An automated defect identification method for distribution network fault handling

Abstract

The present invention proposes an automated defect identification method for distribution network fault processing, including: deploying sensors to collect multi-dimensional data; constructing a quantum computing engine to process data and construct a dynamic knowledge graph; constructing a health status assessment model to form a globally optimized health status assessment model; creating a holographic digital twin, which is simulated according to real-time sensor data and historical data; combining unsupervised learning and online learning to construct a distribution network state prediction model to dynamically predict and detect distribution network faults; using a quantum particle swarm optimization algorithm to calculate the optimization strategy to optimize the global situation; combining mixed reality technology to build an MR system; using an algorithm based on a combination of adaptive neural networks and genetic programming to build an equipment life assessment model to provide a decision for early intervention in faults. The present invention improves the operating quality and fault handling efficiency of distribution network equipment, and provides a solid guarantee for the safe and stable operation of the power system.

Description

Technical Field

[0001] The present invention belongs to the field of communication networks, and in particular relates to an automatic defect identification method for distribution network fault processing. Background Art

[0002] In modern power systems, distribution networks are an important link between power transmission systems and end users, and their operating quality and stability are directly related to the reliability of power supply. However, with the continuous expansion of the scale and increase in complexity of distribution networks, traditional fault handling and defect identification methods have been unable to meet actual needs. Traditional distribution network fault handling mainly relies on manual inspections and rule-based diagnostic methods, which have problems such as slow response speed, low accuracy and high cost. In addition, with the increase in the types and number of power equipment, the amount of data has also exploded. Traditional data processing technology cannot effectively cope with such a huge amount of data, resulting in a further reduction in the efficiency of fault diagnosis and defect identification.

[0003] In the prior art, although the fault diagnosis methods based on big data analysis and machine learning have improved the accuracy of fault identification to a certain extent, they still face the following major problems: First, the heterogeneity and complexity of multi-source information make data fusion and information extraction extremely difficult. The existing methods have data islands when processing multimodal data, making it difficult to form a unified knowledge base. Secondly, traditional health status assessment methods usually only consider a single data source, ignoring the correlation between multimodal data, resulting in incomplete and inaccurate assessment results. Furthermore, the accuracy and real-time performance of equipment operation status prediction technology are insufficient, making it difficult to timely warn of potential faults, often resulting in processing only after the fault occurs. In addition, the existing maintenance strategy optimization methods mostly use heuristic algorithms. Although they have certain optimization effects, they still have problems with local optimality and slow convergence, making it difficult to achieve true closed-loop control. Finally, the application of existing big data technology in equipment life assessment is still not mature enough, and it is impossible to accurately predict the remaining life of the equipment, resulting in unreasonable maintenance and replacement strategies. Summary of the invention

[0004] The purpose of this invention is to propose an automated defect identification method for distribution network fault handling, which significantly improves the operation quality and fault handling efficiency of distribution network equipment by introducing technologies such as quantum computing, holographic digital twins, edge computing, federated learning and quantum heuristic optimization algorithms.

[0005] In order to achieve the above object, the present invention provides an automatic defect identification method for distribution network fault processing, the method comprising:

[0006] S1. Deploy sensors at edge computing nodes to collect multi-dimensional data of distribution network equipment and pre-process the data;

[0007] S2. Build a quantum computing engine to process the preprocessed data and build a dynamic knowledge graph;

[0008] S3. Build a health status assessment model based on federated learning and train it on edge computing nodes respectively. Use federated learning technology to regularly send updates of the health status assessment model to the central server. The central server integrates the edge computing node updates of each edge computing node to form a globally optimized health status assessment model.

[0009] S4. Creating a holographic digital twin, wherein the holographic digital twin simulates and predicts the operating status of distribution network equipment based on real-time sensor data and historical data;

[0010] S5. Combining unsupervised learning and online learning to build a distribution network state prediction model to dynamically predict distribution network faults and detect anomalies;

[0011] S6. Based on the dynamic prediction results, the quantum particle swarm optimization algorithm is used to calculate the optimization strategy to optimize the overall situation;

[0012] S7. According to the optimization strategy, the MR system is built in combination with mixed reality technology to provide real-time status and prediction information of equipment in a virtual environment to assist in the maintenance decision of the distribution network;

[0013] S8. Comprehensively analyze the multi-dimensional data of distribution network equipment according to maintenance decisions, extract key factors affecting life, build equipment life assessment model based on adaptive neural network and genetic programming algorithm, and dynamically adjust the parameters of equipment life assessment model to calculate the remaining life of the equipment, so as to provide early intervention decision for distribution network faults;

[0014] Wherein, the step S3 comprises:

[0015] S301, initialize a health status assessment model at each edge computing node, extract features from multi-dimensional data, obtain a feature vector x, and the target variable is the device health index y;

[0016] S302, training a health status assessment model using local data at each edge node;

[0017] S303, after one round of training, each edge node uploads the updated health status assessment model parameters to the central server;

[0018] S304: The central server receives the model parameters uploaded by all edge nodes, performs weighted average integration, and forms a global model, including:

[0019] Let the weight be α k ,satisfy Global model parameter w global , b global The calculation is as follows:

[0020]

[0021] Where K is the number of edge nodes, w k represents the model weight of the kth node, b k represents the model bias of the kth node, w global represents the global model weight, b global represents the global model bias, where α k According to the data volume N of each edge node k Allocation, expressed as follows:

[0022]

[0023] S305, the central server will integrate the global model parameters w global and b global Broadcast to each edge node to update local model parameters.

