Intelligent power distribution network line state real-time monitoring, analyzing and evaluating system

Abstract

The invention discloses an intelligent power distribution network line state real-time monitoring, analysis and evaluation system, and relates to the technical field of power system monitoring and control, the intelligent power distribution network line state real-time monitoring, analysis and evaluation system is constructed, data is collected in multiple dimensions, features are extracted through edge calculation, data is screened, a dynamic threshold model is constructed at a cloud end, and abnormity is identified. The method has the advantages that a three-dimensional digital twin network is constructed based on GIS data and an SCADA system through the power grid topology modeling module, and a unique'topology fingerprint 'containing electrical connection, physical position and logic control relation is generated for each piece of equipment; an upstream and downstream equipment list, a load influence range and a power flow transfer path are marked, and an association state evaluation module adopts a graph convolutional neural network and can accurately capture electrical coupling, physical proximity and historical fault association effects among equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring and control, and particularly to a real-time monitoring, analysis and evaluation system for the line state of an intelligent distribution network. Background Technique

[0002] With the rapid development of social economy, the demand for electricity continues to climb. As a key link in power supply, the safe and stable operation of the distribution network is crucial. The intelligent distribution network integrates modern information technology, communication technology and power technology, aiming to achieve efficient and reliable power distribution. However, there are still many challenges in the current monitoring, analysis and evaluation of the line state of the distribution network;

[0003] There are certain defects in the existing technologies. First of all, the traditional system adopts a single-device independent evaluation mode, lacking the analysis of the electrical, physical and fault correlations between devices, and it is difficult to detect potential hidden chain faults. Secondly, the traditional manual maintenance plan relies on experience and lacks overall planning, unable to integrate multi-dimensional information, resulting in frequent power outages and high operation and maintenance costs. For this reason, we propose a real-time monitoring, analysis and evaluation system for the line state of an intelligent distribution network. Summary of the Invention

[0004] The present invention realizes the global correlation analysis of device states through topological fingerprints and graph convolutional networks, solves the limitations of traditional single-device evaluation, and uses a dynamic threshold self-learning model to improve the early warning accuracy.

[0005] To achieve the above object, the present invention provides the following technical solution: A real-time monitoring, analysis and evaluation system for the line state of an intelligent distribution network, the analysis and evaluation system includes:

[0006] A multi-dimensional data acquisition module: used to collect electrical parameters, device temperature, mechanical vibration, environmental humidity and discharge signals of the line in real time;

[0007] The edge computing and processing module is used to calculate the kurtosis of the current waveform, the temperature change rate, the main frequency of the vibration signal, the harmonic distortion rate and the amplitude fluctuation coefficient in real time, generate an "abnormal data frame" containing abnormal events after noise reduction, and screen out valid data frames;

[0008] The cloud data analysis module receives the valid data frames, constructs a dynamic threshold self-learning model, generates real-time early warning thresholds that change with time, load and environment based on historical data, and at the same time integrates an abnormal pattern recognition algorithm to detect multi-parameter coupling abnormalities;

[0009] The power grid topology modeling module constructs a three-dimensional digital twin network containing the electrical connection relationships, physical location information, and logical control relationships of devices based on GIS data and the distribution network SCADA system, generates a unique "topological fingerprint" for each device, marks the list of upstream and downstream devices, the load influence range, and the power flow transfer path, generates an association weight matrix between devices, and quantifies the association degrees of electrical connection, physical proximity, and historical fault influence;

[0010] The association status evaluation module uses a graph convolutional neural network to construct a joint evaluation model, takes the state vector of a single device and the association weight matrix between devices as inputs, aggregates the state information of adjacent devices, outputs the corrected device health score and fault propagation probability considering the influence of topological association, and transmits the evaluation results;

[0011] The optimization decision module aims at "the least number of power outages" and "the optimal total life cycle cost of devices", combines the real-time operation constraints of the distribution network, generates an intelligent maintenance plan and operation adjustment strategy, and simulates the strategy effect through the digital twin system.

[0012] As a further solution of the present invention: In the multi-dimensional data acquisition module, the current sensor and the temperature sensor are integrated at the wire joint, and the vibration sensor is installed in the middle of the pole body and designed with a three-axis accelerometer to monitor the vibration signals of the line and the pole caused by external construction vibration, and identify mechanical stress anomalies through the changes in vibration frequency and amplitude.

[0013] As a further solution of the present invention: In the edge computing processing module, the specific methods for calculating the kurtosis of the current waveform, the temperature change rate, the main frequency of the vibration signal, the harmonic distortion rate, and the amplitude fluctuation coefficient are as follows:

[0014] The harmonic distortion rate of the voltage signal is calculated by the following formula:

[0015]

[0016] Among them, THD represents the total harmonic distortion rate, V n is the amplitude of each harmonic voltage, V1 is the amplitude of the fundamental voltage, N is the highest order of the calculated harmonics. When THD exceeds the national standard limit, it is used to evaluate the power quality. When THD in the low-voltage distribution system exceeds 5%, and in the medium-voltage and high-voltage systems, THD exceeds 3%, it is determined that the power quality is abnormal;

[0017] The amplitude fluctuation coefficient is calculated by the ratio of the amplitude difference between adjacent sampling points to the time interval where A(t) is the signal amplitude at the current moment, τ represents the sampling period. When the fluctuation coefficient exceeds 10% / ms, it is determined that the signal has a rapid abnormal fluctuation, and the data frame screening mechanism is triggered.

