Box-type substation panoramic monitoring operation and maintenance system and method fused with edge calculation
By constructing a dynamic causal model on the edge computing unit of the prefabricated substation, and combining panoramic monitoring data acquisition and intelligent analysis, the problems of poor interpretability and insufficient prediction accuracy in fault diagnosis in existing technologies are solved. This achieves accuracy in fault root cause location and equipment status prediction, and improves operation and maintenance efficiency and autonomy.
Patent Information
- Application Number
- CN202511176259.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-21
- Publication Date
- 2025-11-07
AI Technical Summary
Existing monitoring methods for prefabricated substations rely on data-driven black-box models, resulting in poor interpretability of fault diagnosis, insufficient accuracy of trend prediction, poor real-time system response, and difficulty in achieving accurate assessment and effective prediction of equipment status.
The panoramic monitoring and maintenance system adopts edge computing to build a dynamic causal model, use real-time operating data for status updates, and perform intelligent operation and maintenance analysis on the edge computing unit, including fault root cause diagnosis and fault evolution trend prediction. It combines target detection and character recognition technologies to obtain structured data and realize on-site closed-loop data processing.
It enables precise location of fault roots and accurate prediction of future change trajectories, improves the credibility of diagnostic conclusions and the accuracy of predictions, transforms passive response-based operation and maintenance into proactive prevention-based operation and maintenance, reduces dependence on cloud resources and network latency, and supports low-cost intelligent transformation.
Smart Images
Figure CN120914993A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of smart grid and electrical equipment state monitoring, in particular to a panoramic monitoring operation and maintenance system and method for box-type substations based on edge computing. BACKGROUND
[0002] The box-type substation is a key node device in the power distribution network, and its stable and reliable operation is directly related to the safety and quality of regional power supply. Therefore, effective and accurate monitoring and operation and maintenance of the box-type substation is an important link to ensure the stable operation of the power grid.
[0003] The current monitoring and operation and maintenance technology for box-type substations mainly relies on periodic manual inspection and remote automatic monitoring systems. The manual inspection method not only has high labor intensity and low efficiency, but also is easily affected by the subjective experience and responsibility of personnel, making it difficult to achieve continuous and quantitative evaluation of the equipment state. The existing automatic monitoring systems usually collect part of the electrical quantities and environmental quantities of key devices such as transformers and switch cabinets through the installation of sensors, and transmit the data to the master station or cloud platform for centralized analysis. Some more advanced systems introduce video monitoring and data-driven analysis models, such as using machine learning algorithms to train massive historical data to achieve anomaly detection or fault warning.
[0004] However, the above-mentioned existing technology still has disadvantages in practical application. On the one hand, the data perception dimension is relatively single, mainly relying on pre-set electrical quantity and environmental quantity sensors, and the state information of a large number of stock equipment that has not been intelligently transformed cannot be effectively obtained, resulting in information blind area in the portrait construction of the overall operation state of the box-type substation. On the other hand, the existing analysis methods are mostly "black box" models based on data statistical correlation. Although such models can identify data anomalies to some extent, they are difficult to reveal the deep physical causal logic behind the fault. When an alarm occurs, the system often cannot clearly explain the origin of the fault and its conduction path in the device, resulting in insufficient credibility and explainability of the diagnosis result, making it difficult for operation and maintenance personnel to develop accurate troubleshooting strategies. At the same time, the existing technology also has deficiencies in fault prediction. The prediction logic is usually a simple trend extrapolation of a single or a few key parameters, ignoring the internal mechanism of mutual coupling and dynamic evolution of various physical states in the box-type substation as a complex system. This prediction method is difficult to accurately simulate the evolution process of a specific fault in the system, thus making it impossible to make reliable assessment of the future failure risk and remaining life of the equipment, and making the operation and maintenance strategy still biased towards passive response or fixed periodic maintenance, thereby reducing the practicality of the system. SUMMARY
[0005] In view of the deficiencies of the prior art, the panoramic monitoring operation and maintenance system and method of the box-type substation fusing edge computing are provided, which solves the problems of poor fault diagnosis explainability, insufficient trend prediction accuracy and poor system response real-time performance of the existing box-type substation monitoring method due to the dependence on the data-driven black box model and the centralized processing architecture.
