A switch cabinet unmanned inspection method and system

Abstract

The application relates to the technical field, in particular to a switch cabinet unmanned inspection method and system, which comprises the following steps: an association relationship between signal nodes is established through a graph neural network, the time sequence consistency of multi-source signals is improved, instrument readings and temperature and humidity signals can form a hierarchical structure under a unified time framework, the logical association of state layering is strengthened, continuous sampling sequence features are extracted through a long short-term memory network, the response sensitivity and time prediction capability of abnormal signal mutation are improved, the amplitude of temperature difference and current signals is matched under the same time index, early abnormal nodes are accurately identified and a dynamic positioning sequence is established, the path length and energy peak value are reduced through distance and energy joint weighting, the task throughput and equipment availability are improved, the time sequence capturing and trend judgment capability is enhanced in the abnormal identification layer, and the accuracy, continuity and safety evaluation reliability of the inspection decision are improved as a whole.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic inspection control, in particular to a switch cabinet unmanned inspection method and system. BACKGROUND

[0002] The technical field of automatic inspection control, specifically a switch cabinet equipment state detection and operation safety evaluation unmanned inspection method for power system, is based on automatic control technology, combined with equipment operation parameter acquisition, inspection path planning, state recognition algorithm and control logic, to realize periodic detection of internal and external components of high-voltage and medium-voltage switch cabinets. Automatic inspection control technology is usually applied to the operation management of substation, switch station and distribution network equipment, to replace traditional inspection methods, reduce misjudgment, and improve the real-time performance and safety of equipment monitoring. The technical field mainly studies how to realize autonomous inspection, abnormal state recognition, defect alarm and data archiving functions of equipment through control algorithm and sensing recognition module, forms a closed-loop control system, and realizes visualized and intelligent management of equipment state.

[0003] A switch cabinet unmanned inspection method is a technical solution that uses automatic control and intelligent recognition means to perform fixed-point, fixed-time and fixed-sequence inspection of power switch cabinets. The design purpose of the method is to realize automatic detection and abnormal recognition of the running state of the switch cabinet, avoid omissions and delays in the inspection process, and enable the inspection device to perform inspection tasks without human intervention by constructing autonomous inspection logic. The system systematically collects equipment meter readings, indicator light states, temperature and humidity data, and partial discharge signal information. The core goal of the solution is to realize intelligent monitoring of the running state of the switch cabinet, early detection of abnormal faults, and maintenance decision-making assisted by data models to ensure long-term stable operation of the power system.

[0004] The existing technology mainly uses fixed cycles and fixed routes in the operation mode, and signals are usually judged item by item according to independent thresholds. There is a lack of unified time index constraint, and there is a lack of synchronization comparison basis between different channels, making it difficult to reflect short-term mutations and cross-channel correlations in a timely manner. The inspection trigger relies on pre-designed plans, which may miss pre-alarm information, resulting in omissions and delays. The path organization is usually based on the established route and cabinet sequence, without considering the spatial distance fluctuation and energy consumption difference in the same evaluation framework, resulting in long return path and uneven energy consumption distribution, affecting the execution efficiency of continuous tasks. Abnormal positioning relies on single-point overrun, which is difficult to complete mutual verification of temperature difference and current in the same time window, causing the false positive rate to rise and the false disposal cost to increase. The repair window arrangement is disconnected from the actual risk points, limiting the system stability and response speed. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings of the prior art and to provide a switch cabinet unmanned inspection method and system.

[0006] To achieve the above object, the application adopts the following technical scheme: A switch cabinet unmanned inspection method, comprising the following steps:

[0007] S1: Obtain the cabinet door state signal, circuit breaker opening and closing position signal, indicator light on-off state, instrument reading change value and cabinet internal temperature and humidity sensor output value through a sensor, sort the cabinet unit state and align the time, analyze the correlation between signal nodes using a graph neural network, and generate a running state hierarchical information set;

[0008] S2: Based on the running state hierarchical information set, filter the high fluctuation section of the disconnecting switch mechanism, calculate the duration of the continuous signal and count the frequency, determine the priority node by amplitude ratio, record the signal segment index that meets the continuous threshold condition, and construct an inspection trigger timing list;

[0009] S3: Based on the inspection trigger timing list, synchronously compare the temperature measurement component and bus contact temperature difference signal and cable end current fluctuation signal, extract the continuous sampling sequence features using a long short-term memory network, confirm the amplitude pairing under the same time index, and obtain the abnormal node positioning sequence;

[0010] S4: Based on the inspection trigger timing list and the abnormal node positioning sequence, match the time index and node position, compare the spatial distance of the corresponding signal group and filter the aggregation, extract the continuous node index and arrange it according to the cabinet number, and form a centralized inspection target list;

[0011] S5: Based on the centralized inspection target list, read the ground loop and disconnecting switch node coordinates, compare the distance between nodes and energy consumption values, complete the path connection and sorting of continuous nodes, and establish an inspection task scheduling structure.

[0012] As a further scheme of the application, the running state hierarchical information set includes the cabinet door state signal layer, circuit breaker opening and closing position signal layer, indicator light on-off state layer, instrument reading change value layer and cabinet internal temperature and humidity signal layer, the inspection trigger timing list includes high fluctuation section index, continuous signal duration statistical value and priority node number, the abnormal node positioning sequence includes temperature measurement component temperature difference signal index, bus contact temperature difference signal index and cable end current fluctuation signal index, the centralized inspection target list includes time index matching result, node spatial distance comparison result and cabinet number arrangement sequence, and the inspection task scheduling structure includes node path connection sequence, node distance value and energy consumption comparison result.