[0024] Furthermore, the multi-dimensional data of the distribution network equipment is collected and pre-processed; wherein the pre-processing includes denoising and data compression;

[0025] The denoising process uses a filter to denoise the collected data, which is expressed as follows:

[0026]

[0027] Among them, ω c represents the cutoff frequency of the filter, and H(s) represents the transfer function;

[0028] The data compression uses a data compression algorithm to reduce the amount of data, which is expressed as follows:

[0029]

[0030] in, and ψ j,k (t) represent the scaling function and wavelet function respectively, and d j,k Indicates the corresponding coefficient.

[0031] Furthermore, the quantum computing engine is constructed as follows:

[0032] Map the data into quantum states, and transform the data vector x=(x1, x2,…, x n ) is normalized and mapped to the quantum state |ψ>, which is expressed as follows:

[0033]

[0034] in, represents the norm of the data vector;

[0035] Applying quantum Fourier transform to the quantum state |ψ> to extract frequency domain features is expressed as follows:

[0036]

[0037] Where N represents the number of data points. represents the phase factor of Fourier transform;

[0038] Using the measurement operation in quantum computing to extract the correlation between data, let the element a of the correlation matrix A be ij represents the correlation strength between the i-th and j-th data points, obtained through quantum measurement, and is expressed as follows:

[0039] a ij =<ψ i |ψ j >

[0040] Among them, |ψ i > and |ψ j >represent the quantum states corresponding to the i-th and j-th data points respectively.

[0041] Furthermore, the dynamic knowledge graph uses each sensor or data point as a node of the knowledge graph, and the association relationship between the nodes is defined by the elements in the association matrix A obtained in the previous step. At the same time, a regular update mechanism is set to recalculate the association matrix and update the knowledge graph at regular intervals.

[0042] Furthermore, the step S302 specifically includes:

[0043] Calculate the predicted value, expressed as follows:

[0044]

[0045] Among them, w represents the weight vector and b represents the bias term;

[0046] Calculate the loss function, expressed as follows:

[0047]

[0048] Where N represents the number of training samples. and i Represent the predicted value and true value of the i-th sample respectively;

[0049] Gradient descent updates weights and biases, expressed as follows:

[0050]

[0051] Among them, η represents the learning rate, and They represent the partial derivatives of the loss function with respect to weight and bias respectively.

[0052] Furthermore, the step S4 specifically includes:

[0053] S401, constructing a physical model of the equipment according to the physical characteristics of the distribution network equipment;

[0054] S402, initializing the digital twin of the device based on the physical model combined with real-time data and historical data, and updating the state vector of the digital twin in real time according to the real-time multi-dimensional data and the physical model;

[0055] S403, run the physical model, perform simulation based on the real-time updated state vector, predict the state of the equipment at the future moment, and analyze the current state and predict potential failures in combination with the failure modes in the historical data.

[0056] Furthermore, the step S5 specifically includes:

[0057] S501. Extract features based on multi-dimensional data and calculate the statistics of each feature; where the statistics include the mean μ i ,variance and the autocorrelation function R xx (τ), expressed as follows:

[0058]

[0059] Among them, x i (t) represents the value of the i-th feature at time t, N represents the number of data points, and τ represents the time lag;

[0060] The K-means clustering algorithm based on the physical model is used to transform the feature vector Clustering is performed to identify device modes in different states, as shown below:

[0061]

[0062] Among them, k′ represents the number of clusters, C i represents the i-th cluster, μ i represents the center of the i-th cluster;

[0063] S502, based on the clustering results, establish an initial grid state prediction model for each device state, and use the least squares method to estimate the parameters of the initial grid state prediction model;

[0064] S503: Update feature statistics μ based on real-time collected data i (t) and R xx (τ,t), real-time adjustment of cluster center μ i , the recursive least squares (RLS) method is used to dynamically update the parameters of the power grid state prediction model;

[0065] S504: Predicting values ​​based on the updated grid status prediction model The residual ∈(t) with the actual observed value y(t) detects abnormal equipment status. When the system generates an abnormal alarm and records abnormal data, it prompts maintenance personnel to check, which is expressed as follows:

[0066]

[0067] When ∈(t) exceeds a set threshold value θ, it is determined to be abnormal.