[0006] To achieve the above object, the panoramic monitoring operation and maintenance system of the box-type substation fusing edge computing is implemented by the following technical solutions: the panoramic monitoring operation and maintenance system of the box-type substation fusing edge computing comprises a panoramic monitoring data acquisition device configured to acquire real-time operation data of the box-type substation; an edge computing unit connected with the panoramic monitoring data acquisition device; a processing module running on the edge computing unit and configured to: continuously utilize the real-time operation data to drive a dynamic causal model preset on the edge computing unit to update the state in real time, wherein the dynamic causal model comprises a plurality of state nodes and a plurality of causal edges, and the state update specifically refers to real-time updating of the values of each state node; based on the global state of the dynamic causal model after the state update, when a preset trigger condition is monitored, automatically performing corresponding intelligent operation and maintenance analysis; when the trigger condition is determined as an abnormal value of at least one state node, performing fault root cause diagnosis by means of reverse propagation tracing along the causal edges; and when the trigger condition is determined as a deviation of the evolution trend of the value of at least one state node, performing fault evolution trend prediction by means of forward iterative evolution based on the mathematical definition of the causal edges.
[0007] Preferably, the dynamic causal model is characterized in that: the processing module is further configured to perform data fusion and feature extraction on the real-time operation data before performing the state update, so as to uniformly convert the heterogeneous real-time operation data from different sensors into structured state values corresponding to each state node in the dynamic causal model.
[0008] Preferably, the specific manner of the processing module performing the feature extraction comprises: when the real-time operation data is image data acquired by the panoramic imaging unit, the processing module is configured to first locate the display area of an instrument or a meter in the image data through a target detection model, then identify the numerical value from the located display area through a character recognition model, and finally use the identified numerical value as the structured state value to update the corresponding state node.
[0009] Preferably, the processing module is further configured to: propagating the initial cause with the highest confidence in the fault root cause chain as a preset fault, locking the fault state in a temporary virtual dynamic causal model; and calling the fault evolution trend prediction to deduce the future failure time and consequences of the system under the preset fault, for generating an active operation and maintenance strategy.
[0010] Preferably, the specific manner in which the processing module executes the fault evolution trend prediction comprises: based on the global state evolution equation determined by the overall topology of the dynamic causal model, taking the current global state as an initial condition, iteratively solving the state vectors at multiple future time steps to form the future change trajectory.
[0011] Preferably, the processing module is further configured to: calculate a prediction error by comparing the result of the fault evolution trend prediction with subsequent real-time operation data; and based on the prediction error, adaptively adjust the model parameters within the transfer function operators through an online optimization algorithm to realize self-correction of the dynamic causal model.
[0012] Preferably, the processing module realizes self-correction of the dynamic causal model through an online optimization algorithm, the online optimization algorithm being an extended Kalman filter algorithm, and the processing module is configured to augment the model parameter vector encapsulated in all transfer function operators in the dynamic causal model with the global state vector of the dynamic causal model to jointly form an augmented state vector; and based on the augmented state vector, iteratively optimize the model parameter vector by continuously performing a prediction step and an update step; wherein the update step uses the prediction error to calculate a Kalman gain and uses the gain to correct the a priori estimate value generated in the prediction step, thereby realizing simultaneous optimization and adaptive adjustment of the model parameter vector while continuously correcting the global state vector.
[0013] Preferably, the processing module realizes self-evolution of the dynamic causal model, and the processing module is configured to continuously monitor and analyze the distribution of the prediction error over the topology of the dynamic causal model, and when it is identified that the prediction error is long-term and significantly concentrated in a certain high-error region, it is determined that there may be a mismatch between the physical reality and the model structure in the region, and a new causal relationship mining is automatically triggered; The new causal relationship mining is realized by calculating information entropy to quantify the potential information flow strength between the state nodes in the high-error region and other state nodes, and the information flow strength is calculated through the following transfer entropy formula: wherein represents the information flow strength from the source state node to the target state node ; represents the state of the target state node at a future time instant ; represents the state of the target state node at a future time instant ; represents the information flow strength from the source state node to the target state node ; represents the state of the target state node at a future time instant ; represents a historical state sequence consisting of the state of the target state node at the current time instant and its preceding history time instants; represents a historical state sequence consisting of the state of the source state node at the current time instant and its preceding history time instants; represents the probability distribution of the corresponding variable or variable combination; when it is calculated that there exists an information flow strength from a source state node originally connected by a non-causal edge to a target state node within the high-error region, and the strength continuously exceeds a preset structural correlation threshold, the processing module generates a structural reconstruction suggestion for the dynamic causal model, which prompts the operation and maintenance personnel that there may exist an undefined, new physical causal path in the historical state sequence consisting of the state of the target state node at the current time instant and its preceding history time instants; represents a historical state sequence consisting of the state of the source state node at the current time instant and its preceding history time instants; a probability distribution representing a corresponding variable or variable combination; when it is calculated that there is an information flow strength from a source state node originally without a causal edge connection to a target state node within a high error area, and the strength continuously exceeds a preset structural correlation threshold, the processing module generates a structural reconstruction suggestion for the dynamic causal model, which prompts the operation and maintenance personnel that there may be a new physical causal path that has not been defined.
