Power distribution automation terminal fault detection method and system

Abstract

The invention discloses a power distribution automation terminal fault detection method and system, and relates to the technical field of fault detection, and the key points of the technical scheme are that multi-modal data of a power distribution automation terminal are collected in real time through a multi-modal sensor array, and the multi-modal data comprise operation environment parameters, electrical quantity parameters and network topology data; constructing a dynamic coupling factor matrix by quantifying a nonlinear association relationship between the operating environment parameters and the electrical quantity parameters; adjusting a fault detection threshold in combination with the dynamic coupling factor matrix and the network topology data; and based on the adjusted fault detection threshold, performing fault diagnosis on the multi-modal data by using a hierarchical neural network model to generate a preliminary fault positioning result. According to the method, synchronous optimization of fault detection sensitivity and specificity in a complex distribution network environment is realized, and the problems of environment-electrical coupling failure, topology tracking lag, fault positioning fuzziness and the like in the prior art are systematically solved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and more specifically, to a fault detection method and system for a distribution automation terminal. Background Art

[0002] The distribution automation terminal is the core equipment used to realize the automated operation of the distribution network in the power system. It is usually deployed at key nodes such as switch stations, ring main units, pole-mounted switches, and distribution transformers in the distribution network, and undertakes functions such as real-time monitoring, fault detection, remote control, and data interaction. With the improvement of the intelligent level of the distribution automation terminal, traditional fault detection methods are difficult to cope with the multi-modal coupling disturbance problems brought by the high proportion of new energy access.

[0003] At present, some existing technologies record a fault detection method using an environmental parameter and electrical quantity threshold segmented correction strategy, but it does not consider the non-linear time-varying coupling characteristics of environmental factors such as temperature and humidity and harmonic parameters, and is prone to protection misoperation under complex working conditions such as the plum rain season. In addition, some existing technologies also record an overcurrent protection setting method based on topology recognition, but its equivalent impedance calculation depends on a fixed topology template and cannot track the network structure changes caused by the dynamic grid connection of distributed power sources in real time, which is likely to lead to a large deviation in short-circuit capacity calculation. In addition, some existing technologies also use a CNN fault diagnosis model to capture transient waveform features, but it does not construct a joint analysis mechanism for the time-series characteristics of environmental parameters and topological correlation, and there is a positioning fuzzy area in the multi-point grounding fault scenario.

[0004] Therefore, how to research and design a fault detection method and system for a distribution automation terminal that can overcome the above defects is an urgent problem for us to solve at present. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the purpose of the present invention is to provide a fault detection method and system for a distribution automation terminal. Through the collaborative innovation of multi-modal data fusion modeling, topology adaptive threshold setting, and hierarchical intelligent diagnosis, the synchronous optimization of fault detection sensitivity and specificity in a complex distribution network environment is achieved, providing a reliable fault handling technology system for an intelligent distribution network with a high proportion of new energy access, and systematically solving problems such as environmental-electrical coupling failure, topology tracking lag, and fault location ambiguity in the prior art.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In the first aspect, a fault detection method for a distribution automation terminal is provided, including the following steps:

[0008] Collect multimodal data of the distribution automation terminal in real time through a multimodal sensor array, where the multimodal data includes operating environment parameters, electrical quantity parameters, and network topology data;

[0009] Construct a dynamic coupling factor matrix by quantifying the non - linear correlation relationship between the operating environment parameters and the electrical quantity parameters;

[0010] Adjust the fault detection threshold by combining the dynamic coupling factor matrix and the network topology data;

[0011] Based on the adjusted fault detection threshold, use a hierarchical neural network model to perform fault diagnosis on the multimodal data and generate a preliminary fault location result.

[0012] Further, the operating environment parameters include one or more of temperature, humidity, dust concentration, and equipment deformation;

[0013] And, the electrical quantity parameters include one or more of current, voltage, harmonics, and transient waveforms.

