A method and system for monitoring the operating status of cables based on deep learning

Abstract

The present invention discloses a method and system for monitoring the operating state of a cable based on deep learning, comprising the following steps: collecting cable state data obtained in real time by a sensing device; monitoring and performing a preliminary scoring and feedback on the operating state of the cable based on a preset operating range of the cable state data; deeply extracting the cable state data based on feature extraction by HOG to obtain cable state feature data according to a feature screening strategy; and building a cable operating state monitoring model based on a BP neural network optimized by an improved genetic algorithm to perform deep monitoring and feedback on the obtained cable state feature data. The present invention monitors the operating state of the cable based on the cable operating state monitoring model, makes adaptive adjustments according to the current algorithm, and improves the monitoring effect of the BP neural network; and also feeds back the deep monitoring results through various warning methods, which can ensure that the operating state of the cable is received and monitored to the greatest extent, and can quickly handle the abnormal state of the cable.

Description

Technical Field

[0001] The present invention relates to the technical field of cables, and in particular, to a method and system for monitoring the operating state of cables based on deep learning. Background Art

[0002] Power cables are used to transmit and distribute electrical energy. They are commonly used in urban underground power grids, outgoing lines of power stations, internal power supply of industrial and mining enterprises, and underwater transmission lines across rivers and seas. In power lines, the proportion of cables is gradually increasing. Power cables are cable products used to transmit and distribute high-power electrical energy in the main lines of the power system. After the existing power cable operating state monitoring and control system detects an abnormal operation, it usually first adopts a maintenance method to handle the abnormality, and then replaces the entire cable after the maintenance fails. Therefore, once the maintenance fails, the overall efficiency of cable abnormality handling will be low.

[0003] In the prior art, deep learning algorithms, such as BP neural networks, are often introduced in equipment state monitoring to learn and monitor the operating state of cables. The BP neural network can make the state expression better reflect the true state of cable equipment and achieve daily equipment monitoring of cable equipment. However, the learning speed of the BP neural network is slow and it is prone to falling into local extrema, resulting in training failure; it is not conducive to real-time monitoring of the operating state of cables, and the equipment monitoring alarm in the prior art only uses a single alarm method, which cannot ensure that the operating state of cable equipment is received to the greatest extent and achieve the safety guarantee of cable equipment operation. Summary of the Invention

[0004] To solve the technical problems in the background art, the present invention proposes a method and system for monitoring the operating state of cables based on deep learning.

[0005] A method for monitoring the operating state of cables based on deep learning proposed by the present invention includes the following steps:

[0006] S1: Collect the cable state data obtained in real time by the sensing device, and perform filtering processing on the cable state data based on an improved wavelet threshold algorithm;

[0007] S2: Monitor the filtered cable state data, and perform a preliminary score and feedback on the operating state of the cable based on the preset operating range of the cable state data. When the monitored cable state data exceeds the cable state deep monitoring score threshold, go to S3; otherwise, return to S1;

[0008] S3: Based on a feature screening strategy, perform deep extraction on the cable state data extracted based on HOG to obtain cable state feature data;

[0009] S4: Build a cable operation status monitoring model based on an improved genetic algorithm-optimized BP neural network to perform in-depth monitoring and feedback on the obtained cable status characteristic data.

[0010] Preferably, the cable status data in S1 includes cable operation electrical quantity data and cable operation physical quantity data.

[0011] Preferably, the data filtering process in S1 is specifically as follows:

[0012]

[0013]

[0014] Among them, is the th estimated wavelet coefficient of the th layer; is the th true wavelet coefficient of the th layer of the decomposition; is the th selected threshold; is a mathematical constant; is a step function; is a median function; is a high-order factor.

[0015] Preferably, in S2, the cable data is classified into score levels according to the operation range of the cable status data, and the status feedback of the cable data is performed according to the classification result of the current cable data score level.

[0016] Preferably, the score levels include abnormal, general, good, and excellent.

[0017] Preferably, S3 performs feature extraction and feature screening on the cable operation statuses rated as abnormal, general, and good according to the status feedback result in S2.