[0014] Preferably, the processing module is configured to generate a structured proactive operation and maintenance strategy based on the fault root cause chain and the deduced consequences and failure times, the proactive operation and maintenance strategy including a diagnosis conclusion, a predicted consequence, and a maintenance urgency level.
[0015] The box-type substation panoramic monitoring operation and maintenance method based on edge computing includes the following steps: S1. Constructing and maintaining a dynamic causal model on the edge computing unit, the model including a plurality of state nodes representing key physical states of the box-type substation, and a plurality of causal edges for quantifying the dynamic causal influence between the state nodes; S2. Acquiring real-time operation data of the box-type substation through a panoramic monitoring data acquisition device, and updating the values of the state nodes in the dynamic causal model in real time using the real-time operation data to form a global state vector ; S3. Then, based on the updated dynamic causal model, when a preset trigger condition is monitored, performing corresponding intelligent operation and maintenance analysis, the intelligent operation and maintenance analysis including fault root cause diagnosis, fault evolution trend prediction, and online self-evolution of the dynamic causal model, the fault evolution trend prediction being calculated by solving the following global state evolution equation, taking the current global state vector as the initial condition, to iteratively calculate the states at a series of future time steps, thereby forming a future change trajectory: wherein, is the global state vector at the next time, is a system function determined by the overall topological structure of the dynamic causal model is an external input vector, is a model parameter vector containing all internal parameters of the causal edges, and the online self-evolution of the dynamic causal model is to continuously compare the predicted state of the fault evolution trend prediction with the subsequently acquired real-time operation data to calculate a prediction error, and when it is identified that the prediction error is long-term concentrated in a certain high error area, triggering new causal relationship mining, which quantifies the potential information flow strength between the target state node within the high error area and other source state nodes in the graph by calculating the following transfer entropy formula: wherein, represents the information flow strength from the source state node to the target state node , represents the state of the target state node at a future time instant , represents the history state sequence of the target state node , represents the history state sequence of the source state node , represents a probability distribution; if the information flow strength continuously exceeds a preset threshold, a structural reconstruction suggestion for the dynamic causal model is generated.
[0016] The application provides a box-type substation panoramic monitoring operation and maintenance system and method fusing edge computing. The following beneficial effects are achieved: 1. The application constructs and runs a dynamic causal model based on physical mechanism, and when an abnormal state is monitored, the root cause of the fault is accurately located by backtracking along the causal edges defined in the model. Compared with the traditional "black box" diagnosis method relying on data correlation, the diagnosis result of the application is in the form of a fault root cause propagation chain, has clear physical meaning and logical interpretability, and greatly improves the credibility of the diagnosis conclusion.
[0017] 2. The application performs forward iterative evolution based on the overall topology structure of the dynamic causal model to accurately predict the future change trajectory of the device state, and can deduce the failure evolution trend by taking the diagnosed root cause as a preset fault, estimate the future failure time and potential consequences, and thus change the traditional passive response operation and maintenance into active preventive operation and maintenance, providing a solid decision basis for formulating a forward-looking maintenance plan.
[0018] 3. The application introduces an online self-evolution mechanism, which on the one hand uses the extended Kalman filter algorithm to continuously adaptively correct the model parameters according to the error between the prediction result and the real data, and on the other hand analyzes the topological distribution of the prediction error and calculates the transfer entropy to mine potential and undefined causal relationships to optimize the model structure, so as to realize self-correction and self-evolution of the model and ensure long-term adaptability and high precision of the model in the whole life cycle of the device.
[0019] 4、The application can automatically read and extract data of a large number of non-intelligent analog instruments in the box-type substation by adopting the panoramic monitoring data acquisition device, especially the visual analysis technology combining target detection and character recognition model, which realizes low-cost and non-invasive intelligent transformation of the inventory equipment, without the need of large-scale hardware replacement of the original equipment, and can obtain more comprehensive and fine-grained operation data.
