Distribution network power outage monitoring method and system based on artificial intelligence engine and intelligent agent

Abstract

The present invention relates to a power outage monitoring method and system for a distribution network based on an artificial intelligence engine and an intelligent body, including: S1: introducing a sensor intelligent fault detection model, monitoring the sensor state in real time, identifying and eliminating faulty sensors; S2: based on a sensor error compensation model, performing error compensation on sensor data, and fusing data from different sensors to obtain a final fusion feature matrix; S3: based on a multipath communication method, transmitting the sensor data after fusion processing of each edge node to a cloud platform; S4: obtaining the fusion feature matrix of the power outage area, and analyzing the power outage cause based on a power outage analysis model; S5: based on the output of the power outage analysis model, locating the fault point using a distribution network topology diagram and a device status diagram; S6: constructing a visualization unit to display the state and monitoring results of the distribution network in real time. The present invention improves the reliability and safety of the distribution network, and reduces power outage time and economic losses.

Description

Technical Field

[0001] The present invention relates to the field of distribution network monitoring, and in particular to a distribution network power outage monitoring method and system based on an artificial intelligence engine and an intelligent body. Background Art

[0002] In the power network, the operating status of the distribution network directly affects the stability and reliability of power supply. Therefore, it is very important to monitor power outages in substations and switches. The existing distribution automation system (DAS), although it can realize real-time monitoring and remote control of the distribution network, uses various sensors and communication technologies to collect data, but it is limited to specific areas or equipment, and the real-time performance and detection reliability of data need to be improved. As a key data acquisition device, the failure or error of the sensor will directly affect the accuracy of the monitoring results. The collected data may contain noise and errors, which affects the analysis and judgment of the cause of the power outage. In addition, data transmission in complex environments is easily interfered, resulting in data delay or packet loss, affecting the real-time monitoring effect. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a distribution network power outage monitoring method and system based on artificial intelligence engine and intelligent body, which can realize real-time monitoring, rapid response and precise positioning of distribution network faults, and provide reliable power outage risk assessment, greatly improving the reliability and safety of the distribution network, reducing power outage time and economic losses, and providing strong technical support for the stable operation and intelligent management of the distribution network.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A distribution network power outage monitoring method based on an artificial intelligence engine and an intelligent agent, comprising:

[0006] S1: Use sensors to collect real-time data of the power system at each edge node of the distribution network, introduce sensor intelligent fault detection model, monitor sensor status in real time, identify and eliminate faulty sensors;

[0007] S2: Based on the sensor error compensation model, according to the real-time environmental parameters, the sensor data is error compensated, and the data from different sensors are fused based on the data fusion technology to obtain the final fusion feature matrix;

[0008] The sensor error model is constructed to compensate the sensor data for errors according to the real-time environmental parameters, as follows:

[0009] Real-time acquisition of sensor measurements and environmental parameters, including temperature and humidity ;

[0010] Using the determined error coefficient, calculate the error:

[0011] ;

[0012] in, and are the error coefficients of temperature and humidity respectively;

[0013] The sensor measurement value is compensated according to the error model to obtain the true value of the sensor:

[0014] ;

[0015] Said and The least squares fitting method is used to obtain the following:

[0016] Get M groups of historical sensor and environmental data, the mth group of data includes the ambient temperature T m 、Humidity m , sensor measurement value D measured,m and the sensor true value D true,m :

[0017] Compute the error for each group:

[0018] ;

[0019] The error coefficient is obtained by least squares fitting:

[0020]

[0021] S3: Based on the multi-path communication method, the sensor data after fusion processing of each edge node is transmitted to the cloud platform;

[0022] S4: Obtain the fusion feature matrix of the power outage area, and analyze the cause of the power outage based on the power outage analysis model;

[0023] S5: Based on the output of the power outage analysis model, the fault point is located using the distribution network topology diagram and the equipment status diagram;

[0024] S6: Build a visualization unit to display the status and monitoring results of the distribution network in real time, and generate detailed reports and statistical information for operation and maintenance personnel to analyze and make decisions.