[0013] As a further scheme of the application, the specific steps for generating the running state hierarchical information set are:

[0014] The cabinet state time sequence table is generated by the cabinet door state signal, the circuit breaker opening and closing position signal, the indicator light on and off state, the instrument reading change value and the cabinet temperature and humidity sensor output value, calling the channel and aligning the time, using the sampling time pairing and labeling index, calculating the interval difference and sorting the sequence, and generating the cabinet state time sequence table;

[0015] Based on the cabinet state time sequence table, the adjacent sampling difference is compared, the upper and lower boundaries are marked by threshold interval division, the continuous section is aggregated and the amplitude order is arranged, the start and end time and the channel number are recorded, and the state change interval index is generated.

[0016] Based on the state change interval index, the same time index signal group is checked in parallel, the association relationship between signal nodes is constructed by the graph neural network, the connection weight is calculated, the difference boundary cross matching is executed and the hierarchical number is arranged, the section order is sorted and the index is archived, and the running state hierarchical information set is generated.

[0017] As a further scheme of the application, the graph neural network first inputs the cabinet door state signal, the circuit breaker opening and closing position signal, the indicator light on and off state, the instrument reading change value and the cabinet temperature and humidity sensor output value as nodes, establishes the connection edge between nodes according to the sampling index, determines the connection strength by the sampling time difference and the amplitude difference between nodes, and updates the node feature value according to the connection weight in each round of propagation, forms a multi-layer association mapping structure, accumulates and calculates the relationship matrix of the signal group under different time indexes, and outputs the association weight matrix between nodes.

[0018] As a further scheme of the application, the specific steps for constructing the patrol trigger time sequence list are:

[0019] Based on the running state hierarchical information set, the channels of the isolating switch mechanism signal section are read and compared one by one, the fluctuation range is divided by the cumulative screening of the amplitude difference between adjacent points, the section time index is registered and the number is identified, and the high fluctuation section duration table is generated.

[0020] Based on the high fluctuation section duration table, the continuous signal frequency is counted and the amplitude ratio is calculated, the priority node is determined by sorting the high value group in the section, the signal section meeting the duration and amplitude conditions is screened out by threshold interval determination, the position index and the channel number of the signal section are registered one by one, and the patrol trigger time sequence list is generated.

[0021] As a further scheme of the application, the specific steps for obtaining the abnormal node positioning sequence are:

[0022] Based on the inspection trigger timing list, read the temperature measuring component and bus contact temperature difference signal and cable end current fluctuation signal, use long short-term memory network to extract time sequence characteristics of continuous sampling value, calculate the amplitude difference after comparing the channel sampling value with the same time index, time sequence pairing and labeling number are carried out on the comparison result, and a time index pairing matrix is generated;

[0023] Based on the time index pairing matrix, superimpose the same index signal amplitude, accumulate the amplitude value detection mutation section and record the time position, divide the high amplitude section into intervals and sort them, extract the boundary index of the mutation section and register the identification, and generate an abnormal signal identification table;

[0024] Based on the abnormal signal identification table, match the position of the abnormal section in the spatial node sequence, locate the abnormal node by time index and calculate the distance between adjacent nodes, adjust the number of position data and sort them, register the coordinate relationship of each node, and generate an abnormal node positioning sequence.

[0025] As a further scheme of the application, the long short-term memory network first inputs the temperature measuring component, bus contact temperature difference signal and cable end current fluctuation signal into the network structure according to time index, reads the sampling value of each time step in turn and establishes time dependence, records the amplitude and direction change of the previous time signal through unit state, calculates the difference sequence combining the current sampling value, sequentially superimposes the output of each time step to form a time feature vector, and generates a time sequence feature according to the amplitude change trend in the feature vector.

[0026] As a further scheme of the application, the specific steps for forming the centralized inspection target list are:

[0027] Based on the inspection trigger timing list and the abnormal node positioning sequence, match the time index and synchronize the channel, compare the node signal amplitude under the same index and calculate the spatial distance, perform distance partitioning and difference filtering, record the node number and index below the threshold, and generate a node aggregation association table;

[0028] Based on the node aggregation association table, extract the continuous node index and corresponding cabinet number, perform position regularization through adjacent node coordinate comparison, adjust the order of node index of the same cabinet number and record the index sequence, and generate a centralized inspection target list.

[0029] As a further scheme of the application, the specific steps for establishing the inspection task scheduling structure are:

[0030] Based on the centralized inspection target list, read the grounding loop and isolating switch node coordinates, calculate the distance between adjacent nodes by extracting node coordinates and record the values, collect energy consumption data and synchronize comparison, perform hierarchical and weight labeling on distance and energy values, and generate a path weight distribution table;

[0031] Based on the path weight distribution table, the node paths are combined and the order is adjusted, the connection order is determined by comparing the adjacent node distance and weight level item by item, the connection and numbering registration are performed on the path coordinates, the node order is arranged to form a continuous sequence, and the inspection task scheduling structure is established.

[0032] A switch cabinet unmanned inspection system for executing the above-mentioned switch cabinet unmanned inspection method, the system comprises:

[0033] The signal acquisition module: through the cabinet door state signal, the circuit breaker opening and closing position signal, the indicator light on-off state, the instrument reading change and the temperature and humidity output value, sampling and time alignment are carried out, and the running state hierarchical information set is generated;

[0034] The state hierarchical module: based on the running state hierarchical information set, the high fluctuation section of the disconnecting switch mechanism is screened, the duration and frequency of the continuous signal are calculated, the priority node is determined by the amplitude ratio, and the inspection trigger timing list is generated;

[0035] The abnormality recognition module: based on the inspection trigger timing list, the temperature measuring component, the bus contact temperature difference signal and the cable end current fluctuation signal are compared synchronously, the amplitude matching under the same time index is confirmed, and the abnormal node positioning sequence is generated;

[0036] The target aggregation module: based on the inspection trigger timing list and the abnormal node positioning sequence, the time index is matched and the spatial distance is compared, the aggregation node is screened and arranged according to the cabinet number, and the centralized inspection target list is generated;

[0037] The task scheduling module: based on the centralized inspection target list, the ground loop and the disconnecting switch node coordinates are read, the distance and energy consumption between nodes are compared and sorted, and the inspection task scheduling structure is generated.