[0020] 5、The application realizes on-site closed-loop processing and real-time intelligent decision of data by constructing, updating and deploying all intelligent operation and maintenance analysis of the dynamic causal model on the edge computing unit of the box-type substation, which significantly reduces the dependence on cloud computing resources and the occupation of network bandwidth, avoids the delay and cost problems caused by the transmission of massive raw data, and improves the efficiency and autonomy of field operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS
[0021] Figure 1 is a main framework diagram of the application; Figure 2 is a local architecture diagram of the application; Figure 3 is a causal state evolution architecture diagram of the application; Figure 4 is an operation core processing module architecture diagram of the application. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the accompanying drawings of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.
[0023] Please refer to the accompanying drawings of the application Figure 1 - the accompanying drawings of the application Figure 4The embodiment of the present application provides a box-type transformer substation panoramic monitoring operation and maintenance system and method fusing edge calculation, which comprises a panoramic monitoring data acquisition device configured to acquire real-time operation data of the box-type transformer substation; an edge calculation unit connected with the panoramic monitoring data acquisition device; and a processing module running on the edge calculation unit and configured to continuously utilize the real-time operation data to drive a dynamic causal model preset on the edge calculation unit to update a state in real time, wherein the dynamic causal model comprises a plurality of state nodes and a plurality of causal edges, and the state updating specifically refers to updating the value of each state node in real time; based on the global state of the dynamic causal model after the state updating, when a preset trigger condition is monitored, corresponding intelligent operation and maintenance analysis is automatically performed; when the trigger condition is determined as an abnormal value of at least one state node, fault root cause diagnosis is performed in a reverse propagation tracing mode along the causal edges; and when the trigger condition is determined as an evolution trend deviating from the value of at least one state node, fault evolution trend prediction is performed in a forward iterative evolution mode based on the mathematical definition of the causal edges. Specifically, a camera or a plurality of spliced wide-angle cameras arranged at the top end of the box-type transformer are used to acquire panoramic image or video data of internal equipment layout, instrument reading, indicator light state and external environment of the box-type transformer, then a plurality of non-visual sensors are used to monitor the electrical quantity sensors of three-phase current, voltage and partial discharge signal of the transformer, after the original data are acquired, the processing module running on the edge calculation unit performs data fusion and feature extraction, and through the above steps, the system can convert the multi-source heterogeneous field data into state information that can be understood by the model, unified and structured.
[0024] The processing module is further configured to perform data fusion and feature extraction on the real-time operation data before performing state updating, so as to uniformly convert heterogeneous real-time operation data from different sensors into structured state values corresponding to state nodes in the dynamic causal model. The processing module performs feature extraction in the following specific manner: when the real-time operation data is image data obtained by the panoramic imaging unit, the processing module is configured to first locate a display region of an instrument or a meter in the image data by a target detection model, then identify a numerical value from the located display region by a character recognition model, and finally use the identified numerical value as a structured state value to update a corresponding state node. The processing module is further configured to: take an initial cause with the highest confidence in the fault root cause propagation chain as a preset fault, and lock the fault state in a temporary virtual dynamic causal model; and call fault evolution trend prediction to deduce a future failure time and consequences of the system under the preset fault, for generating an active operation and maintenance strategy. The processing module performs fault evolution trend prediction in the following specific manner: based on a global state evolution equation determined by the overall topology of the dynamic causal model, the current global state is taken as an initial condition to iteratively solve state vectors at multiple future time steps, so as to form a future change trajectory. The processing module is further configured to: calculate a prediction error by comparing a result of the fault evolution trend prediction with subsequent real-time operation data; and based on the prediction error, adaptively adjust model parameters in the transfer function operator by an online optimization algorithm, so as to realize self-correction of the dynamic causal model. The processing module realizes self-correction of the dynamic causal model by the online optimization algorithm, which is an extended Kalman filter algorithm. The processing module is configured to augment a model parameter vector encapsulated in all transfer function operators in the dynamic causal model with a global state vector of the dynamic causal model to jointly form an augmented state vector; and based on the augmented state vector, iteratively optimize the model parameter vector by continuously performing a prediction step and an update step; wherein the update step calculates a Kalman gain using the prediction error, and corrects a priori estimation value generated in the prediction step using the gain, so as to constantly correct the global state vector and simultaneously realize synchronous optimization and adaptive adjustment of the model parameter vector. Specifically, the dynamic causal model is composed of state nodes, causal edges and transfer function operators. Each causal edge is associated with a transfer function operator for accurately quantifying the influence of the change of the state of the source node on the change of the state of the target node. After the processing module obtains the message, it uses the obtained structured state value to update the current value of the corresponding state node in the CSEG at each time step t. The current values of all state nodes jointly form a global state vector S_t, which is a “digital snapshot” of the health state of the box at time t and is the basis for all subsequent analyses.