[0025] Furthermore, the sensor intelligent fault detection model includes random forest, LSTM model and fusion layer, and is specifically constructed as follows:

[0026] Get the historical data of sensors at each node of the distribution network, including voltage, current, power and frequency, and the vector representation is:

[0027] ;

[0028] in, are the vector representations of voltage, current, power and frequency respectively;

[0029] Normalize and smooth the sensor data;

[0030] Extract features from the processed data, including mean, standard deviation, maximum and minimum differences;

[0031] Use the extracted features to train random forest and LSTM models;

[0032] ;

[0033] ;

[0034] Among them, P RF is the output probability of the random forest model; is the number of decision trees in the random forest; For the A decision tree for the feature vector The estimated output of is the output probability of the LSTM model; n is the time window length of the LSTM model; the feature vector include ;

[0035] The fusion layer performs weighted averaging of the outputs of the trained random forest and LSTM models;

[0036] ;

[0037] in, is the fusion coefficient, is the final output probability after fusion;

[0038] According to the output result of the fusion layer and the set threshold Compare and determine the fault status in real time:

[0039] ;

[0040] in, It is a fault state.

[0041] Furthermore, based on data fusion technology, the data from different sensors are fused to obtain the final fusion feature matrix X fusion The details are as follows:

[0042] Determine the initial state x0 and covariance matrix P0, set the state transfer matrix F, measurement matrix H, process noise covariance Q and measurement noise covariance R;

[0043] Perform prediction and update steps for each time step;

[0044] There are two data sources and , define the state as , the measurement models are:

[0045] ;

[0046] ;

[0047] in, and is the measurement noise, and the measurement matrices are and , the covariances are R1 and R2 respectively;

[0048] The state transfer equation is:

[0049] ;

[0050] in, is the process noise, with covariance Q;

[0051] The prediction steps are as follows:

[0052] Using the state of the previous moment And the state transfer matrix F is predicted:

[0053] ;

[0054] in, is the predicted state vector at time t;

[0055] Forecast error covariance:

[0056] ;

[0057] in, It's in time Updated covariance; is the covariance at the time of prediction at time t; is the transpose of the state transfer matrix F;

[0058] The update steps are as follows:

[0059] Fuse the two sets of measurement data, combining the measurement matrix H and the measurement noise covariance matrix R:

[0060] ;

[0061] Calculate the Kalman gain:

[0062] ;

[0063] Update the state estimate:

[0064] ;

[0065] in, is the predicted state vector at time t-1;

[0066] Update the error covariance:

[0067] ;

[0068] Iteration time steps, concatenate the state estimates of each time step into a matrix to obtain the final fusion feature matrix:

[0069] .

[0070] Furthermore, the power outage analysis model is built based on CNN and LSTM, as follows:

[0071] Use a convolutional neural network to extract key features from the fused feature matrix, including convolution and pooling operations:

[0072] ;

[0073] ;

[0074] Among them, W c is the convolution kernel of the convolutional neural network, b c is the bias term of the convolutional neural network, ReLU is the activation function; Maxpooling is the pooling operation; is the feature after convolution operation; is the key feature matrix:

[0075] Forward and backward features based on bidirectional LSTM layers to capture key features:

[0076] ;

[0077] ;

[0078] in, and are the forward and backward hidden states at the current time t respectively; and are the forward and backward memory cell states at the current time t respectively; is the output of the bidirectional LSTM layer; is the key feature matrix at the current time t; T is the time step;

[0079] Use layer normalization and skip connections after the output of the bidirectional LSTM layer:

[0080] ;

[0081]

[0082] Among them, LayerNorm is the layer normalization operation; is the LSTM output feature matrix after layer normalization; is the feature matrix after skip connection, and is the sum of the layer normalized features and the convolutional features;

[0083] Use a fully connected layer to output the probability distribution of the cause of the power outage:

[0084]

[0085] in, W f is the weight of the fully connected layer, b f is the bias term, Softmax Used to output probability distribution.

[0086] Furthermore, the power outage analysis model uses the cross entropy loss function to calculate the loss between the model prediction value and the true label, performs backpropagation based on the cross entropy loss, calculates the gradient, and uses the optimization algorithm Adam to update the model parameters to minimize the loss function and obtain the optimal model.