[0018] As a further solution of the present invention: in the cloud data analysis module, the dynamic threshold self-learning model generates a real-time warning threshold through a dual mechanism of time dimension and space dimension:

[0019] In the time dimension, based on the historical 24-hour load-temperature data, the temperature-load correlation curve of the peak period from 18:00 to 22:00 and the trough period from 0:00 to 6:00 of the load is fitted, and the temperature warning threshold of the corresponding period is automatically adjusted according to the real-time load rate;

[0020] In the space dimension, when abnormal environmental parameters are detected by 3 or more sensors adjacent to a certain device, according to the preset neighborhood influence rule, the temperature threshold of the device is dynamically corrected by ±10°C through the fuzzy logic algorithm.

[0021] As a further solution of the present invention: the correlation weight matrix between devices generated by the power grid topology modeling module is composed of a linear combination of electrical connection weight, physical proximity weight and historical fault correlation weight, and the specific formula is as follows:

[0022] W ij = α·W 电气 + β·W 物理 + γ·W 历史 ;

[0023] Wherein, W ij represents the correlation weight between device i and device j, and W 电气 represents the electrical connection weight. When device i and device j are directly electrically connected, that is, they are on the same electrical branch and there is no other device blocking the electrical characteristics in the middle, W 电气 takes the value of 0.8. When device i and device j are indirectly electrically connected through other devices, W 电气 takes the value of 0.3;

[0024] W 物理 represents the physical proximity weight. When the physical distance between device i and device j is less than 50 meters, W 物理 takes the value of 0.5;

[0025] When the physical distance between device i and device j is between 50 meters and 200 meters, W 物理 takes the value of 0.2;

[0026] When the physical distance between device i and device j exceeds 200 meters, W 物理 takes the value of 0;

[0027] W 历史 represents the historical fault correlation weight. When device i and device j have reported alarms due to the same fault chain in the past 12 months, that is, the time interval between the two faults is less than 24 hours and there is a logical correlation between the fault causes, W历史 The value is 0.4. If there is no historical association, then W 历史 The value is 0;

[0028] The weight coefficients α, β, and γ satisfy α + β + γ = 1 and are determined through machine learning training on historical failure data for the past 24 months.

[0029] As a further solution of the present invention: in the correlation state evaluation module, the graph convolutional neural network joint evaluation model determines the correlation evaluation of the device state through the following specific mechanism. First, a state vector S is constructed for each device i =[h i ,l i ,e i :

[0030] Among them, h i is the device health score, and the scoring range is 0 - 100 points. The value is proportional to the device health status. l i is the load rate, that is, the ratio of the actual load of the current device to the rated load, which reflects the load level of the device. e i is the comprehensive value of environmental parameters, which is output after environmental parameter fusion processing and is used to reflect the comprehensive impact of environmental factors on the device state;

[0031] Then, graph convolution operation is performed, and its specific formula is as follows:

[0032] S i ′ = σ(∑ j∈N(i) W ij ·S j +W ii ·S i );

[0033] Among them, N(i) represents the set of adjacent devices of device i, that is, the set composed of devices that are associated with device i in terms of electrical connection, physical location, and logical control. W ij is the association weight between devices and is used to measure the influence degree of adjacent device j on the state of device i. W ii is the association weight of device i itself. σ is the ReLU activation function. Through this graph convolution operation, the state information of adjacent devices is weighted and aggregated, and the corrected device state S i ′ is calculated and output.

[0034] As a further solution of the present invention: in the optimization decision module, the intelligent maintenance plan generated by the optimization decision module is implemented based on an integer programming model, and the global optimization of maintenance resources is specifically achieved through the following mechanism:

[0035] Construct an objective function min(C with the goal of minimizing "power outage impact cost" and "maintenance implementation cost"停电 +C 检修 ), where:

[0036] C 停电 is the user impact cost caused by power outage, which is weighted and calculated by user load level, power outage duration and the number of affected users. The specific calculation formula is as follows:

[0037]

[0038] λ k is the impact coefficient of the k-th type of user, t k is the power outage duration of the k-th type of user, n k is the number of the k-th type of users, and τ is the impact unit price per household per hour;

[0039] C 检修 is the maintenance implementation cost, covering equipment replacement cost, labor hour cost and material transportation cost, specifically:

[0040]

[0041] is the replacement cost of the i-th equipment, is the labor cost of the i-th equipment, is the material cost of the i-th equipment, x i is a 0-1 decision variable, where 1 represents maintaining the i-th equipment and 0 represents not maintaining.

[0042] As a further solution of the present invention: the fault propagation probability output by the associated state evaluation module is realized by constructing a fault propagation path model through a Petri net. At least 3 propagation paths are generated for each abnormal event, including electrical fault propagation, mechanical vibration conduction and environmental impact diffusion, and the occurrence probability of each path is statistically calculated according to historical fault data for early warning of cascading faults.

[0043] As a further solution of the present invention: the abnormal pattern recognition algorithm of the cloud data analysis module adopts a method combining isolation forest and dynamic time warping for unsupervised learning of multi-parameter time series, automatically identifying the combined features of "normal current + sudden temperature rise + abnormal vibration", and generating a new type of fault warning signal.

[0044] As a further solution of the present invention: the analysis and evaluation system further includes a federated learning collaboration module and a blockchain evidence storage module. The federated learning collaboration module is deployed between the edge computing nodes and the cloud to build a distributed federated learning network. Each edge node trains a local anomaly detection model based on local device data and submits model parameter updates to the cloud through a secure aggregation protocol. When the cloud integrates the global model, an adaptive weight allocation mechanism is introduced to dynamically adjust the parameter fusion weight according to the data diversity of the edge nodes. The blockchain evidence storage module performs hash chain storage on multi-dimensional collected data and key analysis results. Using the consortium chain architecture, it allows power grid operators, equipment manufacturers, and regulatory agencies to share data evidence storage through smart contracts to ensure the immutability of data.