[0038] Compared with the prior art, the advantages and positive effects of the present application are:

[0039] 1. In the present application, the correlation between signal nodes is established through the graph neural network, the time sequence consistency of multi-source signals is improved, the instrument reading and temperature and humidity signals can form a hierarchical structure under a unified time framework, the logical association of state layering is strengthened, and the misjudgment rate caused by node isolated analysis is reduced;

[0040] 2. In the present application, the long short-term memory network is used to extract the characteristics of continuous sampling sequence, the response sensitivity and time prediction ability to abnormal signal mutation are improved, the amplitude matching of temperature difference and current signals under the same time index is realized, the early abnormal node is accurately identified, and the dynamic positioning sequence is established;

[0041] 3. In the present application, by jointly weighing the distance and energy consumption, the path length and energy peak are reduced, the task throughput and device availability are improved, the structural integrity of data association is improved at the signal fusion layer, the time sequence capture and trend judgment ability is enhanced at the abnormality identification layer, and the accuracy, continuity and safety evaluation reliability of the inspection decision are improved as a whole. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a workflow diagram of the present application;

[0043] Figure 2 is a system flowchart of the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0045] Example 1

[0046] Please refer to Figure 1 The present application provides a technical scheme: a switch cabinet unmanned inspection method, comprising the following steps:

[0047] S1: Obtain the cabinet door state signal, circuit breaker opening and closing position signal, indicator light on-off state, instrument reading change amount and cabinet temperature and humidity sensor output value through the sensor, sort the cabinet unit state and align the time, analyze the correlation between signal nodes by using graph neural network, and generate running state hierarchical information set;

[0048] S2: Based on the running state hierarchical information set, filter the high fluctuation section of the disconnecting switch mechanism, calculate the duration of the continuous signal and count the frequency, determine the priority node by amplitude ratio, record the signal segment index that meets the continuous threshold condition, and construct the inspection trigger timing list;

[0049] S3: Based on the inspection trigger timing list, synchronously compare the temperature measurement component and bus contact temperature difference signal and cable end current fluctuation signal, extract the continuous sampling sequence features by using long short-term memory network, confirm the amplitude pairing under the same time index, and obtain the abnormal node positioning sequence;

[0050] S4: Based on the inspection trigger timing list and the abnormal node positioning sequence, match the time index and the node position, compare the spatial distance of the corresponding signal group and filter the aggregation, extract the continuous node index and arrange it according to the cabinet number, and form the centralized inspection target list;

[0051] S5: Based on the centralized inspection target list, read the grounding loop and isolating switch node coordinates, compare the path distance and energy consumption value, complete the path connection and ordering of continuous nodes, and establish the inspection task scheduling structure.

[0052] The running state hierarchical information set includes a cabinet door state signal layer, a circuit breaker on-off position signal layer, an indicator light on-off state layer, an instrument reading change amount layer, and a cabinet internal temperature and humidity signal layer. The inspection trigger timing list includes a high fluctuation section index, a continuous signal time length statistical value, and a priority node number. The abnormal node positioning sequence includes a temperature difference signal index of a temperature measurement component, a bus contact temperature difference signal index, and a cable end current fluctuation signal index. The centralized inspection target list includes a time index matching result, a node space distance comparison result, and a cabinet number arrangement sequence. The inspection task scheduling structure includes a node path connection sequence, a node distance value, and an energy consumption comparison result.

[0053] The specific steps for generating the running state hierarchical information set are as follows:

[0054] Through the cabinet door state signal, the circuit breaker on-off position signal, the indicator light on-off state, the instrument reading change amount, and the cabinet internal temperature and humidity sensor output value, the channel is called and aligned in time. The sampling time is paired and indexed. After calculating the interval difference, the sequence is sorted. The cabinet state timing table is generated.

[0055] Based on the cabinet state timing table, the adjacent sampling differences are compared. The threshold interval is divided and the upper and lower boundaries are marked. The continuous sections are aggregated and the amplitude order is arranged. The start and end times and channel numbers are recorded. The state change interval index is generated.

[0056] Based on the state change interval index, the same time index signal group is checked in parallel. The signal node association relationship is constructed and the connection weight is calculated through the graph neural network. The difference boundary cross matching and numbering level are executed. The section order is sorted and the index is archived. The running state hierarchical information set is generated.

[0057] Through the cabinet door state signal, the circuit breaker on-off position signal, the indicator light on-off state, the instrument reading change amount, and the cabinet internal temperature and humidity sensor output value, the five types of input signals are uniformly formatted into a timing matrix according to the time label. The sampling interval is set to 0.5 seconds. The length of each group of sampling sequences is set to 120 data points. The signals are aligned with millisecond timestamps. The synchronization deviation is controlled within 2 milliseconds. After alignment, the channels are sorted from 1 to 5 according to the channel number. The interval difference of the continuous sampling values of each channel is calculated. The difference value is set to 0.001. The difference value sequence length remains consistent with the number of sampling points. After sorting, the channels are arranged in sequence according to the channel number and output in text format. The cabinet state timing table is generated.

[0058] Based on the cabinet status time sequence table, peak detection and boundary marking are performed on the sampling difference sequence of each channel. The upper limit of the difference amplitude is set to ±2.0 and the lower limit is ±0.2. Sampling points with amplitudes exceeding the interval are regarded as fluctuation boundaries. Segments with more than 5 consecutive sampling points are aggregated. The start sampling number and end sampling number of each segment are recorded. The duration is calculated by time span. The frequency of occurrence of each channel segment is counted. Segments with a frequency of more than 3 times and a duration of more than 2 seconds are taken as valid records. The segment numbers are arranged from high to low according to the average amplitude and the channel number information is attached to generate a status change interval index.