[0014] Further, the construction expression of the dynamic coupling factor matrix is:

[0015] C ij =w i f(E i )+λ j g(Q j )+θ ij f(E i )g(Q j );

[0016] Among them, C ij represents the dynamic coupling factor between the i - th type of operating environment parameter and the j - th type of electrical quantity parameter in the dynamic coupling factor matrix; w i represents the dynamic weight coefficient of the i - th type of operating environment parameter; f(E i ) represents the first quantization function of the i - th type of operating environment parameter E i ; λ j represents the dynamic weight coefficient of the j - th type of electrical quantity parameter; g(Q j ) represents the second quantization function of the j - th type of electrical quantity parameter Q j ; θ ij represents the dynamic weight coefficient of the cross - term between the i - th type of operating environment parameter and the j - th type of electrical quantity parameter.

[0017] Further, the adjusting the fault detection threshold by combining the dynamic coupling factor matrix and the network topology data includes:

[0018] Extract the network connection relationships in the network topology data, and generate a topology graph including nodes and branches based on the network connection relationships;

[0019] Use a graph embedding algorithm to convert the topology graph into a topology vector representing the network structure characteristics;

[0020] Calculate the topology difference degree based on the topology vectors at adjacent times, and generate a threshold correction instruction when the topology difference degree is greater than or equal to the difference degree threshold.

[0021] Furthermore, the topology difference degree is the Euclidean distance between the topology vectors at adjacent times.

[0022] Furthermore, the topology difference degree is obtained by performing weighted calculation on the instantaneous difference and trend difference between the topology vectors at adjacent times;

[0023] Among them, the instantaneous difference is the normalized parameter instantaneous fluctuation, and the trend difference is the relative amplitude of the parameter mean change.

[0024] Furthermore, if the fault detection threshold is the overcurrent protection threshold, the adjustment expression is:

[0025]

[0026] Among them, I′ set represents the dynamically adjusted overcurrent protection threshold; I set represents the original overcurrent protection setting value under the reference network topology and standard environment; Z eqb represents the equivalent impedance under the reference network topology; Z eqc represents the equivalent impedance under the current real-time network topology; C iI represents the dynamic coupling factor of the i-th operating environment parameter to the current I; ΔE i represents the normalized offset of the i-th operating environment parameter relative to the reference value; n represents the number of environment parameters affecting the overcurrent protection threshold.

[0027] Furthermore, if the fault detection threshold is the low voltage protection threshold, the adjustment expression is:

[0028]

[0029] Among them, U′ set represents the dynamically adjusted low voltage protection threshold; U set represents the original low voltage protection setting value under the reference conditions; γ represents the empirical coefficient of the influence of the short-circuit capacity change on the voltage; C xV represents the dynamic coupling factor of the x-th operating environment parameter to the voltage V; S scn represents the short-circuit capacity under the current real-time network topology; Ssco represents the short-circuit capacity under the reference network topology; m represents the number of environmental parameters affecting the low-voltage protection threshold.

[0030] Furthermore, the hierarchical neural network model includes a lightweight CNN-LSTM hybrid model deployed on the terminal side and a GNN topology parsing engine deployed on the master station side;

[0031] The lightweight CNN-LSTM hybrid model extracts the transient features of the electrical quantity parameters and the time-series features of the operating environment parameters, and determines the fault probability based on the transient features and the time-series features;

[0032] And, the GNN topology parsing engine fuses the fault probabilities of all network nodes to optimize the fault probability and outputs the preliminary fault location result.

[0033] In a second aspect, a distribution automation terminal fault detection system is provided, which is used to implement a distribution automation terminal fault detection method as described in any one of the first aspects, including:

[0034] A data acquisition module, configured to collect multi-modal data of the distribution automation terminal in real time through a multi-modal sensor array, where the multi-modal data includes operating environment parameters, electrical quantity parameters, and network topology data;

[0035] A matrix construction module, configured to construct a dynamic coupling factor matrix by quantifying the non-linear correlation relationship between the operating environment parameters and the electrical quantity parameters;

[0036] A threshold adjustment module, configured to adjust the fault detection threshold by combining the dynamic coupling factor matrix and the network topology data;