[0068] Further, step S6 specifically includes:

[0069] S601. Represent the maintenance strategy as a vector s, where each element s i represents the parameter of the i-th maintenance action. Let the maintenance strategy space be S. Define the objective function f(s) of the maintenance strategy, which represents the balance between the effect and cost of the strategy:

[0070] f(s)=α·E(s)-β·C(s)

[0071] Among them, E(s) represents the maintenance effect, C(s) represents the maintenance cost, and α and β are weight coefficients;

[0072] S602, mapping the maintenance strategy vector s to a qubit representation, assuming n qubits, each qubit represents an element of the maintenance strategy vector, and the quantum state |ψ> is initialized to a uniform superposition state, which is represented as follows:

[0073]

[0074] Where n represents the number of qubits;

[0075] S603, constructing a quantum algorithm using quantum gate operations, evolving the quantum state, and searching for the optimal solution in the strategy space using the Grover search algorithm;

[0076] S604, measure the quantum state |ψ> and obtain the optimal maintenance strategy s * , where the quantum state measurement is expressed as follows:

[0077] P(|s * >)=| * |ψ)|​2

[0078] Among them, P(|s * >) indicates the optimal state obtained by measurement |s * >, |ψ> represents the probability of quantum state, |s * > indicates the optimal state;

[0079] S605, according to the optimal maintenance strategy s * , update actual maintenance plans and operations, and implement optimized maintenance strategies.

[0080] Furthermore, the S603 specifically includes:

[0081] Mark the target state and apply the objective function f(s), which is expressed as follows:

[0082] O|s>=(-1) f(s) |s>

[0083] Among them, O represents the Oracle operation, which is used to mark the target state, and f(s) represents the objective function of the maintenance strategy;

[0084] Reflect all quantum states to a uniform superposition state and perform an amplification operation, which is expressed as follows:

[0085] D=2|ψ><ψ|-I

[0086] Where D represents the diffusion operation, |ψ> represents the uniform superposition state, and I represents the identity matrix;

[0087] Grover iteration is to repeat the Oracle operation and diffusion operation, and the number of iterations is Where N is the size of the strategy space.

[0088] Furthermore, the remaining life prediction is calculated as follows:

[0089] The health status indicators of the equipment are calculated using the fused multi-dimensional data, as shown below:

[0090] HI(t)=αθ(t)+βv(t)

[0091] Among them, θ(t) represents temperature, v(t) represents vibration, α and β represent weight coefficients;

[0092] Based on the health status indicators, a linear degradation model is used to predict the remaining life of the equipment, which is expressed as follows:

[0093]

[0094] Among them, HI threshold is the threshold of the health status indicator, indicating the critical value at which the equipment needs maintenance or replacement. Indicates the rate of change of the health status indicator.

[0095] The beneficial technical effects of the present invention are at least as follows:

[0096] (1) By introducing quantum computing technology, the present invention greatly improves the processing speed and efficiency of multi-source information and solves the bottleneck problem of traditional methods in processing large-scale, heterogeneous data. The quantum computing engine can quickly process and analyze multimodal data such as electrical parameters and environmental parameters from different devices, providing a basis for building a dynamic knowledge graph.

[0097] (2) The present invention constructs a dynamic knowledge graph that is updated in real time, and records the associations and state changes between devices through efficient data processing results generated by quantum computing. This innovation solves the problem of data islands in the traditional multi-source information extraction process and realizes the effective integration and comprehensive utilization of information.

[0098] (3) The present invention deploys edge computing devices at each node of the distribution network and uses federated learning technology to achieve data sharing and collaborative learning among nodes. The edge computing devices collect and process local multimodal data in real time, and form a globally optimized health status assessment model through federated learning, overcoming the single data source limitation of traditional assessment methods.

[0099] (4) The present invention creates a holographic digital twin for each distribution network device, and performs simulation and prediction through real-time sensor data and historical data, which significantly improves the accuracy and real-time performance of the equipment operation status prediction and solves the problems of inaccurate prediction and delay in existing methods.

[0100] (5) The present invention adopts the quantum particle swarm optimization algorithm (QPSO) and uses the characteristics of quantum computing to achieve global optimization of the maintenance strategy. At the same time, it combines mixed reality technology to assist maintenance personnel in making decisions and operations, improves the execution effect of the maintenance strategy, and solves the problems of local optimality and slow convergence of traditional heuristic algorithms.

[0101] (6) The present invention uses a multimodal data fusion method to evaluate the equipment life. It uses a method combining adaptive neural networks with genetic programming to dynamically adjust the model structure and generate an optimal life evaluation model, thereby solving the problem of insufficient application of existing big data technology in life evaluation and providing more accurate equipment life prediction and maintenance strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0103] Figure 1 This is a flow chart of an automated defect identification method for distribution network fault processing according to the present invention. DETAILED DESCRIPTION

[0104] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0105] like Figure 1 As shown, an embodiment of the present invention provides an automatic defect identification method for distribution network fault processing, the method comprising the following steps S1-S8:

[0106] S1. Deploy sensors at edge computing nodes to collect multi-dimensional data of distribution network equipment and pre-process the data.

[0107] Specifically, a variety of sensors are installed at key nodes of the distribution network (such as substations, distribution cabinets, power lines, etc.), including voltage sensors, current sensors, temperature sensors, and vibration sensors. Each sensor converts the physical signal into a digital signal through a data acquisition module (such as an analog-to-digital converter, ADC) and transmits it to the edge computing device through a standard communication protocol (such as Modbus, IEC61850). Set the sampling frequency f of the sensor s Ensure the timeliness and accuracy of data. Each sensor samples synchronously and uses timestamp synchronization technology to ensure the time consistency of data.

[0108] Edge computing devices (such as industrial PCs and ARM processor devices) are deployed at various data collection points. These devices have certain data storage and processing capabilities.