[0045] Adopting the above technical solutions, compared with the prior art, the beneficial effects of the present invention are as follows:

[0046] 1. The present invention constructs a three-dimensional digital twin network through the power grid topology modeling module based on GIS data and the SCADA system, generates a unique "topology fingerprint" for each device, including electrical connections, physical locations, and logical control relationships, and establishes an association weight matrix between devices, marking the upstream and downstream device lists, load influence ranges, and power flow transfer paths. The association status evaluation module uses a graph convolutional neural network to input the single-device state vector and the association weight matrix into the model, aggregates the state information of adjacent devices, and outputs the corrected device health score and fault propagation probability considering the influence of topological associations, breaking through the limitation of traditional systems that only independently evaluate single devices, realizing the global association analysis of device states, and being able to accurately capture the electrical coupling, physical proximity, and historical fault association effects between devices.

[0047] 2. The present invention constructs a dynamic threshold self-learning model through the cloud data analysis module, fits the load-temperature association curve based on historical data, and dynamically adjusts the warning threshold in combination with the real-time load rate and environmental parameters, avoiding false alarms and missed alarms caused by the influence of traditional fixed thresholds on load fluctuations and environmental changes. At the same time, the anomaly pattern recognition technology integrating the isolated forest and dynamic time warping algorithms can perform unsupervised learning on multi-parameter time series and automatically identify complex coupling anomaly patterns such as "normal current + sudden temperature rise + abnormal vibration". This mechanism effectively solves the problem of low warning efficiency caused by rigid threshold settings and insufficient anomaly pattern recognition ability in traditional systems, significantly improving the anomaly detection accuracy under complex working conditions and greatly reducing the risk of ineffective responses and missed judgments of maintenance personnel.

[0048] 3. The present invention constructs a distributed learning network through a federated learning collaboration module. Edge nodes locally train a local anomaly detection model, and combine it with a cloud-based adaptive weight allocation mechanism to fuse the global model, improving the generalization ability of cross-regional device status evaluation and breaking through the data privacy bottleneck of centralized modeling. The blockchain evidence storage module uses a consortium blockchain architecture to hash multi-dimensional data and analysis results onto the chain, ensuring data immutability and traceability through timestamps and chain structures, strengthening the credibility of fault tracing and multi-party collaboration. At the same time, dynamic threshold self-learning and multi-parameter coupling anomaly detection technologies are used to achieve global correlation analysis of device status and environment-adaptive early warning. Finally, the system significantly improves the accuracy of associated analysis of distribution network devices and the reliability of anomaly detection, reduces manual operation and maintenance costs and power outage risks, forms a complete technical loop of "trusted data-intelligent analysis-precise decision-making", and comprehensively improves the safety, economy and intelligence level of distribution network operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the system process in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The following further describes the specific embodiments of the present invention with reference to the drawings. It should be noted here that the description of these embodiments is for helping to understand the present invention, but does not limit the present invention.

[0051] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0052] Please refer to the attached Figure 1 , An intelligent distribution network line status real-time monitoring, analysis and evaluation system of the present invention, the system includes:

[0053] Multi-dimensional data acquisition module: configured at key positions of the distribution network line, and the key positions include poles, wire joints and switchgear, and include current sensors, temperature sensors, vibration sensors, humidity sensors and ultraviolet imagers, which are used to collect electrical parameters, equipment temperature, mechanical vibration, environmental humidity and discharge signals of the line in real time, and transmit the original data to the edge computing and processing module;

[0054] Edge computing and processing module: built-in lightweight feature extraction algorithm, calculates the kurtosis of the current waveform, the temperature change rate, the main frequency of the vibration signal, the harmonic distortion rate and the amplitude fluctuation coefficient in real time, denoises the data and generates an "abnormal data frame" including the time window before and after the abnormal event, and then transmits the filtered valid data frame to the cloud data analysis module;

[0055] Cloud data analysis module: Receive the valid data frames transmitted by the edge computing processing module, construct a dynamic threshold self-learning model, generate real-time warning thresholds that change with time, load, and environment based on historical data. At the same time, integrate an abnormal pattern recognition algorithm to detect multi-parameter coupling abnormalities, and transmit the processed status data and abnormal event information to the power grid topology modeling module and the associated status evaluation module.

[0056] Power grid topology modeling module: Based on GIS data and the distribution network SCADA system, construct a three-dimensional digital twin network containing the electrical connection relationships, physical location information, and logical control relationships of devices, generate a unique "topology fingerprint" for each device, mark the list of upstream and downstream devices, load influence range, and power flow transfer path, generate an associated weight matrix between devices, quantify the degree of association affected by electrical connection, physical proximity, and historical faults, and transmit the data to the associated status evaluation module.

[0057] Associated status evaluation module: Receive the status data from the cloud data analysis module and the topology model data from the power grid topology modeling module, construct a joint evaluation model using a graph convolutional neural network, take the status vector of a single device and the associated weight matrix between devices as inputs, aggregate the status information of adjacent devices, output the corrected device health score and fault propagation probability considering the influence of topological association, and transmit the evaluation results to the optimization decision-making module.

[0058] Optimization decision-making module: Receive the evaluation results from the associated status evaluation module, take "the least number of power outages" and "the optimal full life cycle cost of devices" as goals, combine the real-time operation constraints of the distribution network, generate an intelligent maintenance plan and operation adjustment strategy, simulate the strategy effect through the digital twin system, and automatically push the optimal plan to the operation and maintenance terminal.

[0059] In an embodiment of the present invention: In the multi-dimensional data acquisition module, a current sensor and a temperature sensor are integrated at the wire joint to synchronously collect the real-time current value and temperature value at the joint to capture the coupling relationship between the joint temperature rise and current abnormality caused by poor contact and overload. A vibration sensor is installed in the middle of the pole body and designed with a three-axis accelerometer to monitor the vibration signals of the line and the pole caused by external construction vibration, and identify mechanical stress abnormalities through changes in vibration frequency and amplitude. The ultraviolet imager is aimed at the surface of the insulator, and a ultraviolet photosensitive sensor is used to capture the ultraviolet light signal generated by corona discharge, and the degradation degree of the insulation performance of the insulator is evaluated by combining the discharge intensity and frequency.

