Cable operation state monitoring method for urban underground pipe gallery

Abstract

The invention discloses a cable operation state monitoring method for an urban underground pipe gallery, and the method comprises the steps: 1, building a cable temperature monitoring classifier based on a support vector machine, and carrying out the self-adaptive parameter setting; 2, classifying the cable data according to months to obtain 12 data sets, and taking all the cable data as one data set, namely 13 data sets in total; respectively training the 13 data sets by adopting a long-short-term memory neural network, and integrating 13 training networks by training prediction precision weighting; and 3, taking the acquired cable temperature data as an input value, and judging the running state of the cable through the integrated network. And optimizing the support vector machine classifier based on a particle swarm algorithm, and judging the operation state of the cable through the optimized support vector machine classifier.

Description

Technical Field

[0001] The invention relates to a cable operation status monitoring method for an urban underground pipe gallery, belonging to the technical field of electric power. Background Art

[0002] Urban underground utility corridors, as an important type of centralized, functional underground public building, are widely used around the world. Within these corridors, specialized pipelines for energy, water supply and drainage, communications, and other specialized services are installed, along with maintenance, hoisting, and monitoring facilities to ensure the smooth operation of the city. These various pipelines are uniformly designed and laid out in strict accordance with specialized requirements.

[0003] With the development of urbanization, more and more cities are using power cables to build underground power grids. If there is an accident with the cables in the power grid, it will cause huge economic losses and seriously affect people's lives. Therefore, the operating status monitoring of cables in underground pipeline corridors has become particularly important. Summary of the Invention

[0004] The present invention designs and develops a cable operation status monitoring method for urban underground pipe corridors, which judges the operation status of the cable based on an optimized support vector machine classifier.

[0005] The technical solution provided by the present invention is:

[0006] A method for monitoring the operating status of cables in urban underground pipe corridors, comprising:

[0007] Step 1: Construct a cable temperature monitoring classifier based on support vector machine and set adaptive parameters;

[0008] Step 2: Classify the cable data by month and obtain 12 data sets. All cable data are regarded as one data set, for a total of 13 data sets.

[0009] The 13 datasets were trained separately using long short-term memory neural networks, and the 13 training networks were integrated by weighting the training prediction accuracy.

[0010] Step 3: Use the cable temperature data as the input value of the cable temperature monitoring classifier and judge the operating status of the cable through the integrated network;

[0011] Abnormal judgment of cable temperature based on long short-term memory network;

[0012] For data sets R1, R2, ... R 12 Perform network training to obtain the LSTM neural network corresponding to each data set;

[0013] Train the LSTM neural network based on the training data set to obtain LSTM0;

[0014] Based on LSTM0, the cable temperature prediction values TPP1 and TPP2 of the prediction day are predicted respectively;

[0015] Calculate the data points (TPP1, TPP2) to the cluster centers CEN1, CEN2, ... CEN in sequence 12 The Euclidean distance is obtained by DF1, DF2, ... DF 12 ;

[0016] Set DF1, DF2, ... DF 12 Medium DF i Minimum, then choose LSTM i As a network for secondary prediction, based on LSTM i Predict and obtain the cable temperature prediction values TPP1' and TPP2' on the prediction day respectively;

[0017] Set the final prediction value to TPP1 F 、TPP2 F ;

[0018] The predicted sequence TPP1 F and TPP2 F The data is input into the support vector machine classifier based on particle swarm optimization to judge the cable status.

[0019] Preferably, the step 1 includes:

[0020] The cable temperature data points are used as training data for the support vector machine classifier to perform binary classification training on abnormal data and normal data. The support vector machine classification line equation is:

[0021] wx+b=0;

[0022] Where w and b are the parameters of the support vector machine classification line equation,

[0023] Assume that the closest sample point to the support vector machine straight line equation in the normal data point set is:

[0024] X1 = (tp1, tp2), the sample point closest to the support vector machine linear equation in the abnormal data point set is: X2 = (tp1', tp2'); (tp1, tp2) and (tp1', tp2') are normal and abnormal data collected by two different temperature sensors, respectively. Assuming the support vector machine classification linear equation is P, the calculation formula for the support vector machine generalization ability evaluation index is:

[0025] E SVM =d(X1,P)+d(X2,P);

[0026] Where d(X1,P) is the Euclidean distance from point X1 to line P, and d(X2,P) is the Euclidean distance from X2 to line P.

[0027] Preferably, it also includes:

[0028] Optimize the support vector machine classifier based on particle swarm optimization;

[0029] According to the characteristics of the support vector machine classifier, the parameter form of particle p is set to p = [c, g], and the fitness of the particle swarm is:

[0030] Fitnes=α SVM +β;

[0031] Where, α SVM is the fitness coefficient, β is a constant;

[0032] Based on the particle swarm algorithm to optimize the support vector machine, find the global optimal particle p best Maximize the fitness value;

[0033] According to p best Obtain optimal parameter settings for the support vector machine.