[0109] Further, the preprocessing includes the following:

[0110] De-noising: Use filters to denoise the collected data. Common filters include low-pass filters, high-pass filters, and band-pass filters. The transfer function of a low-pass filter can be expressed as:

[0111]

[0112] Among them, ω c is the cutoff frequency of the filter. H(s) represents the transfer function, which describes the frequency response of the filter to the input signal, ω c is the cutoff frequency of the filter, which determines the highest frequency component that the filter can pass.

[0113] Data compression: Use data compression algorithms (such as sampling compression or compression based on discrete wavelet transform) to reduce the amount of data and reduce the transmission burden. The basic formula of discrete wavelet transform is:

[0114]

[0115] in, and ψ j,k (t) are the scaling function and wavelet function, and d j,k is the corresponding coefficient.

[0116] Furthermore, efficient transmission protocols such as MQTT and HTTP / 2 are used to ensure low-latency transmission of data between edge computing devices and central servers.

[0117] It is understandable that a circular buffer is configured in the edge computing device to store the data of the recent period to ensure that the data will not be lost when the network is unstable. The processed data is packaged into a standard format (such as JSON, XML) for transmission. Each data packet contains information such as timestamp, sensor ID, data value, etc. to ensure the traceability of the data.

[0118] Preferably, a verification mechanism (such as CRC verification) is introduced during the data transmission process to ensure the integrity and accuracy of the data during the transmission process. A data recovery mechanism is designed to recover the data through a retransmission request when data loss or damage is detected to ensure data integrity.

[0119] S2. Build a quantum computing engine to process the preprocessed data and build a dynamic knowledge graph.

[0120] Specifically, S201, first collect data and receive data packets that have been preliminarily processed and compressed from edge computing nodes. The data packets include information such as timestamp, sensor ID, data value, etc. to ensure the integrity and traceability of the data. Use the corresponding decompression algorithm to restore the data. Assuming that the compression method of discrete wavelet transform is used, the decompression process is:

[0121]

[0122] in, and ψ j,k (t) are the scaling function and wavelet function, and d j,k is the decompression factor.

[0123] S202. Build a quantum computing engine, including:

[0124] Map the data into quantum states. For example, map the data vector x = (x1, x2, ..., x n ) is normalized and mapped to the quantum state |ψ>:

[0125]

[0126] in, is the norm of the data vector.

[0127] Apply quantum Fourier transform to the quantum state |ψ> to extract frequency domain features:

[0128]

[0129] Where N is the number of data points, is the phase factor of the Fourier transform.

[0130] The correlation between data is extracted by using the measurement operation in quantum computing. Let the element a of the correlation matrix A be ij represents the correlation strength between the i-th and j-th data points, obtained through quantum measurement:

[0131] a ij =<ψ i |ψ j > (6)

[0132] Among them, |ψ i > and |ψ j >are the quantum states corresponding to the i-th and j-th data points respectively.

[0133] S203. Each sensor or data point is used as a node of the knowledge graph. The association relationship between nodes is defined by the elements in the association matrix A obtained in the previous step, that is, the weight of the edge. A regular update mechanism is set up to recalculate the association matrix and update the knowledge graph at regular intervals to ensure that it reflects the latest device status and data relationships.

[0134] Furthermore, the shortest path algorithm in graph theory (such as Dijkstra algorithm) is used to search for the shortest path between sensors in the knowledge graph to determine the fault propagation path.

[0135]

[0136] Among them, G is the knowledge graph, w is the weight of the edge, and s is the starting node.

[0137] Furthermore, the nodes and edges of the graph are used for logical reasoning to identify potential faults and defects. The reasoning rules can be based on pattern matching and association rules in the graph, for example:

[0138] If a ij >θ and ajk >θ then a ik =max(a ij , a jk ) (8)

[0139] Here, θ is the threshold of association strength.

[0140] S3. Build a health status assessment model based on federated learning and train it on edge computing nodes respectively. Use federated learning technology to regularly send health status assessment model updates to the central server. The central server integrates the edge computing node updates of each edge computing node to form a globally optimized health status assessment model, including:

[0141] S301: Initialize the local model.

[0142] Specifically, a health status assessment model, such as a linear regression model, is initialized at each edge computing node, and the initial weights are randomly assigned. The local data collected in step S1 is used, including characteristic variables such as voltage, current, temperature, and vibration. Let the characteristic vector be x and the target variable be the equipment health index y.

[0143] S302: Local model training.

[0144] The model is trained using local data at each edge node. Assuming a linear regression model is used, the training process includes the following steps:

[0145] Calculate the predicted value:

[0146]

[0147] Among them, w is the weight vector and b is the bias term.

[0148] Calculate the loss function (mean square error):

[0149]

[0150] Where N is the number of training samples, and i are the predicted value and true value of the i-th sample respectively.

[0151] Gradient descent updates weights and biases:

[0152]

[0153] Where η is the learning rate, and are the partial derivatives of the loss function with respect to weights and biases, respectively.

[0154] S303. After a round of training, each edge node uploads the updated model parameters (weight w and bias b) to the central server. Suppose the model parameter of the kth node is w k and b k .