[0018] Preferably, the feature screening strategy in S3 is specifically as follows:

[0019] Obtain pattern class sets:

[0020]

[0021] Among them, is the th HOG feature of the th cable status data under the th pattern class; is the total number of cable status data under the The number of HOG features for each cable status data;

[0022] There are cable status data under each mode type, and a total of cable status data, then there are HOG features, and the feature set is ;

[0023] Calculate the average distance of samples under the same conditions :

[0024]

[0025] Among them, is the th HOG feature of the th cable status data under the th mode class; is the th HOG feature of the th cable status data under the th mode class;

[0026] Average to obtain the average within-class distance :

[0027]

[0028] The calculated variance factor :

[0029]

[0030] Calculate the average distance between cable status data under different conditions :

[0031]

[0032] Calculate the average between-class distance between cable status data under different conditions :

[0033]

[0034] Among them, is the average distance between cable status data under the th mode;

[0035] Calculate 's variance factor :

[0036]

[0037] Calculate the compensation factor is:

[0038]

[0039] Calculate and normalize:

[0040]

[0041] Among them, is the th cable status feature data obtained by feature processing.

[0042] Preferably, the cable operation status monitoring model described in S4 is specifically built as follows:

[0043] The input layer of the BP neural network contains neurons, the hidden layer contains neurons, and the output layer contains neurons.

[0044] The output of the hidden layer is:

[0045]

[0046] The output of the output layer is:

[0047]

[0048] Among them, is the weight between the th neuron in the input layer and the th neuron in the hidden layer, is the output signal of the th neuron in the input layer, is the weight between the th neuron in the hidden layer and the th neuron in the output layer, is the activation function;

[0049] Optimize the initial weights and output layer thresholds in the BP neural network based on the improved genetic algorithm:

[0050] The population is composed of the combination of chromosomes. Each of the chromosomes contains gene sequences. Iteration is performed through crossover and mutation of the gene sequences. Set the scoring error function as the fitness function value. The greater the fitness value, the greater the probability of being selected;

[0051] The population is optimized based on the following selection probability, crossover probability, and mutation probability: ​

[0052] Selection probability of chromosome is as follows;

[0053]

[0054] wherein, is the reciprocal of the fitness value of the th chromosome;

[0055] The crossover probability of genes on the chromosome is:

[0056]

[0057] wherein, represents the crossover probability of genes, and respectively represent the minimum and maximum crossover probabilities of genes, represents the fitness of genes, represents the minimum fitness of genes, represents the average fitness of genes, is a mathematical constant;

[0058] The mutation probability of genes on the

[0059]

[0060] chromosome is: represents the mutation probability of genes; and respectively represent the minimum and maximum mutation probabilities of genes.

[0061] Preferably, the depth monitoring feedback in S4 is performed based on information alarm, email alarm, phone alarm and network connection alarm methods.

[0062] A cable operation status monitoring system based on deep learning proposed by the present invention includes:

[0063] Data acquisition module: used to acquire cable status data obtained in real time in the sensing device;

[0064] Status monitoring module: used to perform preliminary monitoring, scoring and feedback on cable status data;

[0065] Feature extraction module: used to deeply extract cable status feature data from cable status data based on HOG for feature extraction based on a feature screening strategy;

[0066] Deep monitoring module: It is used to build a cable operation status monitoring model based on a BP neural network optimized by an improved genetic algorithm, and perform deep monitoring feedback on the obtained cable status characteristic data;

[0067] The data acquisition module further includes a data filtering unit, and the data filtering unit is used to perform filtering processing on the cable status data based on an improved wavelet threshold algorithm;

[0068] The status monitoring module includes a status scoring unit and a status feedback unit,

[0069] The status scoring unit performs a preliminary score on the cable operation status based on a preset cable status data operation range;

[0070] The status feedback unit gives feedback based on the scoring result of the status scoring unit.

[0071] A cable operation status monitoring method and system based on deep learning proposed by the present invention have the following beneficial effects:

[0072] The present invention performs real-time monitoring of the cable operation status based on a BP neural network improved by an improved genetic algorithm, and makes adaptive adjustments according to the current algorithm to improve the monitoring effect of the BP neural network; it also gives feedback on the deep monitoring results through various alarm methods, which can ensure that the cable operation status is received and monitored to the greatest extent, and can quickly handle abnormal cable statuses. Description of the Drawings

[0073] Figure 1 It is a flowchart of a cable operation status monitoring method based on deep learning proposed by the present invention;

[0074] Figure 2 It is a monitoring effect diagram of a cable operation status monitoring method based on deep learning proposed by the present invention;

[0075] Figure 3 It is a block diagram of a cable operation status monitoring system based on deep learning proposed by the present invention.