[0037] A fault diagnosis module, configured to perform fault diagnosis on the multi-modal data by using the hierarchical neural network model based on the adjusted fault detection threshold, and generate a preliminary fault location result.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. A distribution automation terminal fault detection method provided by the present invention realizes the synchronous optimization of the fault detection sensitivity and specificity in a complex distribution network environment through the collaborative innovation of multi-modal data fusion modeling, topology adaptive threshold setting, and hierarchical intelligent diagnosis, provides a reliable fault handling technology system for the intelligent distribution network with high proportion of new energy access, and systematically solves the problems of environmental-electrical coupling failure, topology tracking lag, and fuzzy fault location in the prior art;

[0040] 2. The present invention realizes the tensor correlation analysis of temperature and humidity gradient changes and transient harmonic parameters for the first time by constructing a dynamic coupling factor matrix containing cross-term weights, solves the problem of large calculation deviation of the coupling degree in the previous linear correction model under complex working conditions, and can effectively improve the measurement accuracy of the environment-electrical coupling coefficient;

[0041] 3. The present invention uses a graph embedding algorithm to transform the real-time network topology into a 128-dimensional feature vector, dynamically triggers a threshold correction instruction by calculating the topological difference degree between adjacent moments, overcomes the short-circuit capacity calculation error caused by a fixed topology template, can improve the tracking response speed of the impedance correction factor, and can improve the accuracy of overcurrent protection action at the same time;

[0042] 4. The hierarchical neural network model described in the present invention can effectively compress the fuzzy area of multi-point grounding fault location to improve the detection timeliness through feature extraction on the terminal side and topology optimization on the master station side, and through the joint optimization of local transient features and global fault probabilities, realizes the "edge perception-cloud optimization" closed-loop of power system fault detection, can balance real-time performance and accuracy, and provides a reliable intelligent fault handling ability for the distribution network with high proportion of new energy access. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0044] Figure 1 is the flowchart in Embodiment 1 of the present invention;

[0045] Figure 2 is the system block diagram in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with the embodiments and the drawings. The illustrative embodiments and descriptions thereof of the present invention are only used to explain the present invention and do not limit the present invention.

[0047] Embodiment 1: A method for fault detection of a distribution automation terminal, as Figure 1 shown, includes the following steps:

[0048] S1: Real-time collect multi-modal data of the distribution automation terminal through a multi-modal sensor array, and the multi-modal data includes operating environment parameters, electrical quantity parameters, and network topology data;

[0049] S2: Construct a dynamic coupling factor matrix by quantifying the non-linear correlation relationship between the operating environment parameters and the electrical quantity parameters;

[0050] S3: Adjust the fault detection threshold by combining the dynamic coupling factor matrix and the network topology data;

[0051] S4: Based on the adjusted fault detection threshold, use a hierarchical neural network model to perform fault diagnosis on multi-modal data and generate a preliminary fault location result.

[0052] In step S1, the operating environment parameters include, but are not limited to, one or more of temperature, humidity, dust concentration, and equipment deformation; and, the electrical quantity parameters include one or more of current, voltage, harmonics, and transient waveforms.

[0053] In some examples, the multi-modal sensor array adopts an integration of a MEMS (Micro-Electro-Mechanical System) sensor array and a fiber Bragg grating sensor. The MEMS sensor array is, for example, an SHT45 temperature and humidity sensor and a PM2.5 laser dust sensor, and the fiber Bragg grating sensor monitors equipment deformation and stress.

[0054] In addition, the multi-modal sensor array samples at a low frequency of 1Hz during steady-state operation; when an environmental mutation is detected, it automatically switches to high-frequency sampling at 100Hz to capture transient changes; environmental mutations such as a temperature gradient > 5°C / min.

[0055] In some examples, the electrical quantity parameters can adopt a dual-channel redundant acquisition architecture. The main channel in the dual-channel can use an AD7779 to achieve 24-bit synchronous sampling and capture data acquisition such as voltage, current, and harmonics (0 - 5kHz); while the auxiliary channel in the dual-channel uses a Hall sensor and an ADS8588S to monitor transient impact current for fault recording trigger.

[0056] In some examples, the network topology data can be collected by the active reporting method of the GOOSE (Generic Object Oriented Substation Event) protocol, achieving real-time performance at the 1ms level. For example, the intelligent terminal subscribes to the switch position change information published by the SCADA (Supervisory Control and Data Acquisition) system. The intelligent terminal includes a distribution terminal unit DTU and a feeder terminal unit FTU.