[0034] Preferably, the step 2 includes:

[0035] The cable temperature data are classified according to 12 months, and 12 data sets are obtained, which are marked as R1, R2, ..., R 12 ;

[0036] Calculate the cluster center of each type of data and get 12 cluster centers, which are CEN i (i=1,2,...,12), the calculation formula is:

[0037]

[0038] In the formula, i is the clustered category, n is the number of data contained in the i-th category data, X j 、X k are two different data points in the i-th class, d(X j ,X k ) is the data point X j and X k The Euclidean distance of

[0039] When the sum of the Euclidean distances from any data point in each class of data to other data points is the smallest, then this point is the class center of this class of data.

[0040] Preferably, the step three includes:

[0041] The LSTM neural networks corresponding to each data set are marked as follows:

[0042] LSTM1, LSTM2, ... LSTM 12 ;

[0043] TPP1 and TPP2 are denoted as;

[0044] TPP1=[tpp11,tpp12,...tpp1 w ];

[0045] TPP2=[tpp21,tpp22,...tpp2 w ];

[0046] TPP1', TPP2' are denoted as:

[0047] TPP1'=[tpp11',tpp12',...tpp1 w '];

[0048] TPP2'=[tpp21',tpp22',...tpp2 w '];

[0049] Final predicted value TPP1 F 、TPP2 F for:

[0050]

[0051] Where, is the training prediction accuracy of LSTM0 for tp1 data, For LSTM i The training prediction accuracy of tp1 data, is the training prediction accuracy of LSTM0 for tp2 data, For LSTM i Training prediction accuracy on tp2 data.

[0052] Preferably, the prediction effect of the cable status is evaluated based on the average relative error, and the calculation formula of the average relative error is:

[0053]

[0054] Where N is the sample size, Y k is the actual value, and Y is the predicted value.

[0055] The present invention has the following beneficial effects: The present invention uses a particle swarm algorithm to optimize a support vector machine classifier, constructing a cable temperature data classifier. This optimized support vector machine classifier is then used to determine cable status. Historical cable temperature data is categorized by month, and a long-short-term memory neural network is trained using the data from each month. A single long-short-term memory neural network, trained on all historical cable data, is then trained using a single long-short-term memory neural network that has learned all cable temperature data. Based on ensemble learning, this method improves cable temperature data prediction accuracy by weighting training accuracy.

[0056] The predicted value of cable temperature data at future moments is used as the input value of the optimized support vector machine cable temperature data classifier to judge the operating status of the cable at future moments, thereby achieving the purpose of predicting the future status of the cable. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The present invention is a flow chart of the cable operation status monitoring method for urban underground pipeline corridors. DETAILED DESCRIPTION

[0058] The present invention will be described in further detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0059] like Figure 1 As shown, the present invention provides a cable operation status monitoring method for urban underground pipe corridors, which optimizes a support vector machine classifier based on a particle swarm algorithm, constructs a cable temperature data classifier, and judges the status of the cable through the optimized support vector machine classifier, including:

[0060] Step 1: Construct a cable temperature monitoring classifier based on support vector machine and set adaptive parameters;

[0061] Step 2: Classify the cable data by month and obtain 12 data sets. All cable data are regarded as one data set, for a total of 13 data sets.

[0062] The 13 datasets were trained separately using long short-term memory neural networks, and the 13 training networks were integrated by weighting the training prediction accuracy.

[0063] Step 3: Use the cable temperature data as the input value of the cable temperature monitoring classifier and judge the operating status of the cable through the integrated network;

[0064] The cable temperature data points are used as training data for the support vector machine classifier to perform binary classification training on abnormal data and normal data. The support vector machine classification line equation is:

[0065] wx+b=0;

[0066] Where w and b are the parameters of the support vector machine classification line equation,

[0067] Assume that the closest sample point to the support vector machine straight line equation in the normal data point set is:

[0068] X1 = (tp1, tp2), the sample point closest to the support vector machine linear equation in the abnormal data point set is: X2 = (tp1', tp2'); (tp1, tp2) and (tp1', tp2') are normal and abnormal data collected by two different temperature sensors, respectively. Assuming the support vector machine classification linear equation is P, the calculation formula for the support vector machine generalization ability evaluation index is:

[0069] E SVM =d(X1,P)+d(X2,P);

[0070] Where d(X1,P) is the Euclidean distance from point X1 to line P, and d(X2,P) is the Euclidean distance from point X2 to line P. SVM The larger it is, the better the generalization ability of the support vector machine classifier SVM is and the better the classification effect is.