[0025] The processing module implements self-evolution of the dynamic causal model. The processing module is configured to continuously monitor and analyze distribution of prediction errors on a topology of the dynamic causal model. When it is identified that the prediction errors are long-term and significantly concentrated in a high error area, it is determined that there is a possible mismatch between a physical reality and a model structure of the area, and a new causal relationship mining is automatically triggered; The new causal relationship mining is achieved by calculating information entropy to quantify potential information flow strength between state nodes in the high error area and other state nodes. The information flow strength is calculated by the following transfer entropy formula: Wherein represents information flow strength from a source state node to a target state node ; represents a state of the target state node at a future time point ; represents a state sequence of the target state node represents information flow strength from a source state node to a target state node ; represents a state of the target state node at a future time point ; represents a historical state sequence of the target state node at a current time point and before historical time points; represents a historical state sequence of the source state node at a current time point and before historical time points; represents a probability distribution of a corresponding variable or variable combination; when it is calculated that there is information flow strength from a source state node originally connected by no causal edge to a target state node in the high error area, and the strength is continuously higher than a preset structure correlation threshold, the processing module generates a structure reconstruction suggestion for the dynamic causal model. The suggestion prompts an operator that there is a possible new physical causal path that is not defined at a historical state sequence of the current time point and before historical time points; representative source state node at the current time and its previous history state sequence composed of states at the current time and its previous representative probability distribution of the corresponding variable or variable combination; when it is calculated that there is an information flow strength from a source state node originally connected by no causal edge to the target state node in the high error area, and the strength continuously exceeds the preset structural correlation threshold, the processing module generates a structural reconstruction suggestion for the dynamic causal model, which prompts the operation and maintenance personnel that there may be a new physical causal path that has not been defined, and the processing module is configured to generate a structured proactive operation and maintenance strategy based on the fault root cause chain and the deduced consequences and failure time, the proactive operation and maintenance strategy including diagnosis conclusion, predicted consequences and maintenance urgency level; Specifically, after completing state updating, the system does not blindly execute all analyses, but determines subsequent actions through an intelligent triggering mechanism, and when triggered, can quickly and accurately find the source of the anomaly, and the implementation is to perform back propagation and trace on the CSEG graph structure, at each step of the trace, the system calls the inverse evaluation function of the transfer function operator on the causal edge, which evaluates the confidence that the parent node is the cause of the anomaly based on Bayesian inference or inverse operation of the mechanism model under the condition that the child node anomaly is observed, and finally the system outputs one or more fault root cause propagation chains sorted in descending order of confidence, clearly revealing the complete conduction path from the fault source to the final anomaly phenomenon. Through prediction, the system can deduce how other parts of the system will deteriorate and the failure time range of the key safety threshold under the condition that the specific fault continues to exist.
