data and are used for normalizing the degree of abnormality; for each device node Weighted risk value Wherein p is a single abnormal path; Is the device index; is all inclusive of devices Is to screen out the abnormal path set satisfying A set of paths p of e p; is an apparatus Position weight in path p for adjusting contribution of η (p), position weight According to the role definition of the device in the path, the definition rules are as follows: If the device Is the end point of path p, then the position weight =1, Representing full contribution; If the device Is the start of path p, then the position weight =Γ, γ is a starting point weight factor, 0< γ <1, secondary to the source risk; If the device Is the middle point of path p, then the position weights =Δ, δ is the intermediate point weight factor, 0< δ < γ reflecting the weak contribution of the transfer risk.
  7. 7. The method for monitoring the state of the intelligent power grid based on the digital twin system of claim 1, wherein in step 4, the independent risk value of each power grid device is mapped into a discrete risk state; The optimized Bayesian network optimizes the structure and parameters of the Bayesian network by introducing an association rule data mining technology; association rule mining extracts strong dependency relationship among devices from historical data, and calculates support degree of rules through Apriori algorithm Confidence level Wherein, the method comprises the steps of, Is a rule Representing the frequency of simultaneous occurrence of events X and Y; is an association rule representing the dependency relationship between events X and Y; the number of times that events X and Y occur simultaneously in the historical data, M being the total data amount; Is a rule The confidence of (2) represents the probability of occurrence of event Y under the occurrence condition of event X; The method comprises the steps of generating a probability table of a frequent item set guiding condition, wherein the probability table is initialized and then optimized through Bayesian learning, the Bayesian learning updates parameters through posterior probability, and the formula is as follows: wherein, the method comprises the steps of, Is a posterior probability representing the parameters under the condition of the observed data D Probability distribution of (2); is a likelihood function expressed in parameters Probability of observing data D down; is the parameter to be estimated, D is training data, i.e. history equipment state and fault record, P # ) Is a priori distribution of parameters, representing parameters P (D) is the edge probability, which represents the total probability of the data D, and performs normalization.
  8. 8. The method for monitoring the state of a smart grid based on digital twinning as set forth in claim 1, wherein in step 4, risk fusion calculation is performed on the basis of an optimized Bayesian network, and independent risk values are obtained As an evidence variable input network, calculating a comprehensive risk value of the intelligent power grid through Bayesian reasoning, wherein the Bayesian reasoning updates node probability based on the Bayesian theorem: Wherein N is the total number of devices, An independent risk value representing a kth device; The comprehensive risk value, namely posterior probability, of the intelligent power grid represents the probability of occurrence of an event A under the condition of observation evidence B; Is likelihood probability representing the probability of observing evidence B under the condition of system fault occurrence, is derived from the conditional probability table, A represents the system comprehensive risk event, B is the observed equipment risk evidence set, namely the equipment independent risk value The discrete state after mapping, P (A) is the prior probability, representing the initial probability of event A, obtained from the statistics of historical system fault data, P (B) is the edge probability of evidence B, calculated as The reasoning output is posterior probability I.e., the smart grid integrated risk value that quantifies the overall risk level of the system.
  9. 9. The method for monitoring the state of a smart grid based on digital twinning according to any one of claims 1 to 8, wherein the semi-supervised learning model is a trained model, and the training process is as follows: Firstly, carrying out data preprocessing, loading a data set with a label and a non-label, applying normalization processing and data enhancement processing, and dividing training batches; And then initializing a model, randomly initializing parameters of a feature extraction network and an output layer, and initializing a prototype vector based on a feature mean value of a labeled sample, wherein the prototype vector is expressed as: ; Wherein, the Refers to prototype vectors of class k, h Is a sample Is characterized by; is a collection of samples belonging to class k; Then training circulation is carried out, wherein the training circulation comprises forward propagation, loss calculation and backward propagation and optimization, the forward propagation is to input batch data, the characteristics are obtained through a characteristic extraction module, then predictions are generated through an output layer, the loss calculation is combined with supervised loss, unsupervised loss and prototype consistency loss, the backward propagation and optimization uses gradient descent algorithm to update parameters, the total loss is minimized, and the optimization process is that , wherein, Representing the state of the model parameters at the current t-th iteration; Representing the state of the model parameters at the next, i.e. t+1st iteration; Is the learning rate; Is the total loss function with respect to all model parameters Is a gradient of (2); after each training round, recalculating a prototype vector according to the characteristics of the current labeled sample, and ensuring that the prototype vector represents the latest class distribution; finally, convergence check is carried out, and training cycle is repeated until the loss function converges or the maximum iteration number is reached.