[0087] Furthermore, based on the output of the power outage analysis model, the distribution network topology diagram and the equipment status diagram are used to locate the fault point, as follows:

[0088] The distribution network topology is represented as an adjacency matrix A. The possible fault propagation paths are determined using the distribution network topology:

[0089] Get the power of the adjacency matrix , represents the set of all nodes that can be reached from a node through at most k routes;

[0090] Set a maximum path length K and calculate the path matrix P:

[0091] ;

[0092] Comprehensive equipment status information to detect faulty equipment:

[0093] ;

[0094] Where σ is the activation function, and are weight and bias terms respectively, and S is the device status information;

[0095] Based on topology analysis and device status, calculate the fault score of each node:

[0096] ;

[0097] in, and is the weight coefficient, Output the probability distribution of power outage causes for the power outage analysis model

[0098] Sort the fault scores F' and take the ones with the highest scores first. Nodes are selected as candidate failure points;

[0099] Finally, based on the simulated annealing algorithm, the final set of faulty nodes is obtained.

[0100] Furthermore, based on the simulated annealing algorithm, the final set of faulty nodes is obtained, as follows:

[0101] From the top of the breakdown Randomly select the initial fault node set from the nodes to initialize the temperature and cooling coefficient ; Set the initial temperature T init And the cooling rate v, define the fitness function:

[0102]

[0103] in, Is a node The fault score of d(n) represents the difference between the fault node set n and the actual power outage node set, and λ is the adjustment coefficient;

[0104] Loop until the termination condition is met:

[0105] Generate neighboring solutions for the current solution through exchange, replacement and insertion operations ;

[0106] Calculate the fitness of the neighborhood solution and acceptance probability :

[0107] ;

[0108] in, is the current solution, is the number of iterations; is the fitness of the current solution;

[0109] According to the probability of acceptance , decide whether to accept the neighborhood solution:

[0110] ;

[0111] in, Neighborhood solution for the next iteration;

[0112] Update the temperature according to the cooling rate v:

[0113] ;

[0114] When the temperature Reduce to the preset minimum temperature T min Or when the number of iterations reaches the upper limit, the algorithm stops and the final optimal set of faulty nodes and its fitness are output.

[0115] A power outage monitoring system for a distribution network based on an artificial intelligence engine and an intelligent agent, comprising an edge node, a data transmission unit and a cloud platform;

[0116] Use sensors to collect real-time data of the power system at each edge node of the distribution network, and set up a sensor intelligent fault detection model at the edge node to monitor the sensor status in real time and identify and eliminate faulty sensors;

[0117] The edge node is also provided with a sensor error compensation model, which performs error compensation on sensor data according to real-time environmental parameters, and fuses data from different sensors based on data fusion technology;

[0118] The data transmission unit transmits the sensor data after fusion processing of each edge node to the cloud platform based on a multipath communication method;

[0119] The cloud platform includes a power outage analysis module and a visualization module;

[0120] The power outage analysis module analyzes the cause of the power outage based on the power outage analysis model and the fused sensor data of the power outage area; based on the output of the power outage analysis model, the distribution network topology map and the equipment status map are used to quickly locate the fault point;

[0121] The visualization module displays the status and monitoring results of the distribution network in real time, and generates detailed reports and statistical information for analysis and decision-making by operation and maintenance personnel.

[0122] The present invention has the following beneficial effects:

[0123] 1. The present invention can realize real-time monitoring, rapid response, and precise positioning of distribution network faults, and provide a reliable power outage risk assessment and early warning mechanism, which greatly improves the reliability and safety of the distribution network, reduces power outage time and economic losses, and provides strong technical support for the stable operation and intelligent management of the distribution network;

[0124] 2. The present invention collects data in real time through sensors deployed at the edge nodes of the distribution network, and introduces an intelligent fault detection model, which can monitor the status of sensors in real time, identify and eliminate faulty sensors, and ensure the accuracy and reliability of collected data, thereby improving the accuracy of power outage fault detection. Based on the sensor error compensation model and data fusion technology, the present invention can perform error compensation and fusion processing on the collected data, eliminate errors and noise in the data, generate a high-precision fusion feature matrix, and provide high-quality data support for subsequent power outage cause analysis;