[0076] The meanings of the marks in the figure: 100, data acquisition module; 110, data filtering unit; 200, status monitoring module; 210, status scoring unit; 220, status feedback unit; 300, scoring feedback module; 400, deep monitoring module. Detailed Embodiments

[0077] Referring to Figure 1 , this embodiment provides a cable operation status monitoring method based on deep learning, including:

[0078] S1: Collect the cable status data obtained in real time within the sensing device, and perform filtering processing on the cable status data based on the improved wavelet threshold algorithm. The specific steps are as follows:

[0079] Collect two types of cable status data based on the sensor device, namely the electrical quantity data during cable operation (such as current, voltage, and partial discharge amount, etc.) and the physical quantity data during cable operation (such as cable operating temperature, ambient temperature, ambient humidity, etc.).

[0080] Process the cable status data based on operations such as data cleaning, data filtering, missing value filling, and data feature extraction;

[0081] Data cleaning processing includes removing duplicate data and normalizing data with different dimensions, etc., which can effectively improve the data processing speed.

[0082] The specific data filtering processing is as follows:

[0083]

[0084]

[0085] Among them, is the th estimated wavelet coefficient of the th layer; is the th real wavelet coefficient of the th layer of the decomposition; is the th selected threshold; is a mathematical constant; is a step function; is a median function; is a high-order factor;

[0086] Traditional data filtering processing is based on hard threshold for filtering to remove noise coefficients. However, there are situations such as data jumps and over-smoothing in traditional denoising, which reduces the data representation ability. Therefore, the above multi-layer threshold method is adopted for threshold estimation to improve the filtering function and enhance the filtering ability of the function.

[0087] S2: Monitor the cable status data after filtering processing, and perform preliminary scoring and feedback on the cable operation status based on the preset cable status data operation range. When the monitored cable status data exceeds the cable status depth monitoring scoring threshold, transfer to S3; otherwise, return to S1. The specific steps are as follows:

[0088] Score and grade the cable data according to the operating range. The score grades of abnormal, general, good, and excellent correspond to less than 60 points, 60 - 80 points, 80 - 90 points, and more than 90 points respectively. The upper and lower floating ranges of the rated cable status data can be divided according to different types of cable status data for scoring. For example, with a floating range of is considered excellent, is considered good, is considered general, and exceeding is considered abnormal; and conduct status feedback on the cable data according to the scoring grade division result of the current cable data. For example, if the change range of the current cable status data is within and the score is above 90 points, it is determined that the cable operating status at this time is excellent, and there is no need to further monitor the cable operating status, which can save computing resources and improve the monitoring efficiency of the cable operating status; if the change range of the current cable status data is within and the score is between 80 - 90 points, it is determined that the cable operating status at this time is good, and further monitor the cable operating status and provide real-time feedback; if the change range of the current cable operating status data is within to and the score is below 80 points, further monitoring is also carried out; in the actual process of monitoring the cable operation, through real-time data collection and monitoring, it is possible to further monitor and feedback the cable operation status within the good state range of the change range , which can basically ensure that the cable operates in the best working state and improve the operation efficiency of the cable.

[0089] The cable status data performs preliminary monitoring of the cable status based on the preset operating range of the cable status data, and the percentage change between the cable status data and the rated data can be calculated.

[0090] S3: Based on the feature screening strategy, deeply extract the cable status data obtained by feature extraction based on HOG to obtain cable status feature data; the specific steps are as follows:

[0091] Obtain pattern class sets:

[0092]

[0093] Among them, is the th HOG feature of the th cable status data under the th pattern class; is the total number of cable status data under the th class of patterns;

[0094] There are cable status data under each type of mode, and there are a total of cable status data, then there are HOG features, and the feature set is ;

[0095] Calculate the average distance of samples under the same conditions :

[0096]

[0097] Among them, is the th HOG feature of the th cable status data under the th mode class; is the th HOG feature of the th cable status data under the th mode class;

[0098] Average to obtain the average within-class distance :

[0099]

[0100] Calculate the variance factor :

[0101]

[0102] Calculate the average distance between cable status data under different conditions :

[0103]

[0104] Calculate the average between-class distance between cable status data under different conditions :

[0105]

[0106] Among them, is the average distance between cable status data under the th mode;

[0107] Calculate 's variance factor :

[0108]

[0109] Calculate the compensation factor which is:

[0110]

[0111] Calculate and normalize:

[0112]

[0113] where is the th cable status feature data obtained by feature processing.

[0114] The feature extraction algorithm is based on a feature screening strategy. By calculating the inter-class distance between feature classes and the intra-class distance of features, a feature score is obtained according to their ratio, so as to perform data dimensionality reduction and improve the representational ability of data at the same time.