[0155] S304: The central server receives the model parameters uploaded by all edge nodes, performs weighted average integration, and forms a global model. Let the weight be α k ,satisfy The global model parameters are calculated as follows:

[0156]

[0157] Where K is the number of edge nodes. Weight α k According to the data volume N of each edge node k Allocation, that is

[0158] S305, the central server will integrate the global model parameters w global and b global Broadcast to each edge node and update local model parameters. Repeat steps S302-S305 for multiple rounds of iterations until the global model converges.

[0159] S4. Create a holographic digital twin, which simulates and predicts the operating status of distribution network equipment based on real-time sensor data and historical data, including:

[0160] S401, using the real-time sensor data collected in step S1, including characteristic data such as voltage, current, temperature, vibration, etc., set these characteristic vectors as x(t), where t represents time. Obtain the historical operation data of the equipment, including fault records, maintenance records, etc., set these historical data as H.

[0161] Furthermore, a physical model of the device is established based on the physical characteristics of the device. Taking the power transformer as an example, its thermal model and electrical model are established:

[0162] The thermal model:

[0163]

[0164] Where C is the heat capacity, θ(t) is the transformer temperature, P(t) is the power loss, h is the heat dissipation factor, and θ a is the ambient temperature.

[0165] The electrical model:

[0166]

[0167] Where V(t) is the voltage, L is the inductance, i(t) is the current, and R is the resistance.

[0168] S402. Initialize the digital twin of the device based on the physical model, combined with real-time data and historical data. Assume that the state vector of the digital twin is X(t) and the initial state is X(t0). Update the state vector X(t) of the digital twin in real time based on the real-time data x(t) and the physical model. Taking the thermal model as an example, the state update formula is:

[0169]

[0170] Here, Δt is the time step.

[0171] S403: Run the physical model, perform simulation based on the real-time updated state vector X(t), and predict the state of the device at the future moment. Taking the electrical model as an example, predict the change of current:

[0172]

[0173] Furthermore, the current state X(t) is analyzed in combination with the failure mode in the historical data H to predict potential failures. Let the historical failure mode be F and the matching degree between the current state and the failure mode be M:

[0174]

[0175] When M exceeds a certain threshold θ f When the fault occurs, it is predicted.

[0176] S404. In the monitoring system of the control center, the simulation results and predicted status of the digital twin are displayed in real time, including the trend chart of key parameters such as temperature, current, and voltage. When the simulation prediction results show potential faults, the system automatically generates an alarm to prompt maintenance personnel to check and handle them.

[0177] S5. Combining unsupervised learning with online learning to build a distribution network state prediction model to dynamically predict distribution network faults and detect anomalies, including:

[0178] S501, the real-time sensor data x(t) collected from step S1, including voltage, current, temperature, vibration and other characteristics. Calculate the statistics of each characteristic, such as the mean μ i ,variance and the autocorrelation function R xx (τ).

[0179]

[0180]

[0181] Among them, xi (t) is the value of the i-th feature at time t, N is the number of data points, and τ is the time lag.

[0182] Furthermore, the K-means clustering algorithm based on the physical model is used to transform the feature vector Perform clustering to identify device modes in different states.

[0183]

[0184] Where k′ is the number of clusters, C i is the i-th cluster, μ i is the center of the ith cluster.

[0185] S502: Based on the clustering results, an initial prediction model is established for each device state. Taking electrical equipment as an example, the electrical model is adopted:

[0186]

[0187] Among them, V(t) is voltage, L is inductance, i(t) is current, and R is resistance.

[0188] Furthermore, the least squares method is used to estimate the model parameters and

[0189]

[0190] S503. As new data arrives, the feature statistics μ are continuously updated i (t) and R zx (τ,t), real-time adjustment of cluster center μ i . Dynamically update model parameters using recursive least squares (RLS) and

[0191]

[0192]

[0193] Where P(t) is the error covariance matrix, x(t) is the input vector, is the parameter vector.

[0194] S504: Predicted value based on the updated model The residual ∈(t) with the actual observation value y(t) is used to detect abnormal equipment status.

[0195]

[0196] When ∈(t) exceeds the set threshold θ, it is judged as abnormal. When the system generates an abnormal alarm and records the abnormal data, it prompts the maintenance personnel to check.

[0197] S6. Based on the dynamic prediction results, the quantum particle swarm optimization algorithm is used to calculate the optimization strategy to optimize the global situation.

[0198] Specifically, the prediction results of step S5 are used to globally optimize the maintenance strategy using the quantum particle swarm optimization algorithm (QPSO), and the characteristics of quantum computing are used to accelerate the optimization process. The maintenance strategy is regularly updated and adjusted to ensure its optimality and effectiveness. This step significantly improves the efficiency and effect of maintenance strategy optimization and solves the limitations of traditional optimization methods.