[0060] In an embodiment of the present invention: In the edge computing processing module, the specific methods for calculating the kurtosis of the current waveform, the temperature change rate, the main frequency of the vibration signal, the harmonic distortion rate, and the amplitude fluctuation coefficient are as follows:

[0061] The kurtosis calculation for the current signal is obtained through the following formula:

[0062]

[0063] where K is the kurtosis value of the current waveform, N is the number of single sampling points of the current signal, x i is the current sampling value, μ is the mean value, σ is the standard deviation, which is used to identify the distortion of the current waveform caused by impulse loads. When the kurtosis value K exceeds 3.5, it is determined that the distortion of the current waveform is caused by impulse loads;

[0064] The calculation of the rate of change for the temperature signal is calculated through the following formula:

[0065]

[0066] where S is the rate of temperature change per unit time, ΔT represents the amount of temperature change per unit time, Δt represents the sampling interval time, T(t) is the temperature sampling value at the current moment, T(t - τ) is the temperature sampling value at the previous sampling period t - τ, and τ is the sampling interval, which is used to detect abnormal temperature rise events. When the rate of temperature change is greater than 5°C / s, an abnormal event mark is automatically triggered, and multi-sensor data 500 ms before and after the event is intercepted to generate an "abnormal data frame";

[0067] For the main frequency analysis of the vibration signal, the time-domain vibration signal is converted into a frequency-domain signal through fast Fourier transform, and the frequency component with the highest energy ratio is extracted as the main frequency f main to identify the vibration nature through the main frequency:

[0068] The main frequency of line vibration is 2 Hz - 5 Hz, corresponding to the low-frequency vibration of the conductor under the action of wind;

[0069] The main frequency of external construction vibration is 10 Hz - 20 Hz, corresponding to the high-frequency vibration generated by mechanical impact;

[0070] The harmonic distortion rate for the voltage signal is calculated through the following formula:

[0071]

[0072] where THD represents the total harmonic distortion rate, V n is the amplitude of each harmonic voltage, V1 is the amplitude of the fundamental voltage, N is the highest order of harmonics calculated. When THD exceeds the national standard limit, it is determined that the power quality is abnormal;

[0073] The amplitude fluctuation coefficient is the ratio of the amplitude difference between adjacent sampling points to the time interval Calculation, where A(t) is the signal amplitude at the current moment, τ represents the sampling period. When the fluctuation coefficient exceeds 10% / ms, it is determined that the signal has a rapid abnormal fluctuation, and the data frame screening mechanism is triggered.

[0074] In one embodiment of the present invention: in the cloud data analysis module, the dynamic threshold self-learning model generates a real-time warning threshold through a dual mechanism of time dimension and space dimension:

[0075] In the time dimension, based on the historical 24-hour load-temperature data, the temperature-load correlation curve of the peak period from 18:00 to 22:00 and the valley period from 0:00 to 6:00 of the load is fitted, and the temperature warning threshold of the corresponding period is automatically adjusted according to the real-time load rate. For example, when the load rate exceeds 80%, the temperature threshold is dynamically increased from 80°C to 85°C;

[0076] In the space dimension, when the environmental parameters are detected to be abnormal (wind speed > 15 m / s, humidity > 90%) by 3 or more sensors adjacent to a certain device, according to the preset neighborhood influence rule, the temperature threshold of the device is dynamically corrected by ±10°C through the fuzzy logic algorithm. For example, in a high wind speed environment, the vibration of the wire may cause the joint to heat up due to friction, and at this time, the temperature threshold is automatically increased by 10°C to reduce the misjudgment caused by environmental interference.

[0077] In one embodiment of the present invention: the device association weight matrix generated by the power grid topology modeling module is linearly combined by the electrical connection weight, the physical proximity weight, and the historical fault association weight. The specific formula is as follows:

[0078] W ij = α·W 电气 + β·W 物理 + γ·W 历史 ;

[0079] Among them, W ij represents the association weight between device i and device j, and W 电气 represents the electrical connection weight. When device i and device j are directly electrically connected, that is, they are on the same electrical branch and there is no other device blocking the electrical characteristics in the middle, W 电气 takes the value of 0.8. When device i and device j are indirectly electrically connected through other devices, W 电气 takes the value of 0.3;

[0080] W 物理 represents the physical proximity weight. When the physical distance between device i and device j is less than 50 meters, E 物理 takes the value of 0.5. When the physical distance between device i and device j is between 50 meters and 200 meters, W 物理 takes the value of 0.2. When the physical distance between device i and device j exceeds 200 meters, W物理 The value is 0;

[0081] W 历史 represents the historical fault correlation weight. When devices i and j have alarmed due to the same fault chain within the past 12 months, that is, the time interval between the fault occurrences of the two is less than 24 hours and there is a logical correlation in the fault causes, then W 历史 takes the value of 0.4. If there is no historical correlation, then W 历史 takes the value of 0;

[0082] The weight coefficients α, β, and γ satisfy α + β + γ = 1 and are determined through machine learning training on the historical fault data of the past 24 months. The training process is as follows: First, classify and organize the historical fault data according to the electrical connection, physical location, and fault correlation of the devices to construct a training data set. Then, use the support vector machine algorithm to train the training data set, and adjust the values of α, β, and γ with the goal of the highest fault prediction accuracy to finally determine the optimal weight coefficient combination.