[0059] Based on the state change interval index, a graph neural network algorithm is used to construct the association relationship between signal nodes and calculate the connection weight. Each signal node is assigned a fixed number range of 1 to 5, and the node feature dimension is set to 6 dimensions. The input values ​​are door status, circuit breaker position, indicator light status, instrument reading change, temperature and humidity, respectively. The node connection condition is set to establish a connection edge when the time difference is less than 1 second and the amplitude difference is less than 0.5. The connection weight value ranges from 0 to 1, and the initial average value is 0.5. The number of iterations is set to 10 rounds. In each round, the node attribute value is updated according to the connection weight. The final node weights are arranged in descending order, and the node pair numbers and association level information of the top 10 highest weight connection edges are recorded. After sorting, the data is sorted in chronological order and archived into an index to generate a hierarchical information set of operating status.

[0060] The graph neural network first takes the cabinet door status signal, circuit breaker open / close position signal, indicator light on / off status, instrument reading changes, and cabinet temperature and humidity sensor output values ​​as node inputs. It establishes connection edges between nodes based on the sampling index, determines the connection strength by the sampling time difference and amplitude difference between nodes, and updates the node feature values ​​according to the connection weight in each round of propagation to form a multi-layer correlation mapping structure. It performs cumulative calculation and normalization on the relationship matrix of signal groups under different time indices and outputs the correlation weight matrix between nodes.

[0061] Graph neural networks, according to the formula:

[0062]

[0063] in: For the first The node feature matrix of the layer For the first The node feature matrix of the layer For the first The trainable weight matrix of the layer, For degree matrix, For nodes The sum of adjacency weights, As a non-linear activation function, the ReLU function is used in this method. For nodes a sampling time difference between nodes , a time decay scale parameter, a signal amplitude difference between nodes , , an amplitude normalization scale parameter a spatial distance between nodes , , a spatial decay scale parameter, an energy stability standard deviation between nodes , , an energy stability scale parameter, a time difference weight coefficient, an amplitude difference weight coefficient, a spatial distance weight coefficient, an energy stability weight coefficient, a node index;

[0064] The execution process is as follows: the system first takes the cabinet door state, the circuit breaker opening and closing position, the indicator light on and off state, the instrument reading change, and the temperature and humidity signal as input data, aligns and numbers the signals in millisecond level according to the time tag, and defines the signal channel as a node . The system calculates the sampling time difference , the amplitude difference , the spatial distance , and the energy standard deviation between any two nodes , substitutes the above four parameters into the adjacency calculation formula of the graph neural network, forms a composite adjacency weight matrix through the weighted superposition of the time item, the amplitude item, the space item, and the energy item, and fixes the weight proportion of the coefficients , , , of each item after discrete enumeration and error verification, so that each physical quantity participates in weighting under the same dimension. The obtained adjacency weight matrix is subjected to symmetric normalization by the degree matrix , so that the node connection strength remains stable in network propagation. The input feature matrix is updated through weighted propagation and matrix multiplication. The output layer performs nonlinear transformation through the activation function (ReLU) to generate the matrix .

[0065] The specific steps of constructing the inspection triggering time sequence list are as follows:

[0066] Based on the running state hierarchical information set, the channels of the isolating switch mechanism signal section are read item by item, and the continuous sampling difference is compared, the fluctuation range is determined through the cumulative screening of the amplitude difference between adjacent points, the time index of the section is registered and numbered, and the high fluctuation section time length table is generated;

[0067] Based on the high fluctuation section time length table, the continuous signal frequency is counted and the amplitude ratio is calculated, the priority node is determined through the sorting of the high value group in the section, the signal section meeting the duration and amplitude conditions is screened out through the threshold interval judgment, the position index and the channel number of each signal section are registered item by item, and the patrol trigger time sequence list is generated;

[0068] Based on the running state hierarchical information set, the channels of the isolating switch mechanism signal section are read item by item, and the continuous sampling difference is compared, the fluctuation range is determined through the cumulative screening of the amplitude difference between adjacent points, the time index of the section is registered and numbered, and the high fluctuation section time length table is generated;

[0069] Based on the high fluctuation section time length table, the duration and frequency of each fluctuation section are counted, the statistical period is set to 60 seconds, the duration is calculated by multiplying the section length by the sampling interval, the frequency count is based on the number of repeated section numbers, the minimum duration threshold is set to 2 seconds, and the minimum frequency threshold is set to 3 times, the cumulative amplitude total value is obtained by summing the amplitude mean value of each sampling point in the section, and the amplitude ratio is calculated by the proportion of the total amplitude value between sections, the amplitude ratio calculation precision is set to 0.001, the order is sorted from large to small, the top 20% of the sections are taken as priority nodes, the signal section meeting the duration and amplitude conditions is screened out through the threshold interval judgment, and the time index, channel number and amplitude ratio of each signal section are registered item by item, and the registration record is stored in the form of table, and the patrol trigger time sequence list is generated.

[0070] The specific steps of obtaining the abnormal node positioning sequence are:

[0071] Based on the patrol trigger time sequence list, the temperature difference signal between the temperature measuring component and the bus contact point and the cable end current fluctuation signal are read, the long and short term memory network is used to extract the time sequence characteristics of the continuous sampling value, the amplitude difference is calculated after comparing the channel sampling value with the same time index, the comparison result is time sequence paired and numbered, and the time index pairing matrix is generated;

[0072] Based on the time index pairing matrix, the same index signal amplitude is superimposed, the amplitude value of the mutation section is accumulated and the time position is recorded, the high amplitude section is divided into intervals and sorted, the boundary index of the mutation section is extracted and the identification is registered, and the abnormal signal identification table is generated;

[0073] Based on the abnormal signal identification table, the position of the abnormal section in the spatial node sequence is matched, the abnormal node is positioned by time index and the distance between adjacent nodes is calculated, the number of position data is adjusted and the order is sorted, the coordinates of each node are registered, and the abnormal node positioning sequence is generated;