[0115] S4: Build a cable operation status monitoring model based on a BP neural network optimized by an improved genetic algorithm to perform in-depth monitoring and feedback on the obtained cable status feature data; the specific steps are as follows:

[0116] Build a cable operation status monitoring model based on a BP neural network optimized by a genetic algorithm to further monitor and provide status feedback on the cable operation status rated as abnormal, normal, and good:

[0117] The input layer of the BP neural network contains neurons, the hidden layer contains neurons, and the output layer contains neurons.

[0118] The output of the hidden layer is:

[0119]

[0120] The output of the output layer is:

[0121]

[0122] where is the weight between the th neuron in the input layer and the th neuron in the hidden layer, is the output signal of the th neuron in the input layer, is the weight between the th neuron in the hidden layer and the th neuron in the output layer, is the activation function;

[0123] Optimize the initial weights and output layer thresholds in the BP neural network based on an improved genetic algorithm:

[0124] The population consists of Combined by chromosomes, each chromosome contains

[0125] gene sequences. Iteration is performed through crossover and mutation of the gene sequences. The scoring error function is set as the fitness function value. The greater the fitness value, the greater the probability of being selected;

[0126] The selection probability of chromosomes is as follows;

[0127]

[0128] Among them, is the reciprocal of the fitness value of the th chromosome;

[0129] The crossover probability of genes on the chromosome is:

[0130]

[0131] Among them, represents the crossover probability of genes, and respectively represent the minimum crossover probability and the maximum crossover probability of genes, represents the fitness of genes, represents the minimum fitness of genes, represents the average fitness of genes, is a mathematical constant;

[0132] The mutation probability of genes on the

[0133]

[0134] chromosome is: represents the mutation probability of genes; and respectively represent the minimum mutation probability and the maximum mutation probability of genes.

[0135] Through the improvement of the above improved genetic algorithm, it can effectively retain excellent individuals, accelerate the elimination of inferior individuals, coordinate the contradiction between the diversity of the population and the convergence of the algorithm, and perform adaptive adjustment according to the current state of the BP neural network, improving the monitoring effect of the BP neural network.

[0136] In the prior art, the monitoring of the cable operation state based on the BP neural network can learn the instance set with correct cable operation state data and automatically extract reasonable solution rules, that is, it has the ability of self-learning and certain generalization ability. However, the learning speed of the BP neural network is slow and it is easy to fall into the situation of local extreme values, resulting in the failure of training. Therefore, the present invention monitors the cable operation state through the BP neural network improved by the improved genetic algorithm, and the specific monitoring accuracy can be referred to Figure 2 According to Figure 2 it can be clearly observed that the monitoring effect of the BP neural network optimized by the genetic algorithm is better than that of the BP neural network, and the monitoring accuracy of the cable operation state is significantly improved.

[0137] Based on multiple warning methods such as information warning, email warning, phone warning and network connection warning, the feedback of the in-depth monitoring results can ensure that the cable operation state is received and monitored to the greatest extent, and the abnormal state of the cable can be quickly processed.

[0138] Referring to Figure 3 this embodiment provides a cable operation state monitoring system based on deep learning, including:

[0139] Data acquisition module 100: used to acquire the cable state data obtained in real time in the sensing device;

[0140] State monitoring module 200: used to conduct preliminary monitoring, scoring and feedback on the cable state data;

[0141] Feature extraction module 300: used to deeply extract the cable state data based on the HOG for feature extraction to obtain cable state feature data based on the feature screening strategy;

[0142] Deep monitoring module 400: used to build a cable operation state monitoring model based on the BP neural network optimized by the improved genetic algorithm to conduct in-depth monitoring and feedback on the obtained cable state feature data;

[0143] The data acquisition module 100 further includes a data filtering unit 110, and the data filtering unit 110 is used to filter the cable state data based on the improved wavelet threshold algorithm;

[0144] The state monitoring module 200 includes a state scoring unit 210 and a state feedback unit 220,

[0145] The state scoring unit 210 conducts a preliminary score on the cable operation state based on the preset cable state data operation range;

[0146] The state feedback unit 220 conducts feedback based on the scoring result of the state scoring unit 210.

[0147] As described above, it is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent replacements or changes should be covered within the protection scope of the present invention.