[0125] 3. The present invention combines the power outage analysis model with the distribution network topology diagram and the equipment status diagram to quickly analyze the cause of the power outage and accurately locate the fault point, providing effective support for rapid processing and power supply restoration, shortening the power outage time, and improving the reliability and power supply quality of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0126] Figure 1 is a flow chart of the method of the present invention;

[0127] Figure 2 FIG. 4 is a system architecture diagram in one embodiment of the present invention. DETAILED DESCRIPTION

[0128] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0129] refer to Figure 1 In this embodiment, a distribution network power outage monitoring method based on an artificial intelligence engine and an intelligent agent is provided, comprising:

[0130] S1: Use sensors to collect real-time data of the power system at each edge node of the distribution network, introduce sensor intelligent fault detection model, monitor sensor status in real time, identify and eliminate faulty sensors;

[0131] S2: Based on the sensor error compensation model, according to the real-time environmental parameters, the sensor data is error compensated, and the data from different sensors are fused based on the data fusion technology to obtain the final fusion feature matrix;

[0132] The sensor error model is constructed to compensate the sensor data for errors according to the real-time environmental parameters, as follows:

[0133] Real-time acquisition of sensor measurements and environmental parameters, including temperature and humidity ;

[0134] Using the determined error coefficient, calculate the error:

[0135] ;

[0136] in, and are the error coefficients of temperature and humidity respectively;

[0137] The sensor measurement value is compensated according to the error model to obtain the true value of the sensor:

[0138] ;

[0139] Said and The least squares fitting method is used to obtain the following:

[0140] Get M groups of historical sensor and environmental data, the mth group of data includes the ambient temperature T m 、Humidity m , sensor measurement value D measured,m and the sensor true value D true,m :

[0141] Compute the error for each group:

[0142] ;

[0143] The error coefficient is obtained by least squares fitting:

[0144]

[0145] S3: Based on the multi-path communication method, the sensor data after fusion processing of each edge node is transmitted to the cloud platform;

[0146] S4: Obtain the fusion feature matrix of the power outage area, and analyze the cause of the power outage based on the power outage analysis model;

[0147] S5: Based on the output of the power outage analysis model, the fault point is located using the distribution network topology diagram and the equipment status diagram;

[0148] S6: Build a visualization unit to display the status and monitoring results of the distribution network in real time, and generate detailed reports and statistical information for operation and maintenance personnel to analyze and make decisions.

[0149] In this embodiment, the sensor intelligent fault detection model includes a random forest, an LSTM model and a fusion layer, and is specifically constructed as follows:

[0150] Get the historical data of sensors at each node of the distribution network, including voltage, current, power and frequency, and the vector representation is:

[0151] ;

[0152] in, are the vector representations of voltage, current, power and frequency respectively;

[0153] Normalize and smooth the sensor data;

[0154] Extract features from the processed data, including mean, standard deviation, maximum and minimum differences;

[0155] Use the extracted features to train random forest and LSTM models;

[0156] ;

[0157] ;

[0158] Among them, P RF is the output probability of the random forest model; is the number of decision trees in the random forest; For the A decision tree for the feature vector The estimated output of is the output probability of the LSTM model; n is the time window length of the LSTM model; the feature vector include ;

[0159] The fusion layer performs weighted averaging of the outputs of the trained random forest and LSTM models;

[0160] ;

[0161] in, is the fusion coefficient, is the final output probability after fusion;

[0162] According to the output result of the fusion layer and the set threshold Compare and determine the fault status in real time:

[0163] ;

[0164] in, It is a fault state.

[0165] In this embodiment, the data from different sensors are fused based on data fusion technology to obtain the final fusion feature matrix Xfusion The details are as follows:

[0166] Determine the initial state x0 and covariance matrix P0, set the state transfer matrix F, measurement matrix H, process noise covariance Q and measurement noise covariance R;

[0167] Perform prediction and update steps for each time step;

[0168] There are two data sources and , define the state as , the measurement models are:

[0169] ;

[0170] ;

[0171] in, and is the measurement noise, and the measurement matrices are and , the covariances are R1 and R2 respectively;

[0172] The state transfer equation is:

[0173] ;

[0174] in, is the process noise, with covariance Q;

[0175] The prediction steps are as follows:

[0176] Using the state of the previous moment And the state transfer matrix F is predicted:

[0177] ;

[0178] in, is the predicted state vector at time t;

[0179] Forecast error covariance:

[0180] ;

[0181] in, It's in time Updated covariance; is the covariance at the time of prediction at time t; is the transpose of the state transfer matrix F;

[0182] The update steps are as follows:

[0183] Fuse the two sets of measurement data, combining the measurement matrix H and the measurement noise covariance matrix R:

[0184] ;

[0185] Calculate the Kalman gain:

[0186] ;

[0187] Update the state estimate:

[0188] ;

[0189] in, is the predicted state vector at time t-1;

[0190] Update the error covariance:

[0191] ;

[0192] Iteration time steps, concatenate the state estimates of each time step into a matrix to obtain the final fusion feature matrix:

[0193] .

[0194] In this embodiment, based on the multipath communication method, the sensor data after fusion processing of each edge node is transmitted to the cloud platform, as follows:

[0195] Multipath includes: low-latency, high-bandwidth network paths (such as 5G); medium-latency and high-bandwidth paths (such as LTE); high-latency, low-bandwidth paths (such as LoRa)

[0196] The cloud platform is equipped with a data receiving module: it receives and reorganizes data from multiple paths.

[0197] The edge node preprocesses and fuses the collected sensor data to obtain the fused state estimation data.

[0198] Encapsulate the fused data into a data packet, the data packet format includes: data type identifier, timestamp, edge node identifier, data payload, and checksum;

[0199] The optimal path or multiple paths are selected for data transmission based on the current network status (bandwidth, latency, packet loss rate, etc.). The data packet can be divided into multiple data segments, and each data segment is transmitted through a different path.

[0200] The data segments are transmitted through various paths, using encryption and verification mechanisms. The cloud platform receives and reassembles the data segments and provides confirmation feedback.

[0201] In this embodiment, the power outage analysis model is constructed based on CNN and LSTM, as follows:

[0202] Use a convolutional neural network to extract key features from the fused feature matrix, including convolution and pooling operations:

[0203] ;

[0204] ;

[0205] Among them, W c is the convolution kernel of the convolutional neural network, b c is the bias term of the convolutional neural network, ReLU is the activation function; Maxpooling is the pooling operation; is the feature after convolution operation; is the key feature matrix:

[0206] Forward and backward features based on bidirectional LSTM layers to capture key features:

[0207] ;

[0208] ;

[0209] in, and are the forward and backward hidden states at the current time t respectively; and are the forward and backward memory cell states at the current time t respectively; is the output of the bidirectional LSTM layer; is the key feature matrix at the current time t; T is the time step;

[0210] Use layer normalization and skip connections after the output of the bidirectional LSTM layer:

[0211] ;

[0212]

[0213] Among them, LayerNorm is the layer normalization operation; is the LSTM output feature matrix after layer normalization; is the feature matrix after skip connection, and is the sum of the layer normalized features and the convolutional features;

[0214] Use a fully connected layer to output the probability distribution of the cause of the power outage:

[0215]

[0216] in, W f is the weight of the fully connected layer, b f is the bias term, Softmax Used to output probability distribution.

[0217] In this embodiment, the power outage analysis model uses a cross entropy loss function to calculate the loss between the model prediction value and the true label, performs back propagation based on the cross entropy loss, calculates the gradient, and uses the optimization algorithm Adam to update the model parameters to minimize the loss function and obtain the optimal model.

[0218] In this embodiment, based on the output of the power outage analysis model, the distribution network topology diagram and the device status diagram are used to locate the fault point, as follows:

[0219] The distribution network topology is represented as an adjacency matrix A. The possible fault propagation paths are determined using the distribution network topology:

[0220] Get the power of the adjacency matrix , represents the set of all nodes that can be reached from a node through at most k routes;

[0221] Set a maximum path length K and calculate the path matrix P:

[0222] ;

[0223] Comprehensive equipment status information to detect faulty equipment:

[0224] ;

[0225] Where σ is the activation function, and are weight and bias terms respectively, and S is the device status information;

[0226] Based on topology analysis and device status, calculate the fault score of each node:

[0227] ;

[0228] in, and is the weight coefficient, Output the probability distribution of power outage causes for the power outage analysis model,

[0229] Sort the fault scores F' and take the ones with the highest scores first. Nodes are selected as candidate failure points;

[0230] Finally, based on the simulated annealing algorithm, the final set of faulty nodes is obtained.