[0071] The support vector machine classifier is optimized based on the particle swarm algorithm. According to the characteristics of the support vector machine classifier, the parameter form of particle p is set to p = [c, g], and the fitness of the particle swarm is:

[0072] Fitnes=α SVM +β;

[0073] Where, α SVM is the fitness coefficient, and β is a constant.

[0074] Based on the particle swarm algorithm to optimize the support vector machine, find the global optimal particle p best Maximize the fitness value; according to p best Obtain optimal parameter settings for the support vector machine.

[0075] In the present invention, as a preferred embodiment, the cable temperature data is related to the month, and the cable temperature data is classified according to 12 months to obtain 12 data sets, which are marked as R1, R2, ..., R 12 ;

[0076] Calculate the cluster center of each type of data and get 12 cluster centers, which are CEN i (i=1,2,...,12), the calculation formula is:

[0077]

[0078] In the formula, i is the clustered category, n is the number of data contained in the i-th category data, X j 、X k are two different data points in the i-th class, d(X j ,X k ) is the data point X j and X k The Euclidean distance of

[0079] When the sum of the Euclidean distances from any data point in each class of data to other data points is the smallest, then this point is the class center of this class of data.

[0080] Cable temperature data conforms to the characteristics of long-term time series data. In the present invention, as a preferred method, a long short-term memory neural network (LSTM) is selected as the cable temperature prediction algorithm, and a support vector machine (PSO-SVM) classifier optimized by a particle swarm algorithm is combined to perform cable temperature anomaly interpretation, including:

[0081] Based on long short-term memory neural network as abnormal judgment of cable temperature;

[0082] For data sets R1, R2, ... R 12 Perform network training to obtain the LSTM neural network corresponding to each data set; the LSTM neural network corresponding to each data set is marked as:

[0083] LSTM1, LSTM2, ... LSTM 12 ;

[0084] Train the LSTM neural network based on the training data set to obtain LSTM0;

[0085] Based on LSTM0, the cable temperature prediction values TPP1 and TPP2 of the prediction day are predicted respectively;

[0086] TPP1 and TPP2 are denoted as;

[0087] TPP1=[tpp11,tpp12,...tpp1 w ];

[0088] TPP2=[tpp21,tpp22,...tpp2 w ];

[0089] Calculate the data points (TPP1, TPP2) to the cluster centers CEN1, CEN2, ... CEN in sequence 12 The Euclidean distance is obtained by DF1, DF2, ... DF 12 ;

[0090] Set DF1, DF2, ... DF 12 Medium DFi Minimum, then choose LSTM i As a network for secondary prediction, based on LSTM i Predict and obtain the cable temperature prediction values TPP1' and TPP2' on the prediction day respectively;

[0091] TPP1', TPP2' are denoted as:

[0092] TPP1'=[tpp11',tpp12',...tpp1 w '];

[0093] TPP2'=[tpp21',tpp22',...tpp2 w '];

[0094] Set the final prediction value to TPP1 F 、TPP2 F :

[0095]

[0096]

[0097] Where, is the training prediction accuracy of LSTM0 for tp1 data, For LSTM i The training prediction accuracy of tp1 data, is the training prediction accuracy of LSTM0 for tp2 data, For LSTM i Training prediction accuracy on tp2 data.

[0098] TPP1 F and TPP2 F The data is input into the support vector machine classifier optimized by particle swarm algorithm to determine the operating status of the cable.

[0099] Example

[0100] Using cable temperature monitoring data from a power supply bureau over the past year as training samples, a prediction model was developed. A monitoring day was selected and the data was used to perform temperature predictions. Ten temperatures were predicted for any given time period on that day. The results are shown in Table 1.

[0101] In the present invention, as a preference, the prediction effect of the cable status is evaluated based on the average relative error, and the calculation formula of the average relative error is:

[0102]

[0103] Where N is the sample size, Y kis the actual value, and Y is the predicted value.

[0104] Table 1 Temperature prediction results

[0105]

[0106] As shown in Table 1, the proposed monitoring method achieves an average error of 8.22% and a minimum error of 7.68% for predicting the cable temperature tp1 at ten future time points. The average error for predicting the temperature tp2 at ten future time points is 8.19% and a minimum error of 7.61%. These data demonstrate the practicality and rationality of this method. When the predicted values for tp1 and tp2 are input into a support vector machine classifier optimized by a particle swarm algorithm, the calculated cable status is normal, demonstrating the effectiveness of this method.