[0057] In some examples, the network topology data can also be collected through the ModbusTCP communication of the grid-connected inverter. For example, the inverter has a built-in communication module (such as RS485 to Ethernet) and periodically uploads the grid-connected status (sampling at 10Hz).

[0058] In step S2, when quantifying the non - linear correlation between the operating environment parameters and the electrical quantity parameters, the present invention synchronously considers the independent influence of the operating environment parameters, the independent influence of the electrical quantity parameters, and the non - linear coupling effect between the operating environment parameters and the electrical quantity parameters. For example, humidity and harmonics accelerate metal corrosion.

[0059] In some examples, the construction expression of the dynamic coupling factor matrix is:

[0060] C ij =w i f(E i )+λ j g(Q j )+θ ij f(E i )g(Q j );

[0061] Wherein, C ij represents the dynamic coupling factor between the i - th type of operating environment parameter and the j - th type of electrical quantity parameter in the dynamic coupling factor matrix; w i represents the dynamic weight coefficient of the i - th type of operating environment parameter; f(E i ) represents the first quantization function of the i - th type of operating environment parameter E i ; λ j represents the dynamic weight coefficient of the j - th type of electrical quantity parameter; g(Q j ) represents the second quantization function of the j - th type of electrical quantity parameter Q j ; θ ij represents the dynamic weight coefficient of the cross - term between the i - th type of operating environment parameter and the j - th type of electrical quantity parameter.

[0062] In some examples, if the operating environment parameter is temperature, the corresponding first quantization function can adopt linear normalization. For example, the temperature is mapped to the interval [-1, 1] to quantify the degree of deviation from the reference operating condition. The specific expression is:

[0063]

[0064] Wherein, f(E 1 ) represents the first quantization function corresponding to the temperature; T(t) represents the real - time temperature measurement value; T ref represents the reference temperature benchmark, such as taking a value of 25 °C; T max represents the temperature upper - limit threshold, such as taking a value of 70 °C; T min represents the temperature lower - limit threshold, such as taking a value of - 40 °C.

[0065] In some examples, if the operating environment parameter is humidity, the corresponding first quantization function can adopt saturation non - linearity. For example, the humidity saturation effect is simulated by the hyperbolic tangent function. The specific expression is:

[0066]

[0067] Among them, f(E 2 ) represents the first quantization function corresponding to humidity; H(t) represents the real-time humidity measurement value; H crit represents the humidity critical threshold.

[0068] In some examples, if the operating environment parameter is the dust concentration, the corresponding first quantization function can adopt an exponential decay model, aiming to characterize that the impact of high-concentration dust on the equipment decays exponentially, avoiding over-sensitivity of the model under extremely high concentrations. The specific expression is:

[0069]

[0070] Among them, f(E 3 ) represents the first quantization function corresponding to the dust concentration; D(t) represents the real-time dust concentration; D crit represents the critical threshold of the dust concentration, such as taking a value of 1000 μg / m 3 .

[0071] In some examples, if the operating environment parameter is the equipment deformation amount, the corresponding first quantization function can adopt linear normalization to the deformation failure threshold. The specific expression is:

[0072]

[0073] Among them, f(E 4 ) represents the first quantization function corresponding to the equipment deformation amount; S(t) represents the measured equipment deformation amount; S ref represents the initial deformation amount; S fail represents the deformation failure threshold.

[0074] In some examples, if the electrical quantity parameter is current, the corresponding second quantization function can be normalized to the rated current. The specific expression is:

[0075]

[0076] Among them, g(Q 1 ) represents the second quantization function corresponding to the current; represents the instantaneous current change rate; I ra represents the rated current.

[0077] In some examples, if the electrical quantity parameter is voltage, the corresponding second quantization function can adopt a relative value based on the nominal voltage. The specific expression is:

[0078]

[0079] where, g(Q 2 ) represents the second quantization function corresponding to the voltage; V(t) represents the real-time voltage measurement value; V n represents the nominal voltage.