[0074] Based on the inspection trigger timing list, the long short-term memory network algorithm is used to analyze the continuous sequence of temperature measurement component sampling signals, bus contact temperature difference signals and cable end current fluctuation signals, the input sequence length is set to 60 groups of data, each group contains three types of signal value items, the signal sampling interval is 0.5 seconds, the number of input layer neurons is set to 3, corresponding to three types of signal channels, the number of hidden layer units is set to 128, the activation function of the cycle layer is specified as tanh, the number of output layer units is set to 1, the output dimension corresponds to the signal amplitude change value, the gradient clipping threshold is set to 1.0 to limit the parameter update range, the learning rate is set to 0.001, and the iteration number is 100 times. The time step input mode is used to sequentially read the input signal of each sampling and perform state transfer, calculate the amplitude difference between the current input and the previous state at each time step, and record the calculation result as the amplitude change value sequence. After calculation, the three types of signals are compared with the same time index, and the comparison rule is that the temperature difference signal and the current signal are paired under the condition that the time error is less than 1 second. Calculate the amplitude difference value of each paired data and keep three decimal places in absolute value form. The pairing results are summarized to form a time sequence matrix, and a sequential number is added to each index to generate a time index pairing matrix;

[0075] Based on the time index pairing matrix, the same index signal amplitude is superimposed, the sampling interval is set to 0.5 seconds, the signal superposition step is 10 sampling points, the amplitude value interval is 0 to 100, and the detection threshold is set to greater than 20 when the cumulative amplitude change is greater than 20. Marked as a mutation section, perform amplitude accumulation for each sampling point, record the time index and register the time position number at the threshold, divide the high amplitude section into intervals, the interval division step is 5 sampling points, the sorting basis is the cumulative amplitude from large to small, extract the start sampling number and end sampling number of each mutation section, format the number and corresponding time label into integer indexes, and store them in a table structure according to the mutation section order to generate an abnormal signal identification table;

[0076] Based on the abnormal signal identification table, the position of the abnormal section in the spatial node sequence is matched, the preset cabinet node number range is 1 to 20, each node corresponds to a spatial coordinate triplet (X, Y, Z), the unit is millimeter, the time index recorded in the abnormal section is compared with the node time index table, the error tolerance is set to ±1 second, when the matching condition is met, the corresponding node number is located, the coordinate difference value between adjacent nodes is calculated according to the Euclidean distance formula, the result is rounded to two decimal places, the node position number is renumbered, and the numbering order is arranged according to the time index from small to large. The position data is sorted and uniformly formatted into a CSV structure table, the coordinate relationship of each node and the corresponding abnormal number are registered, and the abnormal node positioning sequence is generated.

[0077] The long short-term memory network first inputs the temperature measuring component, the bus contact temperature difference signal and the cable end current fluctuation signal into the network structure according to the time index, reads the sampling values of each time step in turn and establishes the time dependence, records the amplitude and direction change of the signal at the previous moment through the unit state, calculates the difference sequence combined with the current sampling value, sequentially superimposes the outputs of each time step to form a time feature vector, and generates a time sequence feature according to the amplitude change trend in the feature vector;

[0078] The long short-term memory network is according to the formula:

[0079]

[0080] Among them: is the time index is the hidden state vector at the time index is the input vector at the time index , containing the normalized sampling values of temperature difference and current, temperature and humidity, is the hidden state vector of the previous time index, is the unit state vector of the previous time index, is the temperature difference signal amplitude difference value at the time index , unit: Celsius, is the current signal amplitude difference value at the time index , unit: Ampere, is the trigger alignment residual at the time index , unit: second, is the energy consumption standard deviation at the time index , unit: Joule, is the spatial neighborhood factor at the time index , normalized to 0 to 1 by node distance, is the gate activation function taking Sigmoid, is the hyperbolic tangent function, is the element-wise multiplication, forgetting decay coefficient scalar is 0.9, is a weight matrix for input to the gate, is a weight matrix for hidden state to the gate, is a weight matrix for external coupling to the gate, is a bias vector, is a matrix is the th scalar weight in is a matrix is the th scalar weight in is a matrix is the th scalar weight in is a vector is the th bias component in is a vector is the th component of is a vector is the th component of is a vector is the th component of is a scalar of time index;

[0081] Execution process: after the inspection data stream enters the timing processing link, the cabinet door state, circuit breaker opening and closing position, indicator light on and off, instrument reading change and temperature and humidity sequence are normalized and input vectors are formed , the instantaneous amplitude difference of temperature difference signal and current signal with the same index is calculated according to the inspection trigger timestamp to obtain and , the difference between trigger time and sampling time is obtained , the standard deviation of energy sequence is obtained by 60 second sliding window , the adjacent distance is calculated by the three-dimensional coordinates of the cabinet nodes and linearly mapped to the interval 0 to 1 , the and and and and and are linearly synthesized according to the formula and weight matrix and and and bias , and the gating output is obtained by activation, the previous moment unit state and decay coefficient and spatial factor are linearly synthesized according to the formula The nonlinear transformation is performed on the lower part and is multiplied with the gating output to obtain the hidden state output

[0082] The specific steps for forming the centralized inspection target list are as follows:

[0083] Based on the inspection trigger timing list and the abnormal node positioning sequence, the time index is matched and the channel is synchronized, the node signal amplitudes at the same index are compared and the spatial distance is calculated, the distance is partitioned and the difference is screened, the node numbers and indexes below the threshold are recorded, and the node aggregation association table is generated;

[0084] Based on the node aggregation association table, the continuous node indexes are extracted and the cabinet number is corresponded, the position is regularized by comparing the coordinates of adjacent nodes, the order of the node indexes of the same cabinet number is adjusted and the index sequence is recorded, and the centralized inspection target list is generated;