Claims

1. A cable operation status monitoring method based on deep learning, characterized in that: The steps include: S1: Collect the cable status data obtained in real time in the sensor equipment, and filter the cable status data based on the improved wavelet threshold algorithm; S2: Monitor the cable status data after filtering, and make a preliminary score and feedback on the cable operation status based on the preset cable status data operation range. When the monitored cable status data exceeds the cable status depth monitoring score threshold, go to S3, otherwise return to S1; S3: Based on the feature screening strategy, the cable status data extracted based on HOG is deeply extracted to obtain the cable status feature data; The feature screening strategy is as follows: Get A collection of pattern classes: in, For the The following model class The first cable status data HOG features; for Total number of cable status data in class mode; The number of HOG features for each cable status data; In each mode there are Cable status data, total cable status data, then there is HOG features, the feature set is ; Calculate the average distance of samples under the same condition : in, For the The following model class The first cable status data HOG features; For the The following model class The first cable status data HOG features; Averaging to get the average intra-class distance : Calculated variance factor : Calculate the average distance between cable status data under different conditions : Calculate the average inter-class distance between cable status data under different conditions : in, For the The average distance between cable status data in each mode; calculate The variance factor : Calculate the compensation factor for: Calculate and normalize: in, The first Cable status characteristic data; S4: A cable operation status monitoring model is built based on the BP neural network optimized by the improved genetic algorithm to conduct in-depth monitoring and feedback of the acquired cable status characteristic data.

2. The cable operation status monitoring method based on deep learning according to claim 1 is characterized in that: The cable status data in S1 includes cable operation electrical quantity data and cable operation physical quantity data.

3. The cable operation status monitoring method based on deep learning according to claim 1 is characterized in that: The improved wavelet threshold algorithm described in S1 is as follows: in, For the Layer estimated wavelet coefficients; For the decomposition Layer True wavelet coefficients; After selection Threshold value; is a mathematical constant; is a step function; is the median function; is a high-order factor.

4. The cable operation status monitoring method based on deep learning according to claim 1 is characterized in that: In S2, the cable data is graded according to the operating range of the cable status data, and status feedback of the cable data is performed according to the scoring grade division result of the current cable data.

5. The cable operation status monitoring method based on deep learning according to claim 4 is characterized in that: The rating levels include abnormal, fair, good and excellent.

6. The cable operation status monitoring method based on deep learning according to claim 5 is characterized in that: S3 extracts and screens the cable operation states scored as abnormal, normal, and good according to the state feedback results described in S2.

7. The cable operation status monitoring method based on deep learning according to claim 1 is characterized in that: The cable operation status monitoring model described in S4 is specifically constructed as follows: The input layer of the BP neural network contains neurons, and the hidden layer contains neurons, and the output layer contains neurons, Hidden layer output for: Output layer output for: in, The input layer neurons and hidden layer The weights between neurons, The input layer The output signal of a neuron, The hidden layer The neurons in the output layer The weights between neurons, is the activation function; Based on the improved genetic algorithm, the initial weights and output layer thresholds in the BP neural network are optimized: Population by Composed of chromosomes, Chromosomes contain Gene sequences are iterated by crossover and mutation, and the scoring error function is set as the fitness function value. The larger the fitness value, the greater the probability of being selected. The population is optimized based on the following selection probability, crossover probability and mutation probability: The probability of selection of chromosome as follows; in, For the The inverse of the fitness value of the chromosome; The crossover probability of genes on chromosomes for: in, represents the crossover probability of the gene, and Respectively represent the minimum crossover probability and maximum crossover probability of the gene, represents the fitness of the gene, represents the minimum fitness of the gene, represents the average fitness of the gene, is a mathematical constant; The probability of mutation of genes on chromosomes is: in, represents the probability of gene mutation; and They represent the minimum and maximum mutation probabilities of the gene respectively.

8. The cable operation status monitoring method based on deep learning according to claim 1 is characterized in that: The in-depth monitoring feedback described in S4 is based on information alerts, email alerts, telephone alerts and network alerts.

9. A cable operation status monitoring system based on deep learning using the method described in any one of claims 1 to 8, characterized in that: include: Data acquisition module (100): used to collect cable status data acquired in real time in the sensing device; Status monitoring module (200): used for preliminary monitoring, scoring and feedback of cable status data; Feature extraction module (300): used to perform deep extraction on the cable status data extracted based on HOG based on a feature screening strategy to obtain cable status feature data; A deep monitoring module (400): used to build a cable operation status monitoring model based on a BP neural network optimized by an improved genetic algorithm to perform deep monitoring and feedback on the acquired cable status characteristic data; The data acquisition module (100) further comprises a data filtering unit (110), wherein the data filtering unit (110) is used to perform filtering processing on the cable status data based on an improved wavelet threshold algorithm; The state monitoring module (200) comprises a state scoring unit (210) and a state feedback unit (220). The state scoring unit (210) performs a preliminary scoring of the cable operation state based on a preset cable state data operation range; The state feedback unit (220) provides feedback based on the scoring result of the state scoring unit (210).

Images