[0080] In some examples, if the electrical quantity parameter is a harmonic, the corresponding second quantization function can directly adopt the measured value of the total harmonic distortion (THD), and the specific expression is:

[0081] g(Q 3 ) = THD(t);

[0082] where, g(Q 3 ) represents the second quantization function corresponding to the harmonic; THD(t) represents the harmonic distortion rate.

[0083] In some examples, if the electrical quantity parameter is a transient waveform, the corresponding second quantization function can be the Shannon entropy calculated for the transient current, and the specific expression is:

[0084] g(Q 4 ) = H s (I tr );

[0085] where, g(Q 4 ) represents the second quantization function corresponding to the transient waveform; H s (I tr ) represents the Shannon entropy of the transient current I tr .

[0086] In step S3, adjusting the fault detection threshold by combining the dynamic coupling factor matrix and the network topology data includes: extracting the network connection relationship in the network topology data, and generating a topology graph including nodes and branches according to the network connection relationship; using a graph embedding algorithm to convert the topology graph into a topology vector representing the network structure characteristics; calculating the topology difference degree according to the topology vectors at adjacent moments, and generating a threshold correction instruction when the topology difference degree is greater than or equal to the difference degree threshold.

[0087] In some examples, the nodes include but are not limited to distribution equipment nodes, sensor nodes, and power supply nodes. For example, all device IDs are extracted from the device information table as the topology graph nodes, and the types are marked such as terminals, fuses, circuit breakers, etc. The sensor IDs can also be mounted as affiliated nodes under the main device nodes to form a "device-sensor" hierarchical relationship. In addition, a substation or a superior power supply node (such as "A Substation") can be added as the root node of the topology graph.

[0088] In some examples, the branch includes but is not limited to physical connection branches, logical on / off states, and sensor data associations. For example, physical connections between devices are generated according to adjacent device fields, such as cables and busbars. The branch can also be marked as effectively connected or disconnected by combining the on / off states in real-time data. In addition, the sensor deformation can be mapped to the corresponding device node attributes.

[0089] In some examples, the graph embedding algorithm can adopt GraphSAGE (Graph Sample and Aggregate). GraphSAGE extracts topological features layer by layer from local to global through neighbor sampling and feature aggregation, and generates low-dimensional vectors.

[0090] Specifically, the initial features of each node include two types of information: static attributes and dynamic states. Static attributes such as device type (terminal / fuse / circuit breaker), rated current, and voltage. Dynamic states such as real-time switch state (0 / 1), power of photovoltaic inverter, etc.

[0091] Then, neighbor aggregation processing is performed on the nodes. The specific expression is:

[0092]

[0093] Among them, represents the feature vector of node r after the k-th iteration; W agg represents the trainable weight matrix; represents the feature vector of node r after the (k - 1)-th iteration; represents the feature vector of node s after the k-th iteration; Ν(r) represents the set of direct neighbors of node r; σ(·) represents the activation function; MEAN(·) represents the mean aggregation operation of neighbor node features; CONCAT(·) represents the concatenation operation.

[0094] After K iterations, the feature vectors of node r are concatenated into the global topological vector V t = [v 1 , v 2 ,..., v K , and the dimension is fixed at 128.

[0095] Among them, v K is the K-th dimension in the global topological vector. For example, v 1 - v 32 characterizes features such as node degree distribution and hub node identification, such as tie switches and substations; v 33 - v 64 characterizes features such as loop network features and the hierarchical depth of radial networks. Loop network features such as loop impedance distribution; v 65 - v96 The characterized features are DG access features, such as the inverter power direction marker and the output fluctuation frequency; v 97 -v 128 The characterized features are the fault propagation path weight and the probability of the overload hot spot area.

[0096] In some examples, the topological difference degree is the Euclidean distance between the topological vectors at adjacent moments, and the specific expression is:

[0097] ΔT=||V t -V t-Δt || 2 ;

[0098] Among them, ΔT represents the topological difference degree; V t represents the topological vector at the current moment; V t-Δt represents the topological vector at the previous moment.

[0099] In some examples, the topological difference degree can also be obtained by weighted calculation of the instantaneous difference and the trend difference between the topological vectors at adjacent moments; among them, the instantaneous difference is the normalized parameter instantaneous fluctuation, and the trend difference is the relative amplitude of the parameter mean change.