  10. 10. A digital twinning-based smart grid state monitoring system, comprising: One or more processors; Storage means for storing one or more programs that, when executed by the one or more processors, cause the one or more processors to implement the digital twinning-based smart grid state monitoring method of any one of claims 1 to 9.

Description

Smart grid state monitoring method and system based on digital twinning Technical Field The application belongs to the technical field of power grid monitoring, relates to a smart power grid state monitoring method and system, and particularly relates to a digital twinning-based smart power grid state monitoring method and system. Background Along with the continuous expansion of the scale of the intelligent power grid and the continuous improvement of the operation complexity, the real-time state monitoring and risk early warning of the power grid equipment become core requirements for guaranteeing the safe and stable operation of the power system. The digital twin technology is widely applied to the intelligent power grid state monitoring field by virtue of the accurate mapping and simulation capability of the digital twin technology on a physical system, and builds a virtual model synchronous with the physical power grid by fusing multi-source information such as meteorological data, load data, equipment state data and the like, so as to provide support for operation state analysis and abnormality early warning. In the process, scene recognition is performed based on the multi-source data of the power grid and the corresponding digital twin model templates are matched, which is a key premise for realizing efficient monitoring, and the core is that a simulation model which is attached to the actual running conditions is quickly constructed through accurate butt joint of scene labels and a template library. According to a traditional intelligent power grid state monitoring method based on digital twinning, firstly, multi-source data in a power grid operation process are collected, corresponding scene labels are obtained through data processing and feature extraction, then the scene labels are compared with a model template library constructed in advance one by one, then corresponding model templates in the library are directly called, construction of a digital twinning model is completed based on the templates, and further follow-up state simulation and monitoring work is carried out. However, the traditional method only depends on a predefined fixed template in a template library, when a scene tag acquired through multi-source data belongs to a novel operation scene, an adaptive model template cannot be generated based on unique features of the current scene, so that effective basis is lacking in construction of a digital twin model, actual operation states of a power grid are difficult to accurately reflect, and finally, the adaptability of intelligent power grid state monitoring to complex and changeable operation scenes is limited, and comprehensive and accurate state coverage and risk early warning cannot be realized. Disclosure of Invention In order to solve the technical problems, the invention provides a digital twin-based intelligent power grid state monitoring method and system, which are used for optimizing a Bayesian network by combining association rule mining, fusing equipment risks to obtain a comprehensive risk value, improving scene recognition robustness and risk assessment accuracy, and effectively supporting intelligent power grid state monitoring and operation and maintenance decision. The technical scheme adopted by the method is that the intelligent power grid state monitoring method based on digital twinning comprises the following steps: step 1, generating a scene tag based on a semi-supervised learning model according to acquired power grid data, and generating a model template matched with the scene tag according to the scene tag and a semi-supervised learning model template library; Step 2, constructing a power grid digital twin model based on the model template, and generating an abnormal path set based on the power grid digital twin model; step 3, calculating independent risk values of single power grid equipment based on the abnormal path set; and 4, generating a comprehensive risk value of the intelligent power grid according to the independent risk value of each power grid device. Preferably, in step 1, the semi-supervised learning model is composed of an input processing module, a feature extraction module, a dynamic threshold self-training module, a prototype consistency constraint module and an output layer; The device comprises an input processing module, a feature extraction module, a dynamic threshold self-training module and a threshold self-training module, wherein the input processing module is in charge of receiving scene feature vectors and generating diversified input views by applying standardized and data enhancement technologies, the feature extraction module is used for mapping input data to a high-dimensional feature space based on a deep neural network and extracting semantic features, and the dynamic threshold self-training module is integrated with STDCT algorithms and comprises a confidence coefficient calculator and a threshold adjuster. The confidence