[0199] The following steps are involved:

[0200] S601, maintenance strategy modeling. The maintenance strategy is represented as a vector s, where each element s i represents the parameters of the ith maintenance action. Suppose the maintenance strategy space is Define the objective function f(s) of the maintenance strategy, which represents the balance between the effectiveness and cost of the strategy:

[0201] f(s)=α·E(s)-β·C(s) (27)

[0202] Among them, E(s) represents the maintenance effect, C(s) represents the maintenance cost, and α and β are weight coefficients. Assume that the maintenance effect can be measured by improving the equipment health ΔH and reducing the failure rate ΔF. The maintenance cost includes the material cost C m and labor cost C l :

[0203] E(s)=w1ΔH(s)+w2ΔF(s) (28)

[0204] C(s)=C m (s)+C l (s) (29)

[0205] S602, quantum state initialization. Map the maintenance strategy vector s to a qubit representation. Suppose there are n qubits, each of which can represent an element of the maintenance strategy vector. The quantum state |ψ> is initialized to a uniform superposition state:

[0206]

[0207] Here n is the number of qubits.

[0208] S603. Use quantum gate operations (such as Hadamard gates and CNOT gates) to construct quantum algorithms and evolve quantum states. Use the Grover search algorithm to search for the optimal solution in the strategy space. The Grover iteration process includes:

[0209] Oracle operation: mark the target state and apply the target function f(s):

[0210] O|s>=(-1) f(s) |s> (31)

[0211] Among them, O is the Oracle operation used to mark the target state, and f(s) is the objective function of the maintenance strategy.

[0212] Parallel diffusion operation: reflect all quantum states into a uniform superposition state for amplification:

[0213] D=2|ψ><ψ|-I (32)

[0214] Where D is the diffusion operation, |ψ> is the uniform superposition state, and I is the unit matrix.

[0215] Grover iteration: Repeat the Oracle operation and diffusion operation, the number of iterations is Where N is the size of the strategy space. S604, measure the quantum state |ψ> to obtain the optimal maintenance strategy s * The quantum state measurement formula is:

[0216] P(|s * >)=|〈s * |ψ>| 2 (33)

[0217] Among them, P(|s * >) is the optimal state to be measured |s * >, |ψ> is the probability of quantum state, |s * > is the optimal state. According to the optimal maintenance strategy s * , update actual maintenance plans and operations, and implement optimized maintenance strategies.

[0218] S7. Based on the optimization strategy, the MR system is built in combination with mixed reality technology to provide real-time status and prediction information of equipment in a virtual environment to assist in the maintenance decision of the distribution network.

[0219] Specifically, the following steps are included:

[0220] S701. The real-time sensor data x(t) collected from step S1, including characteristics such as voltage, current, temperature, and vibration, are integrated and processed. Based on the physical model, the digital model of the device is created using the holographic digital twin of step S4 and associated with the real-time data. The state vector X(t) of the digital model represents the current operating state of the device.

[0221] S702. Deploy mixed reality devices (such as HoloLens or AR glasses) that can obtain and display equipment status information in real time. Develop dedicated MR applications that can connect to the data integration platform to receive and display equipment status and maintenance suggestions in real time.

[0222] S703. Display the real-time status of the device in the MR device, including key parameters such as temperature, current, and voltage. Use a graphical interface and a three-dimensional model to intuitively display the current status and historical change trends of the device. Use the status prediction model in step S5 to display the possible future status changes and potential failure points of the device. Use color or icon identification to remind maintenance personnel to pay attention to possible abnormal areas.

[0223] S704. According to the maintenance strategy optimized in step S6, the MR system intuitively displays the specific location and steps of each maintenance operation on the equipment model. Maintenance personnel can see the operation instructions on the three-dimensional equipment model through the MR device. Maintenance personnel can interact with the MR system through gestures or voice commands to obtain more detailed maintenance step instructions and precautions. The MR system can update the maintenance steps in real time according to the operation progress to ensure the accuracy and consistency of the operation.

[0224] S705. The MR system records the operation process of the maintenance personnel in real time, including operation time, operation steps and other information, forming a complete maintenance record. The maintenance personnel can feedback the problems and suggestions in the operation through the MR equipment, and the system will adjust and optimize according to the feedback information to provide reference for subsequent maintenance work.

[0225] S8. Comprehensively analyze the multi-dimensional data of distribution network equipment based on maintenance decisions, extract key factors affecting life, use an algorithm based on a combination of adaptive neural networks and genetic programming to build an equipment life assessment model, and dynamically adjust the parameters of the equipment life assessment model to calculate the remaining life of the equipment, providing decisions for early intervention in distribution network failures.

[0226] Specifically, the decision data from step 7 is used to comprehensively analyze the multi-dimensional data of the equipment and extract the key factors affecting the lifespan. The method combining adaptive neural network and genetic programming is used to dynamically adjust the evaluation model structure and generate the optimal equipment lifespan evaluation model. This step improves the accuracy and reliability of lifespan evaluation and provides a scientific equipment maintenance and replacement strategy.

[0227] The step S8 comprises the following steps:

[0228] S801, collect data from different sensors in step S1, including voltage, current, temperature, vibration and other characteristics, and set these characteristic vectors as x i (t). The data from multiple sensors are fused to form a unified data set. The multi-source data is fused into a single feature vector X(t) using the weighted average method, where the weights are determined based on the importance and reliability of each sensor:

[0229]

[0230] Among them, w i is the weight of the i-th sensor, satisfying

[0231] S802. Establish a physical model of the equipment based on its working principle and physical characteristics. Take a power transformer as an example, and establish its thermal model and electrical model:

[0232] Thermal Model:

[0233]

[0234] Where C is the heat capacity, θ(t) is the transformer temperature, P(t) is the power loss, h is the heat dissipation coefficient, θ a is the ambient temperature.