[0085] Based on the inspection trigger timing list and the abnormal node positioning sequence, the time indexes of the two groups of data are compared and the channel is synchronized, the time index format is set as an integer second timestamp, the time error tolerance is set as ±1 second, when the time index difference is less than or equal to the tolerance, it is determined as a matching item, the node signal amplitudes after matching are compared, the amplitude unit is set as ampere and Celsius, the amplitude value range is 0 to 120, the sampling accuracy is set as 0.01, the node coordinates of each matching item are read and the spatial distance is calculated, the spatial coordinate unit is set as millimeter, the distance threshold is set as 500 millimeters, the calculation result is kept to two decimal places, and the distance value is partitioned according to the distance value, the partition standard is 0 to 100 millimeters, 100 to 300 millimeters, and 300 to 500 millimeters, the node difference of each interval is screened, the screening condition is that the amplitude difference of adjacent nodes is less than 5 and the spatial distance is lower than the threshold 500 millimeters, the aggregated node is recorded when the condition is met, the recorded items are sorted in ascending order of time index and saved as a table file, and the node aggregation association table is generated;

[0086] Based on the node aggregation association table, the continuity of the aggregated node indexes is identified, the continuous node determination condition is set as the adjacent index difference being not greater than 1, the nodes meeting the condition are grouped, the cabinet number range is set as 1 to 30, the nodes are matched according to the cabinet number field, the coordinate comparison is performed on the nodes under the same cabinet number, the comparison rule is that when the X coordinate difference between adjacent nodes is less than 50 millimeters, the Y coordinate difference is less than 30 millimeters, and the Z coordinate difference is less than 30 millimeters, it is determined as the node of the same cabinet, the position of the node of the same cabinet is regularized, the order of the adjusted node sequence is renumbered, the number format is “cabinet number+node serial number”, the node serial number is incremented from 001, the node number, spatial coordinate and time index are written into the table, the node index sequence in each cabinet is recorded, the record structure with consistent numbering sequence is arranged, and the centralized inspection target list is generated.

[0087] The specific steps for establishing the patrol task scheduling structure are:

[0088] Based on the centralized patrol target list, read the coordinates of the grounding loop and isolating switch nodes, calculate the distance between adjacent nodes by extracting the node coordinates and record the values, collect energy consumption data and synchronize comparison, classify and weight the distance and energy values, and generate a path weight distribution table;

[0089] Based on the path weight distribution table, combine the node paths and adjust the order, compare the distance and weight level of adjacent nodes item by item to determine the connection order, perform connectivity and numbering registration on the path coordinates, arrange the node order to form a continuous sequence, and establish the patrol task scheduling structure;

[0090] Based on the centralized patrol target list, read the coordinates of the grounding loop and isolating switch nodes, set the node coordinate format to (X, Y, Z) with a unit of millimeters, the coordinate value range is 0 to 6000, the number of nodes is set to 30, extract the coordinates of every two nodes in order and calculate the Euclidean distance between nodes, the distance result is rounded to three decimal places, and recorded in the node distance table. At the same time, collect the energy consumption data when the nodes are running, the sampling frequency is set to 1 Hz, the sampling period is 120 seconds, the energy unit is joule, align and compare the sampling data by timestamp, the deviation tolerance is set to ±1 second, calculate the distance value of each node pair and normalize it with the average energy consumption value, the normalization interval is 0 to 1, assign weight coefficients 0.7 and 0.3 to distance data and energy data respectively, get the path comprehensive weight value by weighted superposition, the weight precision is set to three decimal places, arrange the paths in ascending order of weight value and record the path number, node pair number and weight level, store all recorded data in a structured table file in order, and generate a path weight distribution table;

[0091] Based on the path weight distribution table, combine and adjust the path connection order, set the initial node as No. 1 and the terminal node as No. 30, call the weight value and distance value of each path in the path weight distribution table for cross comparison, set the weight difference threshold to 0.05 and the distance difference threshold to 100 millimeters, when the weight difference and distance difference of two paths are less than the threshold at the same time, merge them into the same continuous segment, rearrange the node numbers in the continuous segment in timestamp order, connect the sorted nodes, check the continuity of the nodes according to the continuity of the node coordinates during connection, if the coordinate deviation exceeds 50 millimeters, insert an intermediate node to supplement the record, renumber the path numbers after connection is completed, the numbering rule is three-digit integer continuous numbering, starting from 001 and accumulating, register and arrange the node sequence, weight level and distance total of each path, form the path scheduling record arranged in execution order, and generate the patrol task scheduling structure.

[0092] Please refer to Figure 2The application discloses an unmanned inspection system for a switch cabinet.

[0093] The signal acquisition module samples and time-aligns the cabinet door state signal, the circuit breaker opening and closing position signal, the indicator light on-off state, the instrument reading change and the temperature and humidity output value, and generates the running state hierarchical information set.

[0094] The state hierarchical module filters the high fluctuation section of the isolating switch mechanism based on the running state hierarchical information set, calculates the duration and frequency of the continuous signal, determines the priority node through the amplitude ratio, and generates the inspection trigger timing list.

[0095] The abnormality identification module synchronously compares the temperature measurement component, the bus contact temperature difference signal and the cable end current fluctuation signal based on the inspection trigger timing list, confirms the amplitude pairing under the same time index, and generates the abnormal node positioning sequence.

[0096] The target aggregation module matches the time index and compares the spatial distance based on the inspection trigger timing list and the abnormal node positioning sequence, filters the aggregation node, arranges the cabinet number, and generates the centralized inspection target list.

[0097] The task scheduling module reads the ground loop and the isolating switch node coordinates based on the centralized inspection target list, compares and sorts the distance and energy consumption between nodes, and generates the inspection task scheduling structure.

[0098] The above is only a preferred embodiment of the application, and does not limit the application in other forms. Any skilled person in the art can modify or change the above disclosed technical content to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made on the basis of the technical essence of the application to the above embodiments still belong to the protection scope of the application technical scheme.