[0100] Specifically, the calculation expression of the topological difference degree is:

[0101]

[0102] Among them, Q represents the number of topological parameters participating in the calculation, and the topological parameters are such as node voltage, current, switch state, etc.; w q represents the weight of the qth topological parameter; V t,q represents the real-time measurement value of the qth topological parameter at time t; V t-Δt,q represents the measurement value of the qth topological parameter at time t - Δt, where Δt is the sliding window; δ q represents the statistical standard deviation of the qth topological parameter within the sliding window; λ represents the trend factor, which is the weight coefficient for balancing the instantaneous change and the long-term trend change; μ t,q represents the average value of the qth topological parameter within the current sliding window; μ t-Δt,q represents the average value of the qth topological parameter within the previous sliding window.

[0103] The present invention uses a graph embedding algorithm to transform the real-time network topology into a 128-dimensional feature vector, dynamically triggers a threshold correction instruction by calculating the topological difference degree at adjacent moments, overcomes the short-circuit capacity calculation error caused by the fixed topological template, can improve the tracking response speed of the impedance correction factor, and can improve the accuracy of the overcurrent protection action at the same time.

[0104] In some examples, if the fault detection threshold is the overcurrent protection threshold, the adjustment expression is:

[0105]

[0106] where I′ set represents the dynamically adjusted overcurrent protection threshold; I set represents the original overcurrent protection setting value under the reference network topology and standard environment; Z eqb represents the equivalent impedance under the reference network topology; Z eqc represents the equivalent impedance under the current real-time network topology; C iI represents the dynamic coupling factor of the i-th operating environment parameter to the current I; ΔE i represents the normalized offset of the i-th operating environment parameter relative to the reference value; n represents the number of environment parameters affecting the overcurrent protection threshold.

[0107] The overcurrent protection threshold correction logic with weighted dynamic coupling factors proposed by the present invention solves the problem of overlapping protection ranges caused by DG (distributed generation) grid connection by fusing the network impedance ratio and environmental offset in real time, and can effectively compress the protection action coordination step difference.

[0108] In some examples, if the fault detection threshold is the low voltage protection threshold, the adjustment expression is:

[0109]

[0110] where U′ set represents the dynamically adjusted low voltage protection threshold; U set represents the original low voltage protection setting value under the reference conditions; γ represents the empirical coefficient of the influence of short-circuit capacity change on voltage; C xV represents the dynamic coupling factor of the x-th operating environment parameter to the voltage V; S scn represents the short-circuit capacity under the current real-time network topology; S sco represents the short-circuit capacity under the reference network topology; m represents the number of environment parameters affecting the low voltage protection threshold.

[0111] In step S4, the hierarchical neural network model includes a lightweight CNN-LSTM hybrid model deployed on the terminal side and a GNN topology parsing engine deployed on the master station side; the lightweight CNN-LSTM hybrid model extracts the transient features of electrical quantity parameters and the time series features of operating environment parameters, and determines the fault probability based on the transient features and time series features; and, the GNN topology parsing engine fuses the fault probabilities of all network nodes to optimize the fault probability and outputs a preliminary fault location result.

[0112] Specifically, the lightweight CNN-LSTM hybrid model mainly extracts transient features from electrical quantity parameters, and combines with the time-series changes of operating environment parameters to calculate the local fault probability.

[0113] The CNN (Convolutional Neural Network) module in the lightweight CNN-LSTM hybrid model uses 3 layers of 1D convolution, with the number of channels being 16, 32, and 64 respectively, and the convolution kernel size being 5. After each layer, ReLU and MaxPooling (pooling size 2) are connected. In the present invention, depthwise separable convolution is used to reduce the computational amount, and the number of model parameters can be compressed to within 50KB. The LSTM (Long Short-Term Memory Network) module in the lightweight CNN-LSTM hybrid model uses a single-layer LSTM, with the number of its hidden units being 32, and the input being the time series of operating environment parameters. Finally, the 128-dimensional feature vector output by the CNN module is concatenated with the 32-dimensional vector output by the LSTM module and input into the fully connected layer; the finally output fault probability takes values in [0,1], indicating the fault possibility of the local node. The GNN (Graph Neural Network) topology parsing engine mainly fuses the fault probabilities of all nodes in the network, and then optimizes the positioning result in combination with the network topology relationship. For example, suppressing isolated false alarms and enhancing associated faults.