[0235] Electrical Model:

[0236]

[0237] Among them, V(t) is voltage, L is inductance, i(t) is current, and R is resistance.

[0238] S803. Calculate the health status index (HealthIndex, HI) of the device using the fused data X(t), assuming that the health status is related to factors such as temperature and vibration:

[0239] HI(t)=αθ(t)+βv(t) (37)

[0240] Among them, θ(t) is temperature, v(t) is vibration, and α and β are weight coefficients.

[0241] Based on the health status indicators, a linear degradation model is used to predict the remaining useful life (RUL) of the equipment:

[0242]

[0243] Among them, HI thresholdis the threshold of the health status indicator, indicating the critical value at which the equipment needs maintenance or replacement. is the rate of change of the health status indicator.

[0244] S804. Develop reasonable maintenance strategies and optimize equipment maintenance plans based on the predicted remaining life. Ensure timely maintenance before equipment failure to reduce downtime and maintenance costs. Dynamically adjust maintenance strategies based on changes in real-time data and health status indicators to ensure reliability and stability of equipment operation.

[0245] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0246] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0247] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0248] Although embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. An automated defect identification method for distribution network fault processing, characterized in that: The method comprises the following steps: S1. Deploy sensors at edge computing nodes to collect multi-dimensional data of distribution network equipment and pre-process the data; S2. Build a quantum computing engine to process the preprocessed data and build a dynamic knowledge graph; S3. Build a health status assessment model based on federated learning and train it on edge computing nodes respectively. Use federated learning technology to regularly send updates of the health status assessment model to the central server. The central server integrates the edge computing node updates of each edge computing node to form a globally optimized health status assessment model. S4. Creating a holographic digital twin, wherein the holographic digital twin simulates and predicts the operating status of distribution network equipment based on real-time sensor data and historical data; S5. Combining unsupervised learning and online learning to build a distribution network state prediction model to dynamically predict distribution network faults and detect anomalies; S6. Based on the dynamic prediction results, the quantum particle swarm optimization algorithm is used to calculate the optimization strategy to optimize the overall situation; S7. According to the optimization strategy, the MR system is built in combination with mixed reality technology to provide real-time status and prediction information of equipment in a virtual environment to assist in the maintenance decision of the distribution network; S8. Comprehensively analyze the multi-dimensional data of distribution network equipment according to maintenance decisions, extract key factors affecting life, build equipment life assessment model based on adaptive neural network and genetic programming algorithm, and dynamically adjust the parameters of equipment life assessment model to calculate the remaining life of the equipment, so as to provide early intervention decision for distribution network faults; Wherein, the step S3 comprises: S301, initialize a health status assessment model at each edge computing node, extract features from multi-dimensional data, obtain a feature vector x, and the target variable is the device health index y; S302, training a health status assessment model using local data at each edge node; S303, after one round of training, each edge node uploads the updated health status assessment model parameters to the central server; S304: The central server receives the model parameters uploaded by all edge nodes, performs weighted average integration, and forms a global model, including: Let the weight be α k ,satisfy Global model parameter w global , b global The calculation is as follows: Where K is the number of edge nodes, w k represents the model weight of the kth node, b k represents the model bias of the kth node, w global represents the global model weight, b global represents the global model bias, where α k According to the data volume N of each edge node k Allocation, expressed as follows: S305, the central server will integrate the global model parameters w global and b global Broadcast to each edge node to update local model parameters.

2. The method for automatic defect identification of distribution network fault processing according to claim 1 is characterized in that: The multi-dimensional data of the distribution network equipment is collected and pre-processed; wherein the pre-processing includes denoising and data compression; The denoising process uses a filter to denoise the collected data, which is expressed as follows: Among them, ω c represents the cutoff frequency of the filter, and H(s) represents the transfer function; The data compression uses a data compression algorithm to reduce the amount of data, which is expressed as follows: in, and ψ j,k (t) represent the scaling function and wavelet function respectively, and d j,k Indicates the corresponding coefficient.

3. The method for automatic defect identification of distribution network fault processing according to claim 1 is characterized in that: The quantum computing engine is constructed as follows: Map the data into quantum states, and transform the data vector x=(x1, x2,…, x n ) is normalized and mapped to the quantum state |ψ>, which is expressed as follows: in, represents the norm of the data vector; Apply quantum Fourier transform to the quantum state |ψ> to extract frequency domain features, which is expressed as follows: Where N represents the number of data points. represents the phase factor of Fourier transform; Using the measurement operation in quantum computing to extract the correlation between data, let the element a of the correlation matrix A be ij represents the correlation strength between the i-th and j-th data points, obtained through quantum measurement, and is expressed as follows: a ij =<ψ i |ψ j > Among them, |ψ i > and |ψ j >represent the quantum states corresponding to the i-th and j-th data points respectively.