[0231] Preferably, in this embodiment, the final set of faulty nodes is obtained based on a simulated annealing algorithm, as follows:

[0232] From the top of the breakdown Randomly select the initial fault node set from the nodes to initialize the temperature and cooling coefficient ; Set the initial temperature T init And the cooling rate v, define the fitness function:

[0233]

[0234] in, Is a node The fault score of d(n) represents the difference between the fault node set n and the actual power outage node set, and λ is the adjustment coefficient;

[0235] Loop until the termination condition is met:

[0236] For the current solution , generate neighborhood solutions through exchange, replacement and insertion operations ;

[0237] Calculate the fitness of the neighborhood solution and acceptance probability :

[0238] ;

[0239] in, is the current solution, is the number of iterations; is the fitness of the current solution;

[0240] According to the probability of acceptance , decide whether to accept the neighborhood solution:

[0241] ;

[0242] in, Neighborhood solution for the next iteration;

[0243] Update the temperature according to the cooling rate v:

[0244] ;

[0245] When the temperature Reduce to the preset minimum temperature T min Or when the number of iterations reaches the upper limit, the algorithm stops and the final optimal set of faulty nodes and its fitness are output.

[0246] refer to Figure 2,The present invention also provides another embodiment, a power outage monitoring system for a distribution network based on an artificial intelligence engine and an intelligent agent, comprising: an edge node, a data transmission unit and a cloud platform;

[0247] Use sensors to collect real-time data of the power system at each edge node of the distribution network, and set up a sensor intelligent fault detection model at the edge node to monitor the sensor status in real time and identify and eliminate faulty sensors;

[0248] The edge node is also provided with a sensor error compensation model, which performs error compensation on sensor data according to real-time environmental parameters, and fuses data from different sensors based on data fusion technology;

[0249] The data transmission unit transmits the sensor data after fusion processing of each edge node to the cloud platform based on a multipath communication method;

[0250] The cloud platform includes a power outage analysis module, a power outage risk assessment module and a visualization module;

[0251] The power outage analysis module analyzes the cause of the power outage based on the power outage analysis model and the fused processed sensor data of the power outage area;

[0252] Based on the output of the power outage analysis model, the distribution network topology diagram and equipment status diagram are used to quickly locate the fault point, and finally intelligent restoration is performed based on the intelligent agent;

[0253] The power outage risk assessment module performs power outage risk assessment on normal areas based on the power outage risk assessment model, and issues early warning for high-risk areas;

[0254] The visualization module displays the status and monitoring results of the distribution network in real time, and generates detailed reports and statistical information for analysis and decision-making by operation and maintenance personnel.

[0255] In this embodiment, the visualization model uses HTML5 and front-end framework development to display the real-time status of the distribution network, and uses the D3.js graphics library to create interactive data visualization charts to present complex data to users in an intuitive way. GIS technology (such as Leaflet and OpenLayers) is introduced to display the geographical distribution of the distribution network on the map, and display the status and fault information of each node in real time.

[0256] The visualization module provides a real-time monitoring diagram of the overall operation status of the distribution network, showing the key parameters of each substation and line. Users can click on specific nodes in the diagram to view detailed data, such as voltage, current, fault location and cause, and obtain comprehensive operation information. Ensure that the visualization interface can update data changes in real time to provide operation and maintenance personnel with the latest distribution network status information.

[0257] The system can automatically generate operation reports based on the day, week, and month. The reports contain key statistical data and analysis results, such as power load, number of failures, etc. Operation and maintenance personnel can select time periods and monitoring indicators according to specific needs, and the system will generate corresponding customized reports.