[0083] In an embodiment of the present invention: In the association status evaluation module, the graph convolutional neural network joint evaluation model determines the association evaluation of the device status through the following specific mechanism: First, construct a status vector S i = [h i , l i , e i for each device:

[0084] Among them, h i is the device health score, and the scoring range is 0 - 100 points. The value is proportional to the device health status. l i is the load rate, that is, the ratio of the actual load of the current device to the rated load, which reflects the load degree of the device. e i is the comprehensive value of environmental parameters, which is obtained by fusing environmental parameters and is used to reflect the comprehensive impact of environmental factors on the device status. Then, perform graph convolutional operation, and its specific formula is as follows:

[0085] S i ' = σ(∑ j∈N(i) W ij ·S j + W ii *S i );

[0086] Among them, N(i) represents the set of adjacent devices of device i, that is, the set composed of devices that are associated with device i in terms of electrical connection, physical location, and logical control. W ij is the association weight between devices, which is comprehensively determined according to the electrical connection tightness, physical distance, and historical fault correlation between devices and is used to measure the influence degree of adjacent device j on the status of device i. Wii is the associated weight of device i itself. σ is the ReLU activation function, which can introduce non-linear transformation and enhance the expression ability of the model. Through this graph convolution operation, the state information of adjacent devices is weighted and aggregated, and the corrected device state S i ′ is calculated and output.

[0087] In an embodiment of the present invention: in the optimization decision-making module, the intelligent maintenance plan generated by the optimization decision-making module is implemented based on an integer programming model, and the global optimization of maintenance resources is specifically achieved through the following mechanism:

[0088] Construct an objective function min(C 停电 +C 检修 ) with the goal of minimizing the "power outage impact cost" and the "maintenance implementation cost", where:

[0089] C 停电 is the user impact cost caused by the power outage, which is calculated by weighting the user load level, the power outage duration, and the number of affected users. The specific calculation formula is as follows:

[0090]

[0091] λ k is the impact coefficient of the k-th type of user, t k is the power outage duration of the k-th type of user, n k is the number of the k-th type of users, and τ is the impact unit price per household per hour;

[0092] C 检修 is the maintenance implementation cost, which covers the equipment replacement cost, the labor cost, and the material transportation cost. Specifically:

[0093]

[0094] is the replacement cost of the i-th device, is the labor cost of the i-th device, is the material cost of the i-th device, and x i is a 0-1 decision variable, where 1 indicates that the i-th device is to be maintained, and 0 indicates that it is not to be maintained;

[0095] Grid operation constraints: load transfer capacity limit, equipment rated capacity constraint, and power outage time window limit during maintenance;

[0096] Cooperative constraints of associated devices: the maintenance plans of adjacent devices need to be merged into the same power outage window to reduce repeated power outages. Adjacent devices are those with a physical distance < 50 meters or direct electrical connection;

[0097] Objective function solution algorithm

[0098] The branch and bound method is used to solve the integer programming model, and the specific process is as follows:

[0099] (1) Solving the relaxation problem: First, ignore the integer constraints and solve the linear programming relaxation problem to obtain an initial feasible solution;

[0100] (2) Branching process: Branch the non-integer decision variable x i to generate sub-problems x i = 0 and x i = 1;

[0101] (3) Bounding and pruning: Calculate the lower bound of the sub-problem. If the lower bound is higher than the current optimal solution, prune it; otherwise, continue branching until all sub-problems are solved to obtain the global optimal solution;

[0102] Strategy generation and simulation verification

[0103] Input the obtained maintenance plan into the digital twin system to simulate the following content:

[0104] (1) The power flow transfer path and equipment load changes during maintenance;

[0105] (2) The power outage impact range and load transfer effect;

[0106] (3) The time cost of maintenance implementation and the rationality of resource scheduling (including the optimization of manual working hours and material transportation paths).

[0107] By comparing the simulation results of different schemes (including power outage duration and cost consumption), automatically select the scheme with "the fewest power outage times" and "the optimal full life cycle cost".

[0108] In one embodiment of the present invention: The fault propagation probability output by the associated state evaluation module is realized by constructing a fault propagation path model through a Petri net. At least 3 possible propagation paths are generated for each abnormal event, including electrical fault propagation, mechanical vibration conduction, and environmental impact diffusion, and the occurrence probability of each path is statistically calculated based on historical fault data for early warning of cascading faults.

[0109] In one embodiment of the present invention: The abnormal pattern recognition algorithm of the cloud data analysis module adopts a method combining isolation forest and dynamic time warping to perform unsupervised learning on multi-parameter time series, automatically identify the combined features of "normal current + sudden temperature rise + abnormal vibration", and generate a new type of fault warning signal.

[0110] In one embodiment of the present invention: The analysis and evaluation system further includes a federated learning collaboration module and a blockchain evidence storage module, where:

[0111] The federated learning collaboration module is deployed between the edge computing nodes and the cloud to build a distributed federated learning network. Each edge node trains a local anomaly detection model (such as Isolation Forest, LSTM time series model) based on local device data, and submits model parameter updates to the cloud through a secure aggregation protocol (differential privacy). When the cloud integrates the global model, an adaptive weight allocation mechanism is introduced to dynamically adjust the parameter fusion weights according to the data diversity of the edge nodes (device type, regional environment differences). The formula is as follows:

[0112]

[0113] Among them, D i and D j are the data sets of the i-th and j-th edge nodes respectively. The entropy value (D i ) and the entropy value (D j ) both reflect the complexity of data features, realizing the collaborative optimization of the cross-regional device status model and improving the generalization ability;

[0114] Blockchain evidence storage module:

[0115] Hash the multi-dimensional collected data and key analysis results (abnormal data frames, health scores) and upload them to the blockchain. Adopt the consortium blockchain architecture to allow power grid operators, equipment manufacturers and regulatory agencies to share data evidence through smart contracts to ensure the immutability of data. The formula is:

[0116] Hash(data i ,timestamp i )→Block n ←Block n-1 ;

[0117] Among them, a chain structure is constructed through the timestamp i and the previous block hash value Block n-1 to provide a credible basis for fault tracing and liability determination.