4. The method for automatic defect identification of distribution network fault processing according to claim 3 is characterized in that: The dynamic knowledge graph uses each sensor or data point as a node of the knowledge graph. The association relationship between the nodes is defined by the elements in the association matrix A obtained in the previous step. At the same time, a regular update mechanism is set to recalculate the association matrix and update the knowledge graph at regular intervals.

5. The method for automatic defect identification of distribution network fault processing according to claim 1 is characterized in that: The step S302 specifically includes: Calculate the predicted value, expressed as follows: Among them, w represents the weight vector and b represents the bias term; Calculate the loss function, expressed as follows: Where N represents the number of training samples. and i Represent the predicted value and true value of the i-th sample respectively; Gradient descent updates weights and biases, expressed as follows: Among them, η represents the learning rate, and They represent the partial derivatives of the loss function with respect to weight and bias respectively.

6. The method for automatic defect identification of distribution network fault processing according to claim 1 is characterized in that: The step S4 specifically includes: S401, constructing a physical model of the equipment according to the physical characteristics of the distribution network equipment; S402, initializing the digital twin of the device based on the physical model combined with real-time data and historical data, and updating the state vector of the digital twin in real time according to the real-time multi-dimensional data and the physical model; S403, run the physical model, perform simulation based on the real-time updated state vector, predict the state of the equipment at the future moment, and analyze the current state and predict potential failures in combination with the failure modes in the historical data.

7. The method for automatic defect identification of distribution network fault processing according to claim 1 is characterized in that: The step S5 specifically includes: S501. Extract features based on multi-dimensional data and calculate the statistics of each feature; where the statistics include the mean μ i ,variance and the autocorrelation function R xx (τ), expressed as follows: Among them, x i (t) represents the value of the i-th feature at time t, N represents the number of data points, and τ represents the time lag; The K-means clustering algorithm based on the physical model is used to transform the feature vector Clustering is performed to identify device modes in different states, as shown below: Among them, k′ represents the number of clusters, C i represents the i-th cluster, μ i represents the center of the i-th cluster; S502, based on the clustering results, establish an initial grid state prediction model for each device state, and use the least squares method to estimate the parameters of the initial grid state prediction model; S503: Update feature statistics μ based on real-time collected data i (t) and R xx (τ,t), real-time adjustment of cluster center μ i , the recursive least squares method is used to dynamically update the parameters of the power grid state prediction model; S504: Predicting values ​​based on the updated grid status prediction model The residual ∈(t) with the actual observed value y(t) detects abnormal equipment status. When the system generates an abnormal alarm and records abnormal data, it prompts maintenance personnel to check, which is expressed as follows: When ∈(t) exceeds a set threshold value θ, it is determined to be abnormal.

8. The method for automatic defect identification of distribution network fault processing according to claim 1 is characterized in that: Step S6 specifically includes: S601. Represent the maintenance strategy as a vector s, where each element s i represents the parameter of the i-th maintenance action. Let the maintenance strategy space be S. Define the objective function f(s) of the maintenance strategy, which represents the balance between the effect and cost of the strategy: f(s)=α·E(s)-β·C(s) Among them, E(s) represents the maintenance effect, C(s) represents the maintenance cost, and α and β are weight coefficients; S602, mapping the maintenance strategy vector s to a qubit representation, assuming n qubits, each qubit represents an element of the maintenance strategy vector, and the quantum state |ψ> is initialized to a uniform superposition state, which is represented as follows: Where n represents the number of qubits; S603, constructing a quantum algorithm using quantum gate operations, evolving the quantum state, and searching for the optimal solution in the strategy space using the Grover search algorithm; S604, measure the quantum state |ψ> and obtain the optimal maintenance strategy s * , where the quantum state measurement is expressed as follows: P(|s * >)=| * |ψ>| 2 ​ Among them, P(|s * >) indicates the optimal state obtained by measurement |s * >, |ψ> represents the probability of quantum state, |s * > indicates the optimal state; S605, according to the optimal maintenance strategy s * , update actual maintenance plans and operations, and implement optimized maintenance strategies.

9. The method for automatic defect identification of distribution network fault processing according to claim 8, characterized in that: The S603 specifically includes: Mark the target state and apply the objective function f(s), which is expressed as follows: O|s>=(-1) f(s) |s> Among them, O represents the Oracle operation, which is used to mark the target state, and f(s) represents the objective function of the maintenance strategy; Reflect all quantum states to a uniform superposition state and perform an amplification operation, which is expressed as follows: D=2|ψ><ψ|-I Where D represents the diffusion operation, |ψ> represents the uniform superposition state, and I represents the identity matrix; Grover iteration is to repeat the Oracle operation and diffusion operation, and the number of iterations is Where N is the size of the strategy space.

10. The method for automatic defect identification of distribution network fault processing according to claim 1, characterized in that: The remaining life prediction is calculated as follows: The health status indicators of the equipment are calculated using the fused multi-dimensional data, as shown below: HI(t)=αθ(t)+βv(t) Among them, θ(t) represents temperature, v(t) represents vibration, α and β represent weight coefficients; Based on the health status indicators, a linear degradation model is used to predict the remaining life of the equipment, which is expressed as follows: Among them, HI threshold is the threshold of the health status indicator, indicating the critical value at which the equipment needs maintenance or replacement. Indicates the rate of change of the health status indicator.

Images