[0114] An innovative cascaded architecture of the terminal-side CNN-LSTM hybrid model (50KB lightweight design) and the master-station-side GNN topology parsing engine is designed. Through the joint optimization of local transient feature extraction (24-bit AD sampling) and the global fault probability propagation matrix, the fuzzy area of multi-point grounding fault location is reduced from 200 meters in the traditional method to within 50 meters, and the detection timeliness is improved by 60%.

[0115] The hierarchical neural network model described in the present invention can effectively compress the fuzzy area of multi-point grounding fault location through feature extraction on the terminal side and topology optimization on the master-station side, and through the joint optimization of local transient feature extraction and global fault probability, so as to improve the detection timeliness, realize the "edge perception-cloud optimization" closed-loop of power system fault detection, and can balance real-time performance and accuracy, providing a reliable intelligent fault handling ability for the distribution network with a high proportion of new energy access.

[0116] Embodiment 2: A distribution automation terminal fault detection system, which is used to implement a distribution automation terminal fault detection method as described in Embodiment 1, as Figure 2 shown, including a data acquisition module, a matrix construction module, a threshold adjustment module, and a fault diagnosis module.

[0117] Among them, the data acquisition module is used to collect multi-modal data of the distribution automation terminal in real time through a multi-modal sensor array. The multi-modal data includes operating environment parameters, electrical quantity parameters, and network topology data. The matrix construction module is used to construct a dynamic coupling factor matrix by quantifying the non-linear correlation relationship between the operating environment parameters and the electrical quantity parameters. The threshold adjustment module is used to adjust the fault detection threshold by combining the dynamic coupling factor matrix and the network topology data. The fault diagnosis module is used to perform fault diagnosis on the multi-modal data based on the adjusted fault detection threshold by using a hierarchical neural network model and generate a preliminary fault location result.

[0118] Working principle: Through the collaborative innovation of multi-modal data fusion modeling, topology adaptive threshold setting, and hierarchical intelligent diagnosis, the present invention realizes the synchronous optimization of the sensitivity and specificity of fault detection in a complex distribution network environment, provides a reliable fault handling technology system for an intelligent distribution network with a high proportion of new energy access, and systematically solves problems such as environmental-electrical coupling failure, topology tracking lag, and fuzzy fault location in the prior art.

[0119] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can 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.

[0120] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0121] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1The functions specified in one or more boxes.

[0122] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the steps of the functions specified in one Figure 1 one process or more processes and / or boxes Figure 1 step of the functions specified in one or more boxes.

[0123] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting faults in a distribution automation terminal, characterized in that: The following steps are involved: Collect multimodal data of distribution automation terminals in real time through a multimodal sensor array, wherein the multimodal data includes operating environment parameters, electrical quantity parameters and network topology data; Constructing a dynamic coupling factor matrix by quantifying the nonlinear correlation between the operating environment parameters and the electrical quantity parameters; adjusting a fault detection threshold in combination with the dynamic coupling factor matrix and the network topology data; Based on the adjusted fault detection threshold, a hierarchical neural network model is used to perform fault diagnosis on the multimodal data to generate a preliminary fault location result.

2. A distribution automation terminal fault detection method according to claim 1, characterized in that: The operating environment parameters include one or more of temperature, humidity, dust concentration and equipment deformation; Furthermore, the electrical quantity parameters include one or more of current, voltage, harmonics and transient waveforms.