[0118] Embodiment, please refer to the appendix Figure 1 , for the monitoring and early warning of the poor contact fault of the wire joint;

[0119] The feeder L1 of the 10kV distribution network in City A contains 50 poles. The wire joint of the #15 pole has an oxidation hazard due to long-term operation. It is necessary to monitor the current-temperature coupling anomaly at the joint in real time through this system and give an early warning of the poor contact fault;

[0120] The multi-dimensional data acquisition module is deployed

[0121] Sensor configuration:

[0122] Current sensor: Install a Rogowski coil current sensor (model: RC-100A, accuracy ±0.5%, sampling frequency 1 kHz, the sensor model is only an example, and those skilled in the art can replace it with an equivalent device according to actual needs) at the wire joint of Tower #15 to synchronously collect the three-phase current waveforms;

[0123] Temperature sensor: Use an infrared temperature measurement module (model: TMD-20, accuracy ±1°C, response time 200 ms), which is closely attached to the joint surface through thermal conductive silicone, and collect the temperature once every 50 ms;

[0124] Vibration sensor: Install a three-axis accelerometer (model: ADXL345, range ±16g, sampling frequency 500 Hz) in the middle of the tower to monitor wire vibration or external vibration;

[0125] Parameter acquisition process:

[0126] The sensor sends the original data to the edge computing terminal (model: ECU-2000) through the RS-485 bus, along with a hardware timestamp (accuracy ±1 μs);

[0127] Data processing by the edge computing processing module

[0128] Feature extraction algorithm:

[0129] Calculation of current kurtosis:

[0130]

[0131] Among them, the number of single sampling points N = 1024, x i is the current sampling value, x i = [10.1, 10.2, 10.3,...], and the calculation method of the mean value μ is:

[0132]

[0133] The calculation method of the standard deviation σ is:

[0134]

[0135] Then:

[0136]

[0137] If K > 3.5, it is determined as an impact load.

[0138] Calculation of temperature change rate:

[0139] The temperature T(t) at the current moment t = 80°C, and the temperature sampling value at the previous sampling period t - τ (τ = 50 ms) T(t - τ) is 75°C, then the temperature change rate:

[0140]

[0141] Since 100 °C / s is greater than 5 °C / s, an abnormal event flag is triggered, and data for 500 ms before and after the event is intercepted to generate an "abnormal data frame" (including current, temperature, and vibration signals).

[0142] Output result:

[0143] During normal operation, the aggregated steady-state data is uploaded every 10 seconds. When an abnormal data frame is detected, it is immediately uploaded to the cloud via the 5G network.

[0144] Dynamic early warning of the cloud data analysis module

[0145] Dynamic threshold adjustment:

[0146] Time dimension:

[0147] Load and temperature data of feeder L1 at different times in the past 3 years are collected, and a total of n = 10,000 groups of data are obtained. After linear regression fitting, for the peak period (18:00 - 22:00), the linear regression equation is obtained:

[0148] T = 0.5L + 20;

[0149] Among them, L is the load, and the current real-time load rate is 85%. Assuming the full load is 100 A, the current load L = 85 A. According to the fitting equation, the temperature threshold is calculated:

[0150] T = 0.5L + 20 = 0.5×85 + 20 = 62.5 °C;

[0151] Since the load rate exceeds 80%, the temperature threshold is dynamically reduced from 80 °C to 62.5 + 5 = 67.5 °C.

[0152] Spatial dimension:

[0153] The wind speed sensors near towers #14 and #16 detect a wind speed v = 18 m / s, and the number of adjacent towers n = 2. According to the fuzzy logic rule, it belongs to the situation of "if v ∈ V3 and n ∈ N3", and the temperature threshold adjustment amount ΔT = 10 °C. The joint temperature threshold is temporarily increased from 67.5 °C to 67.5 + 10 = 77.5 °C to avoid misjudging the temperature rise caused by vibration friction.

[0154] Abnormal mode recognition:

[0155] The collected vibration signals are processed by fast Fourier transform. Assume that the vibration signal x(t) is transformed by FFT to obtain the frequency spectrum x(f). The frequency corresponding to the maximum amplitude in the frequency spectrum is 3 Hz. Combining the current kurtosis K = 4.2 > 3.5 and the temperature change rate S = 100 °C / s > 5 °C / s calculated previously, it is determined as "initial fault of poor contact", and the warning level is set to level II (yellow warning).

[0156] Power grid topology modeling and correlation assessment

[0157] Topological fingerprint generation:

[0158] The "topological fingerprint" of this joint includes: electrical connection weight W 电气 = 0.8 (directly connected to the switch of tower #15), physical proximity weight W 物理 = 0.5 (the distances from tower #14 and tower #16 are both < 50 meters), historical fault correlation weight W 历史 = 0.4 (two similar faults occurred on the same feeder within the past 12 months).

[0159] After training with historical fault data, the optimal weight coefficients are determined as α = 0.4, β = 0.3, γ = 0.3, and the correlation weight matrix between devices:

[0160] W ij = α·W 电气 + β·W 物理 + γ·W 历史 = 0.4×0.8 + 0.3×0.5 + 0.3×0.4 = 0.59.

[0161] Graph convolutional network evaluation:

[0162] State vector S i = [h i = 70 (health degree), l i = 75% (load rate), e i = 0.8 (comprehensive value of high - temperature environment)];

[0163] Let the state vector of adjacent device j be S j = [h j = 65, l i = 70%, e j = 0.7];

[0164] W ij = 0.59, W ii = 0.8;

[0165] S i ′ = σ(∑ j∈N(i) W ij ·S j + W ii·S i ) =

[0166] [0.59×65 + 0.8×70, 0.59×0.7 + 0.8×0.75, 0.59×0.7 + 0.8×0.8]

[0167] = [93.35, 1.013, 1.013];

[0168] After being processed by the ReLU activation function σ(x) = max(0, x), the output is the corrected state vector S i ′ = [93.35, 1.013, 1.013], and the health degree h i = 93.35, indicating that the health state of the joint has further deteriorated due to the vibration of adjacent equipment.