Claims

1. A switch cabinet unmanned inspection method, characterized in that, The method comprises the following steps: S1: obtaining the cabinet door state signal, the circuit breaker opening and closing position signal, the indicator light on and off state, the instrument reading change value and the cabinet temperature and humidity sensor output value through the sensor, sorting the cabinet unit state and aligning the time, using the graph neural network to analyze the correlation between the signal nodes, and generating the running state hierarchical information set; The running state hierarchical information set comprises a cabinet door state signal layer, a circuit breaker opening and closing position signal layer, an indicator light on and off state layer, an instrument reading change value layer and a cabinet temperature and humidity signal layer; The specific steps for generating the running state hierarchical information set are as follows: Through the cabinet door state signal, the circuit breaker opening and closing position signal, the indicator light on and off state, the instrument reading change value and the cabinet temperature and humidity sensor output value, the channel is called and the time is aligned, the sampling time is matched and the index is labeled, the interval difference value is calculated, the sequence is sorted, and the cabinet state time sequence table is generated; Based on the cabinet state time sequence table, the adjacent sampling difference values are compared, the threshold interval is divided and the upper and lower boundaries are marked, the continuous sections are aggregated and the amplitude order is arranged, the start and end times and the channel numbers are recorded, and the state change interval index is generated; Based on the state change interval index, the same time index signal groups are checked in parallel, the correlation between the signal nodes is constructed through the graph neural network, the connection weight is calculated, the difference boundary cross matching is performed and the level is numbered, the section order is arranged and the index is archived, and the running state hierarchical information set is generated; S2: Based on the running state hierarchical information set, the high fluctuation section of the disconnecting switch mechanism is screened, the duration of the continuous signal is calculated and the frequency is counted, the priority node is determined through the amplitude ratio, the signal segment index meeting the continuous threshold condition is recorded, and the inspection trigger time sequence list is constructed; The specific steps for constructing the inspection trigger time sequence list are as follows: Based on the running state hierarchical information set, the channel of the isolating switch mechanism signal section is continuously sampled and compared, the sampling frequency is set to 2 Hz , the total sampling points are 240, the amplitude change value of each channel is calculated by absolute difference and kept to three decimal places, the amplitude difference cumulative threshold is set to 1.5, the window width is 5 sampling points, the amplitude difference between each adjacent sampling point is accumulated by cycle, if the cumulative value exceeds the threshold, it is determined as a fluctuation section, after the detection is completed, the starting sampling index and the ending sampling index of the section are recorded, the time index unit is set to seconds, the section time range is calculated by multiplying the starting sampling time and the sampling interval, the detection result is numbered and identified, the numbering rule is composed of two integer numbers according to the channel number and the detection order, the section number is sequentially increased from 001, the result is arranged in time sequence and registered in the record table, and a high fluctuation section duration table is generated; The amplitude ratio is obtained by summing the amplitude mean value of each sampling point in the section to obtain the cumulative amplitude total value, and the amplitude ratio is calculated by the proportion of the amplitude total value between sections; S3: Based on the inspection trigger time sequence list, the temperature difference signal of the temperature measurement component and the bus contact and the cable end current fluctuation signal are compared synchronously, the long short-term memory network is used to extract the continuous sampling sequence features, the amplitude pairing under the same time index is confirmed, the abnormal node positioning sequence is obtained, and the amplitude pairing is the amplitude difference calculated by comparing the channel sampling values under the same time index; S4: Based on the inspection trigger time sequence list and the abnormal node positioning sequence, the time index matching and the node position are associated, the spatial distance of the corresponding signal group is compared and screened, the continuous node index is extracted and arranged according to the cabinet number, and the centralized inspection target list is formed; S5: Based on the centralized inspection target list, the ground loop and the disconnecting switch node coordinates are read, the distance and energy consumption value between nodes are compared and the order is adjusted, the path connection and sorting of the continuous nodes are completed, the inspection task scheduling structure is established, and the energy consumption value is the energy consumption data when the node is running.

2. The switch cabinet unmanned inspection method according to claim 1, characterized in that, The inspection trigger timing list comprises a high fluctuation section index, a continuous signal duration statistical value and a priority node number, the abnormal node positioning sequence comprises a temperature measuring assembly temperature difference signal index, a bus contact temperature difference signal index and a cable end current fluctuation signal index, the centralized inspection target list comprises a time index matching result, a node space distance comparison result and a cabinet number arrangement sequence, and the inspection task scheduling structure comprises a node path connection sequence, a node distance value and an energy consumption comparison result.

3. The switch cabinet unmanned inspection method according to claim 1, characterized in that, The graph neural network first takes a cabinet door state signal, a circuit breaker opening and closing position signal, an indicator light on-off state, an instrument reading change amount and a cabinet temperature and humidity sensor output value as node input, establishes a connection edge between nodes according to a sampling index, determines a connection strength through a sampling time difference and an amplitude difference between nodes, and updates a node feature value according to a connection weight in each round of propagation to form a multi-layer correlation mapping structure, accumulatively calculates and normalizes a relationship matrix of a signal group under different time indexes, and outputs a correlation weight matrix between nodes.

4. The switch cabinet unmanned inspection method according to claim 1, characterized in that, The specific steps for constructing the inspection trigger timing list further comprise: Based on the high fluctuation section duration table, the frequency of continuous signals is counted and the amplitude ratio is calculated, the priority node is determined through sorting of high value groups in the section, the signal segments meeting the duration and amplitude conditions are screened out through threshold interval determination, the position index and the channel number of the signal segment are registered item by item, and the inspection trigger timing list is generated.

5. The switch cabinet unmanned inspection method according to claim 1, characterized in that, The specific steps for obtaining the abnormal node positioning sequence are: Based on the inspection trigger timing list, the temperature measuring assembly, the bus contact temperature difference signal and the cable end current fluctuation signal are read, the long short-term memory network is used to extract the time sequence features of the continuous sampling values, the amplitude difference is calculated after comparing the channel sampling values of the same time index, the comparison results are paired and numbered in time sequence, and a time index pairing matrix is generated; Based on the time index pairing matrix, the amplitudes of the signals of the same index are superimposed, the amplitude value of the mutation section is accumulated and the time position is recorded, the high amplitude section is divided into intervals and sorted, the boundary index of the mutation section is extracted and the identifier is registered, and an abnormal signal identifier table is generated; Based on the abnormal signal identifier table, the position of the abnormal section in the space node sequence is matched, the abnormal node is positioned through the time index and the distance between adjacent nodes is calculated, the number of the position data is adjusted and the order is sorted, the coordinate relationship of each node is registered, and an abnormal node positioning sequence is generated.