[0026] The box-type substation panoramic monitoring operation and maintenance method fuses edge computing, and comprises the following steps: S1. Constructing and maintaining a dynamic causal model on the edge computing unit, the model comprising a plurality of state nodes representing key physical states of the box-type substation and a plurality of causal edges for quantifying the dynamic causal influence between the state nodes; S2. Acquiring real-time operation data of the box-type substation through a panoramic monitoring data acquisition device, and updating the values of the state nodes in the dynamic causal model in real time using the real-time operation data to form a global state vector ; S3. Then, based on the updated dynamic causal model, when a preset triggering condition is monitored, performing corresponding intelligent operation and maintenance analysis, the intelligent operation and maintenance analysis including fault root cause diagnosis, fault evolution trend prediction, and online self-evolution of the dynamic causal model, the fault evolution trend prediction being to solve the following global state evolution equation to predict the evolution trend of the current global state vector As initial conditions, the states of a series of future time steps are calculated through forward iteration, thus forming the future trajectory of change: in, It is the global state vector at the next time step. The system function is determined by the overall topology of the dynamic causal model. It is an external input vector. It is a model parameter vector containing all the internal parameters of the causal edges. The online self-evolution of the dynamic causal model is achieved by continuously comparing the predicted state of the fault evolution trend with the subsequently acquired real-time running data to calculate the prediction error. When the prediction error is identified as being concentrated in a certain high-error region for a long time, new causal relationship mining is triggered. This mining quantifies the potential information flow intensity between the target state node in the high-error region and other source state nodes in the graph by calculating the following transfer entropy formula: in, Represents the source state node To the target state node Information flow intensity, Represents the target state node In the future state, Represents the target state node Historical state sequence Represents the source state node Historical state sequence Represents a probability distribution; if the intensity of this information flow remains above a preset threshold, a structural reconstruction suggestion for the dynamic causal model is generated. In summary, this invention provides a panoramic monitoring and maintenance system and method for prefabricated substations that integrates edge computing. By calculating the transfer entropy between high-error nodes and all other nodes in the model, a dynamic causal model based on physical mechanisms is constructed and run. When an abnormal state is detected, backpropagation is performed along the causal edges defined in the model to accurately locate the root cause of the fault. Compared with the traditional "black box" diagnostic method that relies on data correlation, this invention can greatly improve the credibility of the diagnostic conclusions.
[0027] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A box-type substation panoramic monitoring operation and maintenance system integrated with edge computing, comprising a panoramic monitoring data acquisition device, characterized in that, configured to acquire real-time operation data of the box-type substation; an edge computing unit connected with the panoramic monitoring data acquisition device; a processing module running on the edge computing unit and configured to: continuously drive a preset dynamic causal model on the edge computing unit to update states in real time using the real-time operation data, wherein the dynamic causal model includes multiple state nodes and multiple causal edges, and the state update specifically refers to real-time updating of values of the state nodes; based on the global state of the dynamic causal model after state update, automatically perform corresponding intelligent operation and maintenance analysis when a preset trigger condition is monitored; when the trigger condition is determined as an abnormal value of at least one state node, perform fault root cause diagnosis by means of backtracking along the causal edges; and when the trigger condition is determined as a deviation of the evolution trend of the value of at least one state node, perform fault evolution trend prediction by means of forward iterative evolution based on the mathematical definition of the causal edges.
2. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 1, characterized in that, The dynamic causal model is characterized in that: Before performing the state update, the processing module is further configured to perform data fusion and feature extraction on the real-time operation data, so as to uniformly convert heterogeneous real-time operation data from different sensors into structured state values corresponding to each state node in the dynamic causal model.
3. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 2, characterized in that, The specific way in which the processing module performs the feature extraction includes: when the real-time operation data is image data acquired by the panoramic imaging unit, the processing module is configured to first locate the display area of an instrument or a meter in the image data through a target detection model, then identify the numerical value from the located display area through a character recognition model, and finally use the identified numerical value as the structured state value to update the corresponding state node.
4. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 3, characterized in that, The processing module is further configured to: lock the fault state of the preset fault in a temporary virtual dynamic causal model by taking the initial cause with the highest confidence in the fault root cause propagation chain as the preset fault; and call the fault evolution trend prediction to deduce the future failure time and consequences of the system under the preset fault, for generating an active operation and maintenance strategy.
5. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 1, characterized in that, The specific way in which the processing module performs the fault evolution trend prediction includes: based on the global state evolution equation determined by the overall topology structure of the dynamic causal model, iteratively solving state vectors at multiple time steps in the future to form the future change trajectory, taking the current global state as the initial condition.
6. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 1 or 5, characterized in that, The processing module is further configured to: calculate the prediction error by comparing the result of the fault evolution trend prediction with subsequent real-time operation data; and based on the prediction error, adaptively adjust the model parameters in the transfer function operator through an online optimization algorithm to realize self-correction of the dynamic causal model.
7. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 6, characterized in that, The processing module implements self-correction of the dynamic causal model through an online optimization algorithm, which is an extended Kalman filter algorithm, and is configured to augment a model parameter vector encapsulated in all transfer function operators in the dynamic causal model to be adjusted in the dynamic causal model with a global state vector of the dynamic causal model to jointly constitute an augmented state vector; and based on the augmented state vector, iteratively optimize the model parameter vector by continuously performing a prediction step and an update step; wherein the update step uses the prediction error to calculate a Kalman gain and uses the gain to correct the prior estimate value generated in the prediction step, thereby realizing simultaneous optimization and adaptive adjustment of the model parameter vector while continuously correcting the global state vector. 8.The edge fusion computing based box-type substation panoramic monitoring operation and maintenance system according to claim 1, characterized in that, The processing module implements self-evolution of the dynamic causal model, and is configured to continuously monitor and analyze the distribution of the prediction error on the topology structure of the dynamic causal model, and when identifying that the prediction error is long-term and significantly concentrated in a certain high-error area, it is determined that there may be a mismatch between the physical reality and the model structure of the area, and a new causal relationship mining is automatically triggered; The new causal relationship mining is realized by calculating information entropy to quantify the potential information flow strength between the state nodes in the high-error area and other state nodes, and the information flow strength is calculated through the following transfer entropy formula: wherein represents the information flow strength from the source state node to the target state node ; represents the state of the target state node at a future time instant ; representative of a target state node representative of an information flow strength from a source state node to a target state node representative target state node at a future time state; representative target state node at the current time and before it a history state sequence consisting of the states of the target state node at the current time representative source state node at the current time instant and the states of the previous history state sequence a probability distribution representing a corresponding variable or combination of variables; when it is calculated that there is an information flow strength from a source state node originally without a causal edge connection to a target state node within the high error area, and the strength continuously exceeds a preset structural correlation threshold value, the processing module generates a structural reconstruction suggestion for the dynamic causal model, which prompts the operation and maintenance personnel that there may be a new physical causal path that has not been defined at the current time and the historical state sequence composed of the states of the previous historical time representative source state node at the current time and previous historical state sequence a probability distribution representing a corresponding variable or combination of variables; and in the event that a strength of information flow from a source state node that was originally not causally connected to a target state node within the high error region is calculated, and the strength remains above a pre-defined structural correlation threshold, the processing module generates a structural reconfiguration suggestion for the dynamic causal model, the suggestion indicating to an operator that there can be an undefined, new physical causal path.
9. The fusion edge computing based box-type substation panoramic monitoring operation and maintenance system according to claim 1, characterized in that, The processing module is configured to generate a structured active operation and maintenance strategy based on the fault root cause chain and the deduced consequences and failure times, and the active operation and maintenance strategy includes a diagnosis conclusion, a predicted consequence, and a maintenance emergency level.
10. A method for panoramic monitoring and operation and maintenance of a box-type substation fused with edge computing, characterized in that, The edge fusion computing-based panoramic monitoring and operation and maintenance system of the box-type substation according to any one of claims 1-9, comprising the following steps: S1, constructing and maintaining a dynamic causal model on the edge computing unit, the model including a plurality of state nodes representing key physical states of the box-type substation, and a plurality of causal edges for quantifying dynamic causal effects between the state nodes; S2, acquiring real-time operation data of the box-type substation by a panoramic monitoring data acquisition device, and updating values of each state node in the dynamic causal model in real time by using the real-time operation data to form a global state vector ; S3. Then, based on the updated dynamic causal model, when a preset triggering condition is monitored, a corresponding intelligent operation and maintenance analysis is performed, the intelligent operation and maintenance analysis including fault root cause diagnosis, fault evolution trend prediction, and online self-evolution of the dynamic causal model, the fault evolution trend prediction being to solve the following global state evolution equation to obtain a future global state vector As an initial condition, the state at a future series of time steps is iteratively calculated forward, forming a future trajectory of change: wherein, is the global state vector at next time instant, is a system function determined by the overall topology of the dynamic causal model is an external input vector, is a model parameter vector containing all internal parameters of the causal edges, the online self-evolution of the dynamic causal model is achieved by continuously comparing the predicted state of the fault evolution trend prediction with the subsequently acquired real-time running data to calculate the prediction error, and when it is identified that the prediction error is long-term concentrated in a certain high error area, a new causal relationship mining is triggered, which quantifies the potential information flow strength between the target state node in the high error area and other source state nodes in the graph by calculating the following transfer entropy formula: wherein, represents the information flow strength from a source state node to a target state node , represents the state of the target state node at a future time instant , represents the history state sequence of the target state node , represents the history state sequence of the source state node , represents a probability distribution; if the information flow strength persists above a pre-defined threshold, a structural reconfiguration suggestion for the dynamic causal model is generated.