3. A distribution automation terminal fault detection method according to claim 1, characterized in that: The construction expression of the dynamic coupling factor matrix is: C ij =w i f(E i )+λ j g(Q j )+θ ij f(E i )g(Q j ); Among them, C ij represents the dynamic coupling factor between the i-th type of operating environment parameter and the j-th type of electrical quantity parameter in the dynamic coupling factor matrix; w i represents the dynamic weight coefficient of the i-th type of operating environment parameters; f(E i ) represents the i-th type of operating environment parameter E i The first quantization function of j Indicates the dynamic weight coefficient of the jth type of electrical quantity parameter; g(Q j ) represents the jth electrical quantity parameter Q j The second quantization function of θ ij It represents the dynamic weight coefficient of the cross term between the i-th operating environment parameter and the j-th electrical quantity parameter.

4. A distribution automation terminal fault detection method according to claim 1, characterized in that: The adjusting the fault detection threshold in combination with the dynamic coupling factor matrix and the network topology data includes: Extracting the network connection relationship in the network topology data, and generating a topology graph including nodes and branches according to the network connection relationship; Using a graph embedding algorithm to convert the topological graph into a topological vector representing network structure characteristics; The topology difference is calculated according to the topology vectors at adjacent moments, and a threshold correction instruction is generated when the topology difference is greater than or equal to a difference threshold.

5. A distribution automation terminal fault detection method according to claim 4, characterized in that: The topological difference is the Euclidean distance between the topological vectors at adjacent moments.

6. A distribution automation terminal fault detection method according to claim 4, characterized in that: The topological difference is obtained by weighted calculation of the instantaneous difference and trend difference between the topological vectors at adjacent moments; The instantaneous difference is the normalized instantaneous fluctuation of the parameter, and the trend difference is the relative amplitude of the change in the mean value of the parameter.

7. A distribution automation terminal fault detection method according to claim 1, characterized in that: If the fault detection threshold is the overcurrent protection threshold, the adjustment expression is: Among them, I′ set Indicates the overcurrent protection threshold after dynamic adjustment; I set Represents the original overcurrent protection setting under the reference network topology and standard environment; Z eqb Represents the equivalent impedance under the benchmark network topology; Z eqc Represents the equivalent impedance under the current real-time network topology; C iI Indicates the dynamic coupling factor of the i-th operating environment parameter to the current I; ΔE i It represents the normalized offset of the ith operating environment parameter relative to the reference value; n represents the number of environmental parameters that affect the overcurrent protection threshold.

8. A distribution automation terminal fault detection method according to claim 1, characterized in that: If the fault detection threshold is a low voltage protection threshold, the adjustment expression is: Among them, U′ set Indicates the low voltage protection threshold after dynamic adjustment; U set represents the original low voltage protection setting under the reference conditions; γ represents the empirical coefficient of the influence of short-circuit capacity change on voltage; C xV represents the dynamic coupling factor of the xth operating environment parameter to the voltage V; S scn Represents the short-circuit capacity under the current real-time network topology; S sco represents the short-circuit capacity under the benchmark network topology; m represents the number of environmental parameters that affect the low voltage protection threshold.

9. A distribution automation terminal fault detection method according to claim 1, characterized in that: The hierarchical neural network model includes a lightweight CNN-LSTM hybrid model deployed on the terminal side and a GNN topology parsing engine deployed on the master station side; The lightweight CNN-LSTM hybrid model extracts the transient characteristics of the electrical quantity parameters and the time series characteristics of the operating environment parameters, and determines the fault probability according to the transient characteristics and the time series characteristics; Furthermore, the GNN topology parsing engine integrates the fault probabilities of all nodes in the entire network to optimize the fault probabilities and outputs the preliminary fault location results.

10. A distribution automation terminal fault detection system, characterized in that: The system is used to implement a distribution automation terminal fault detection method as described in any one of claims 1 to 9, comprising: A data acquisition module is used to collect multimodal data of the distribution automation terminal in real time through a multimodal sensor array, wherein the multimodal data includes operating environment parameters, electrical quantity parameters and network topology data; A matrix construction module, used for constructing a dynamic coupling factor matrix by quantifying the nonlinear correlation between the operating environment parameters and the electrical quantity parameters; A threshold adjustment module, used to adjust a fault detection threshold in combination with the dynamic coupling factor matrix and the network topology data; The fault diagnosis module is used to perform fault diagnosis on the multimodal data based on the adjusted fault detection threshold using a hierarchical neural network model to generate a preliminary fault location result.

Images