[0169] Optimizing the decision-making module policy generation

[0170] Maintenance plan:

[0171] By calculating with the integer programming model, the objective function is min(C 停电 + C 检修 );

[0172] C 停电 is the cost of user impact caused by power outage. The power outage duration t = 2 hours, the number of affected users n = 120 households, and the impact coefficient λ k = 1.0, and the impact unit price per household per hour τ = 10 yuan / (household·hour). Then:

[0173] C 停电 = λ k × τ × t × n = 1.0 × 10 × 2 × 120 = 2400;

[0174] C 检修 is the maintenance implementation cost. The number of maintenance equipment m = 1, the material cost C 材料 = 5000 yuan, the man-hour h = 15 hours, the basic labor preparation cost per single equipment b = 500 yuan / unit, and the operation cost per unit man-hour d = 200 yuan / hour. Then:

[0175] C 检修 = b × m + C 材料 + d × h = 500 × 1 + 5000 + 200 × 15 = 8500 yuan;

[0176] The total cost of the original maintenance plan for this joint is:

[0177] C = C 停电 + C 检修 = 2400 + 8500 = 10900;

[0178] Compare the load transfer scheme (transfer through the tie switch of #10 pole tower, the power outage impact is reduced to 0 households, and only the loss cost of 2,000 yuan for transfer equipment is increased). The total cost of the load transfer scheme is C′ = 8,500 + 2,000 = 10,500 yuan;

[0179] Finally, a "load transfer + live maintenance" scheme is generated and pushed to the operation and maintenance terminal after being verified by digital twin simulation.

[0180] According to the content of the above embodiments, it can be concluded that through the multi-dimensional data acquisition module to collect real-time data at key positions of the distribution network, the feature extraction and data screening of the edge computing processing module, the dynamic threshold adjustment and abnormal pattern recognition of the cloud data analysis module, the construction of a three-dimensional digital twin network and the correlation weight matrix by the power grid topology modeling module, the global correlation analysis by the correlation state evaluation module, and the formulation of intelligent maintenance and operation strategies by the optimization decision-making module, the limitations of the traditional system in equipment evaluation, early warning and decision-making are effectively broken through.

[0181] Although the present invention is disclosed above in a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. An intelligent distribution network line status real-time monitoring, analysis and evaluation system, characterized in that, The analysis and evaluation system includes: A multi-dimensional data acquisition module: used to collect electrical parameters, equipment temperature, mechanical vibration, environmental humidity, and discharge signals of the line in real time; The edge computing and processing module is used to calculate the kurtosis of the current waveform, the temperature change rate, the main frequency of the vibration signal, the harmonic distortion rate, and the amplitude fluctuation coefficient in real time, generate an "abnormal data frame" containing abnormal events after noise reduction, and screen valid data frames; The cloud data analysis module receives the valid data frames, constructs a dynamic threshold self-learning model, generates real-time warning thresholds that change with time, load, and environment based on historical data, and at the same time integrates an abnormal pattern recognition algorithm to detect multi-parameter coupling abnormalities; The power grid topology modeling module constructs a three-dimensional digital twin network containing equipment electrical connection relationships, physical location information, and logical control relationships based on GIS data and the distribution network SCADA system, generates a unique "topological fingerprint" for each equipment, marks the list of upstream and downstream equipment, the load influence range, and the power flow transfer path, generates an association weight matrix between equipment, and quantifies the association degree of electrical connection, physical proximity, and historical fault influence; The association state evaluation module uses a graph convolutional neural network to construct a joint evaluation model, takes the state vector of a single equipment and the association weight matrix between equipment as inputs, aggregates the state information of adjacent equipment, outputs the corrected equipment health score and fault propagation probability considering the influence of topological association, and transmits the evaluation results; The optimization decision module aims at "the least number of power outages" and "the optimal full life cycle cost of equipment", combines the real-time operation constraints of the distribution network, generates an intelligent maintenance plan and operation adjustment strategy, and simulates the strategy effect through the digital twin system.

2. The real-time monitoring, analysis and evaluation system for the line status of an intelligent distribution network according to claim 1, wherein: In the multi-dimensional data acquisition module, the current sensor and the temperature sensor are integrated at the wire joint, and the vibration sensor is installed in the middle of the pole body. It is designed with a three-axis accelerometer to monitor the vibration signal of the line and the pole vibration caused by external construction vibration, and identify mechanical stress abnormalities through the changes in vibration frequency and amplitude.

3. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 2, characterized in that: In the edge computing and processing module, the specific methods for calculating the kurtosis of the current waveform, the temperature change rate, the main frequency of the vibration signal, the harmonic distortion rate, and the amplitude fluctuation coefficient are as follows: The harmonic distortion rate of the voltage signal is calculated by the following formula: Among them, THD represents the total harmonic distortion rate, V n is the amplitude of each harmonic voltage, V1 is the fundamental wave voltage amplitude, N is the highest order of harmonics calculated. When THD exceeds the national standard limit, it is used to evaluate the power quality. When THD in the low-voltage distribution system exceeds 5%, and in the medium-voltage and high-voltage systems exceeds 3%, it is determined that the power quality is abnormal; The amplitude fluctuation coefficient is calculated by the ratio of the amplitude difference between adjacent sampling points to the time interval. Among them, A(t) is the signal amplitude at the current moment, and τ represents the sampling period. When the fluctuation coefficient exceeds 10% / ms, it is determined that the signal has rapid abnormal fluctuations, and the data frame screening mechanism is triggered.

4. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 3, characterized in that: In the cloud data analysis module, the dynamic threshold self-learning model generates real-time warning thresholds through a dual mechanism of time dimension and space dimension: In the time dimension, based on the historical 24-hour load-temperature data, fit the temperature-load correlation curves for the peak period from 18:00 to 22:00 and the trough period from 0:00 to 6:00 of the load, and automatically adjust the temperature warning thresholds for the corresponding periods according to the real-time load rate; In the space dimension, when an abnormal environmental parameter is detected by 3 or more sensors adjacent to a certain equipment, according to the preset neighborhood influence rule, the temperature threshold of the equipment is dynamically corrected by ±10°C through the fuzzy logic algorithm.

5. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 4, characterized in that: The association weight matrix between equipment generated by the power grid topology modeling module is composed of a linear combination of electrical connection weights, physical proximity weights, and historical fault association weights. The specific formula is as follows: W ij = α·W 电气 + β·W 物理 + γ·W 历史 ; Among them, W ij represents the association weight between device i and device j, and W 电气 represents the electrical connection weight. When device i and device j are directly electrically connected, that is, in the same electrical branch and there is no other device affecting the electrical characteristics blocking in the middle, W 电气 takes the value of 0.

8. When device i and device j are indirectly electrically connected through other devices, W 电气 takes the value of 0.3; W 物理 represents the physical proximity weight. When the physical distance between device i and device j is less than 50 meters, W 物理 takes a value of 0.5; When the physical distance between device i and device j is between 50 meters and 200 meters, W 物理 takes the value of 0.2; When the physical distance between device i and device j exceeds 200 meters, W 物理 takes the value of 0; W 历史 represents the historical fault correlation weight. When devices i and j have reported alarms due to the same fault chain in the past 12 months, that is, the time interval between the occurrences of their faults is less than 24 hours and there is a logical correlation between the fault causes, W 历史 takes the value of 0.

4. If there is no historical correlation, then W 历史 takes the value of 0; The weight coefficients α, β, and γ satisfy α + β + γ = 1 and are determined through machine learning training on historical fault data over the past 24 months.

6. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 5, characterized in that: In the associated state evaluation module, the graph convolutional neural network joint evaluation model determines the associated evaluation of the device state through the following specific mechanism. First, a state vector S is constructed for each device i =[h i ,l i ,e i : Among them, h i is the device health score, with a scoring range of 0 - 100 points, and the value is proportional to the device health status. l i is the load rate, that is, the ratio of the actual load of the current device to the rated load, which reflects the load level of the device. e i is the comprehensive value of environmental parameters, which is output after the environmental parameter fusion process and is used to reflect the comprehensive impact of environmental factors on the device state; Then, perform graph convolution operation, and its specific formula is as follows: S i′ = σ(∑ j∈N(i) W ij ·S j + W ii ·S i ); Among them, N(i) represents the set of adjacent devices of device i, that is, the set composed of devices that are associated with device i in terms of electrical connection, physical location, and logical control. W ij is the association weight between devices, which is used to measure the influence degree of adjacent device j on the state of device i. W ii is the association weight of device i itself. σ is the ReLU activation function. Through this graph convolution operation, the state information of adjacent devices is weighted and aggregated to calculate and output the corrected device state S i′ .

7. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 6, characterized in that: In the optimization decision-making module, the intelligent maintenance plan generated by the optimization decision-making module is based on an integer programming model and specifically realizes the global optimization of maintenance resources through the following mechanism: Construct an objective function with the goal of minimizing the "power outage impact cost" and the "maintenance implementation cost": min(C 停电 +C 检修 ), where: C 停电 The user impact cost caused by power outages is calculated by weighting the user load level, power outage duration, and the number of affected users. The specific calculation formula is as follows: λ k is the influence coefficient for the k-th type of user, t k is the power outage duration for the k-th type of user, n k is the number of the k-th type of users, and τ is the influence unit price per household per hour; C 检修 It is the maintenance implementation cost, covering equipment replacement cost, labor hour cost and material transportation cost, specifically: is the replacement cost of the i-th device, is the labor cost of the i-th device, is the material cost of the i-th device, x i is a 0-1 decision variable, where 1 means the i-th device is repaired and 0 means it is not repaired.

8. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 7, characterized in that: The fault propagation probability output by the associated state evaluation module is realized by constructing a fault propagation path model through a Petri net. At least three propagation paths are generated for each abnormal event, including electrical fault propagation, mechanical vibration conduction, and environmental impact diffusion. And the occurrence probabilities of each path are statistically calculated based on historical fault data for early warning of cascading faults.

9. The real-time monitoring, analysis and evaluation system for the line status of an intelligent distribution network according to claim 8, wherein: The abnormal pattern recognition algorithm of the cloud data analysis module adopts a method combining isolation forest and dynamic time warping to perform unsupervised learning on multi-parameter time series, automatically identify the combined feature of "normal current + sudden temperature rise + abnormal vibration", and generate a new fault warning signal.

10. An intelligent distribution network line status real-time monitoring, analysis and evaluation system according to claim 9, characterized in that: The analysis and evaluation system also includes a federated learning collaboration module and a blockchain evidence storage module. The federated learning collaboration module is deployed between the edge computing node and the cloud to construct a distributed federated learning network. Each edge node trains a local anomaly detection model based on local device data and submits model parameter updates to the cloud through a secure aggregation protocol. When the cloud integrates the global model, an adaptive weight allocation mechanism is introduced to dynamically adjust the parameter fusion weight according to the data diversity of the edge nodes. The blockchain evidence storage module performs hash chain storage on multi-dimensional collected data and key analysis results, adopts a consortium chain architecture, and allows power grid operators, equipment manufacturers, and regulatory agencies to share data evidence storage through smart contracts to ensure the immutability of data.

Images