[0107] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for monitoring the cable operation status of an urban underground pipeline corridor, characterized in that: include: Step 1: Construct a cable temperature monitoring classifier based on support vector machine and set adaptive parameters; Step 2: Classify the cable data by month and obtain 12 data sets. All cable data are regarded as one data set, for a total of 13 data sets. The 13 datasets were trained separately using long short-term memory neural networks, and the 13 training networks were integrated by weighting the training prediction accuracy. Step 3: Use the cable temperature data as the input value of the cable temperature monitoring classifier and judge the operating status of the cable through the integrated network; Abnormal judgment of cable temperature based on long short-term memory network; For data sets R1, R2, ... R 12 Perform network training to obtain the LSTM neural network corresponding to each data set; Train the LSTM neural network based on the training data set to obtain LSTM0; Based on LSTM0, the cable temperature prediction values TPP1 and TPP2 of the prediction day are predicted respectively; Calculate the data points (TPP1, TPP2) to the cluster centers CEN1, CEN2, ... CEN in sequence 12 The Euclidean distance is obtained by DF1, DF2, ... DF 12 ; Set DF1, DF2, ... DF 12 Medium DF i Minimum, then choose LSTM i As a network for secondary prediction, based on LSTM i Predict and obtain the cable temperature prediction values TPP1' and TPP2' on the prediction day respectively; Set the final prediction value to TPP1 F 、TPP2 F ; The predicted sequence TPP1 F and TPP2 F The data is input into the support vector machine classifier based on particle swarm optimization to judge the cable status.

2. The cable operation status monitoring method for urban underground pipe gallery according to claim 1 is characterized in that: The step one comprises: The cable temperature data points are used as training data for the support vector machine classifier to perform binary classification training on abnormal data and normal data. The support vector machine classification line equation is: wx+b=0; Where w and b are the parameters of the support vector machine classification line equation, Assume that the closest sample point to the support vector machine straight line equation in the normal data point set is: X1 = (tp1, tp2), the sample point closest to the support vector machine linear equation in the abnormal data point set is: X2 = (tp1', tp2'); (tp1, tp2) and (tp1', tp2') are normal and abnormal data collected by two different temperature sensors, respectively. Assuming the support vector machine classification linear equation is P, the calculation formula for the support vector machine generalization ability evaluation index is: E SVM =d(X1,P)+d(X2,P); Where d(X1,P) is the Euclidean distance from point X1 to line P, and d(X2,P) is the Euclidean distance from X2 to line P.

3. The cable operation status monitoring method for urban underground pipe gallery according to claim 2 is characterized in that: Also includes: Optimize the support vector machine classifier based on particle swarm optimization; According to the characteristics of the support vector machine classifier, the parameter form of particle p is set to p = [c, g], and the fitness of the particle swarm is: Fitness SVM +β; Where, α SVM is the fitness coefficient, β is a constant; Based on the particle swarm algorithm to optimize the support vector machine, find the global optimal particle p best Maximize the fitness value; According to p best Obtain optimal parameter settings for the support vector machine.

4. The cable operation status monitoring method for urban underground pipe gallery according to claim 3 is characterized in that: The second step includes: The cable temperature data are classified according to 12 months, and 12 data sets are obtained, which are marked as R1, R2, ..., R 12 ; Calculate the cluster center of each type of data and get 12 cluster centers, which are , CEN i (i=1,2,...,12), the calculation formula is: In the formula, i is the clustered category, n is the number of data contained in the i-th category data, X j 、X k are two different data points in the i-th class, d(X j ,X k ) is the data point X j and X k The Euclidean distance of When the sum of the Euclidean distances from any data point in each class of data to other data points is the smallest, then this point is the class center of this class of data.

5. The cable operation status monitoring method for urban underground pipe gallery according to claim 4 is characterized in that: The step three includes: The LSTM neural networks corresponding to each data set are marked as follows: LSTM1,LSTM2,...LSTM 12 ; TPP1 and TPP2 are denoted as; TPP1=[tpp11,tpp12,...tpp1 w ]; TPP2=[tpp21,tpp22,...tpp2 w ]; TPP1', TPP2' are denoted as: TPP1'=[tpp11',tpp12',...tpp1 w ']; TPP2'=[tpp21',tpp22',...tpp2 w ']; Final predicted value TPP1 F 、TPP2 F for: Where, is the training prediction accuracy of LSTM0 for tp1 data, For LSTM i The training prediction accuracy of tp1 data, is the training prediction accuracy of LSTM0 for tp2 data, For LSTM i Training prediction accuracy on tp2 data.

6. The cable operation status monitoring method for urban underground pipe gallery according to claim 5 is characterized in that: The prediction effect of cable status is evaluated based on the average relative error. The calculation formula of the average relative error is: Where N is the sample size, Y k is the actual value, and Y is the predicted value.

Images