[0258] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0259] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0260] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0261] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0262] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A distribution network power outage monitoring method based on artificial intelligence engine and intelligent agent, characterized in that: include: S1: Use sensors to collect real-time data of the power system at each edge node of the distribution network, introduce sensor intelligent fault detection model, monitor sensor status in real time, identify and eliminate faulty sensors; S2: Based on the sensor error compensation model, according to the real-time environmental parameters, the sensor data is error compensated, and the data from different sensors are fused based on the data fusion technology to obtain the final fusion feature matrix; Construct a sensor error compensation model and perform error compensation on sensor data according to real-time environmental parameters, as follows: Real-time acquisition of sensor measurements and environmental parameters, including temperature and humidity ; Using the determined error coefficient, calculate the error: ； in, and are the error coefficients of temperature and humidity respectively; The sensor measurement value is compensated according to the error model to obtain the true value of the sensor: ； Said and The least squares fitting method is used to obtain the following: Get M groups of historical sensor and environmental data, the mth group of data includes the ambient temperature T m 、Humidity m , sensor measurement value D measured,m and the sensor true value D true,m : Compute the error for each group: ; The error coefficient is obtained by least squares fitting: ； S3: Based on the multi-path communication method, the sensor data after fusion processing of each edge node is transmitted to the cloud platform; S4: Obtain the fusion feature matrix of the power outage area, and analyze the cause of the power outage based on the power outage analysis model; S5: Based on the output of the power outage analysis model, the fault point is located using the distribution network topology diagram and the equipment status diagram; S6: A visualization unit is constructed to display the status and monitoring results of the distribution network in real time, and to generate detailed reports and statistical information for analysis and decision-making by operation and maintenance personnel; The sensor intelligent fault detection model includes a random forest, an LSTM model and a fusion layer, and is specifically constructed as follows: Get the historical data of sensors at each node of the distribution network, including voltage, current, power and frequency, and the vector representation is: ； in, are the vector representations of voltage, current, power and frequency respectively; Normalize and smooth the sensor historical data; Extract features from the processed data, including mean, standard deviation, maximum and minimum differences; Use the extracted features to train random forest and LSTM models; ； ； Among them, P RF is the output probability of the random forest model; is the number of decision trees in the random forest; For the A decision tree for the feature vector The estimated output of is the output probability of the LSTM model; n is the time window length of the LSTM model; the feature vector include ; The fusion layer performs weighted averaging of the outputs of the trained random forest and LSTM models; ； in, is the fusion coefficient, is the final output probability after fusion; According to the output result of the fusion layer and the set threshold Compare and determine the fault status in real time: ； in, It is a fault state.

2. A distribution network power outage monitoring method based on artificial intelligence engine and intelligent agent according to claim 1, characterized in that: The data fusion technology is based on fusing the data from different sensors to obtain the final fusion feature matrix X fusion The details are as follows: Determine the initial state x0 and covariance matrix P0, set the state transfer matrix F, measurement matrix H, process noise covariance Q and measurement noise covariance R; Perform prediction and update steps for each time step; There are two data sources and , define the state as , the measurement models are: ； ； in, and is the measurement noise, and the measurement matrices are and , the covariances are R1 and R2 respectively; The state transfer equation is: ； in, is the process noise, with covariance Q; The prediction steps are as follows: Using the state of the previous moment And the state transfer matrix F is predicted: ； in, is the predicted state vector at time t; Forecast error covariance: ； in, It's in time Updated covariance; is the covariance at the time of prediction at time t; is the transpose of the state transfer matrix F; The update steps are as follows: Fuse the two sets of measurement data, combining the measurement matrix H and the measurement noise covariance matrix R: ； Calculate the Kalman gain: ； Update the state estimate: ； in, For in time The predicted state vector of Update the error covariance: ； Iteration time steps, concatenate the state estimates of each time step into a matrix, and obtain the final fusion feature matrix: 。 3. A distribution network power outage monitoring method based on artificial intelligence engine and intelligent agent according to claim 2, characterized in that: The power outage analysis model is built based on CNN and LSTM, as follows: Use a convolutional neural network to extract key features from the fused feature matrix, including convolution and pooling operations: ； ； Among them, W c is the convolution kernel of the convolutional neural network, b c is the bias term of the convolutional neural network, ReLU is the activation function; Maxpooling is the pooling operation; is the feature after convolution operation; is the key feature matrix: Forward and backward features based on bidirectional LSTM layers to capture key features: ； ； in, and are the forward and backward hidden states at the current time t respectively; and are the forward and backward memory cell states at the current time t respectively; is the output of the bidirectional LSTM layer; is the key feature matrix at the current time t; T is the time step; Use layer normalization and skip connections after the output of the bidirectional LSTM layer: ； ； Among them, LayerNorm is the layer normalization operation; is the LSTM output feature matrix after layer normalization; is the feature matrix after skip connection, and is the sum of the layer normalized features and the convolutional features; Use a fully connected layer to output the probability distribution of the cause of the power outage: ； in, W f is the weight of the fully connected layer, b f is the bias term, Softmax Used to output probability distribution.