6. The switchgear unmanned inspection method according to claim 1, characterized in that, The long short-term memory network first inputs the temperature measuring assembly, the bus contact temperature difference signal and the cable end current fluctuation signal into the network structure according to the time index, reads the sampling values of each time step in turn and establishes a time dependence relationship, records the amplitude and direction change of the signal at the previous time step through the unit state, calculates the difference sequence combining the current sampling value, sequentially superimposes the outputs of each time step to form a time feature vector, and generates a time sequence feature according to the amplitude change trend in the feature vector.

7. The switchgear unmanned inspection method according to claim 1, characterized in that, The specific steps for forming the centralized inspection target list are: Based on the patrol trigger timing list and the abnormal node positioning sequence, time indexes are matched and channels are synchronized, distance partitioning and difference screening are performed by comparing node signal amplitudes at the same index and calculating spatial distances, node numbers and indexes below a threshold are recorded, and a node aggregation association table is generated; Based on the node aggregation association table, continuous node indexes are extracted and corresponding cabinet numbers are recorded, position normalization is performed by comparing adjacent node coordinates, the order of node indexes of the same cabinet number is adjusted and the index sequence is recorded, and a centralized patrol target list is generated.

8. The switchgear unmanned inspection method according to claim 1, characterized in that, The specific steps for establishing the patrol task scheduling structure are: Based on the centralized patrol target list, ground return circuit and isolating switch node coordinates are read, adjacent node distances are calculated by extracting node coordinates and the values are recorded, energy consumption data are collected and synchronized comparison is performed, distance and energy values are classified and weighted, and a path weight distribution table is generated; Based on the path weight distribution table, node paths are combined and the order is adjusted, the connection order is determined by comparing adjacent node distances and weight levels item by item, path coordinates are connected and numbered, node order is arranged to form a continuous sequence, and a patrol task scheduling structure is established.

9. A switch cabinet unmanned inspection system, characterized in that, The switch cabinet unmanned patrol method according to any one of claims 1-8, the system comprising: A signal acquisition module: cabinet door state signals, circuit breaker opening and closing position signals, indicator light on and off states, instrument reading change amounts, and indoor temperature and humidity output values are sampled and time-aligned to generate a running state hierarchical information set; The running state hierarchical information set includes a cabinet door state signal layer, a circuit breaker opening and closing position signal layer, an indicator light on and off state layer, an instrument reading change amount layer, and an indoor temperature and humidity signal layer; The specific steps for generating the running state hierarchical information set are: Cabinet door state signals, circuit breaker opening and closing position signals, indicator light on and off states, instrument reading change amounts, and indoor temperature and humidity sensor output values are used to call channels and align time, sample time points are paired and indexed, interval differences are calculated and sorted sequences are generated, and a cabinet state timing table is generated; Based on the cabinet state timing table, adjacent sampling differences are compared, threshold intervals are divided and upper and lower boundaries are marked, continuous sections are aggregated and amplitude order is arranged, start and end times and channel numbers are recorded, and a state change interval index is generated; Based on the state change interval index, the same time index signal groups are checked in parallel, signal node association relationships are constructed and connection weights are calculated through a graph neural network, difference boundary intersection matching is performed and hierarchical numbering is executed, section order is arranged and the index is archived, and a running state hierarchical information set is generated. A state hierarchical module: based on the running state hierarchical information set, the isolating switch mechanism high fluctuation section is screened, the duration and frequency of continuous signals are calculated, the priority node is determined by the amplitude ratio, and a patrol trigger timing list is generated. The specific steps for constructing the patrol trigger timing list are: Based on the running state hierarchical information set, the channel of the isolating switch mechanism signal section is continuously sampled and compared, the sampling frequency is set to 2 Hz , the total sampling points are 240, the amplitude change value of each channel is calculated by absolute difference and kept to three decimal places, the amplitude difference cumulative threshold is set to 1.5, the window width is 5 sampling points, the amplitude difference between each adjacent sampling point is accumulated by cycle, if the cumulative value exceeds the threshold, it is determined as a fluctuation section, after the detection is completed, the starting sampling index and the ending sampling index of the section are recorded, the time index unit is set to seconds, the section time range is calculated by multiplying the starting sampling time and the sampling interval, the detection result is numbered and identified, the numbering rule is composed of two integer numbers according to the channel number and the detection order, the section number is sequentially increased from 001, the result is arranged in time sequence and registered in the record table, and a high fluctuation section duration table is generated; The amplitude ratio is obtained by summing the amplitude mean values of each sampling point in the section to obtain the cumulative amplitude total value, and the amplitude ratio is calculated based on the proportion of the amplitude total value between sections. Anomaly identification module: based on the patrol trigger timing list, synchronously compare the temperature measurement component, bus contact temperature difference signal and cable end current fluctuation signal, confirm the amplitude pairing under the same time index, generate the abnormal node positioning sequence, and the amplitude pairing is the amplitude difference calculated after comparing the channel sampling value through the same time index; Target aggregation module: based on the patrol trigger timing list and the abnormal node positioning sequence, match the time index and compare the spatial distance, filter the aggregation nodes and arrange them according to the cabinet number, and generate the centralized patrol target list; Task scheduling module: based on the centralized patrol target list, read the grounding loop and isolating switch node coordinates, compare and sort the distance between nodes and energy consumption, generate the patrol task scheduling structure, and the energy consumption is the energy consumption data when the node runs.

Images