4. A distribution network power outage monitoring method based on artificial intelligence engine and intelligent agent according to claim 3, characterized in that: The power outage analysis model uses a cross entropy loss function to calculate the loss between the model prediction value and the true label, performs back propagation based on the cross entropy loss, calculates the gradient, and uses the optimization algorithm Adam to update the model parameters to minimize the loss function and obtain the optimal model.

5. A distribution network power outage monitoring method based on artificial intelligence engine and intelligent agent according to claim 1, characterized in that: The output of the power outage analysis model is used to locate the fault point using the distribution network topology diagram and the equipment status diagram, as follows: The distribution network topology is represented as an adjacency matrix A. The possible fault propagation paths are determined using the distribution network topology: Get the power of the adjacency matrix , represents the set of all nodes that can be reached from a node through at most k routes; Set a maximum path length K and calculate the path matrix P: ； Comprehensive equipment status information to detect faulty equipment: ； Among them, σ is the activation function, W s and b s are weight and bias terms respectively, and S is the device status information; Based on topology analysis and device status, calculate the fault score of each node: ； in, and is the weight coefficient, Output the probability distribution of power outage causes for the power outage analysis model Sort the fault scores F' and take the ones with the highest scores first. Nodes are selected as candidate failure points; Finally, based on the simulated annealing algorithm, the final set of faulty nodes is obtained.

6. The power outage monitoring method for distribution network based on artificial intelligence engine and intelligent agent according to claim 5 is characterized in that: The final set of faulty nodes is obtained based on the simulated annealing algorithm, as follows: From the top of the breakdown Randomly select the initial fault node set from the nodes to initialize the temperature and cooling coefficient ; Set the initial temperature T init And the cooling rate v, define the fitness function: in, Is a node The fault score of d(n) represents the difference between the fault node set n and the actual power outage node set, and λ is the adjustment coefficient; Loop until the termination condition is met: Generate neighboring solutions for the current solution through exchange, replacement and insertion operations ; Calculate the fitness of the neighborhood solution and the probability of acceptance : ; in, is the current solution, is the number of iterations; is the fitness of the current solution; According to the probability of acceptance , decide whether to accept the neighborhood solution: ； in, Neighborhood solution for the next iteration; Update the temperature according to the cooling rate v: ; When the temperature Reduce to the preset minimum temperature T min Or when the number of iterations reaches the upper limit, the algorithm stops and the final optimal set of faulty nodes and its fitness are output.

7. A system based on the distribution network power outage monitoring method based on an artificial intelligence engine and an intelligent agent as described in any one of claims 1 to 6, characterized in that: include: Edge nodes, data transmission units, and cloud platforms; Use sensors to collect real-time data of the power system at each edge node of the distribution network, and set up a sensor intelligent fault detection model at the edge node to monitor the sensor status in real time and identify and eliminate faulty sensors; The edge node is also provided with a sensor error compensation model, which performs error compensation on sensor data according to real-time environmental parameters, and fuses data from different sensors based on data fusion technology; The data transmission unit transmits the sensor data after fusion processing of each edge node to the cloud platform based on a multipath communication method; The cloud platform includes a power outage analysis module and a visualization module; The power outage analysis module analyzes the cause of the power outage based on the power outage analysis model and the fused processed sensor data of the power outage area; Based on the output of the power outage analysis model, the fault point is quickly located using the distribution network topology diagram and equipment status diagram; The visualization module displays the status and monitoring results of the distribution network in real time, and generates detailed reports and statistical information for analysis and decision-making by operation and maintenance personnel.

Images