Fire extinguishing decision method for cable tunnel based on multi-sensor data fusion and fwa-bp neural network algorithm
By using multi-sensor data fusion and the FWA-BP neural network algorithm, the problem of high false alarm rate in the automatic fire alarm system for cable tunnels was solved, and the safe and stable operation of cable tunnels and timely fire suppression were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- STATE GRID SHANGHAI MUNICIPAL ELECTRIC POWER CO
- Filing Date
- 2022-11-14
- Publication Date
- 2026-05-29
Smart Images
Figure CN115905998B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of online monitoring and fault diagnosis of power equipment, specifically a cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm. Background Technology
[0002] In production sites with concentrated power equipment, such as substations and power plants, power cables and communication cables are typically laid in cable trenches. In actual production, due to site conditions, construction costs, and various other reasons, cable trenches often have unfavorable characteristics such as narrow spaces and dense cable density. These factors hinder routine line maintenance and repair, and easily lead to fires due to short circuits, overloads, increased resistance, etc., resulting in serious production accidents. Statistics on cable fire accidents show that in the past 20 years, my country has experienced numerous cable-related fire accidents, especially cable fires in thermal power plants and substations, totaling approximately 140 incidents, with over 70% of these cable fires causing severe losses.
[0003] However, in most cases, a series of warning events must occur before a cable fire breaks out. These events include abnormally high cable temperatures, the generation of smoke particles, and other warning signals. If fire risks can be identified and faults eliminated in the early stages of a fire, fire accidents can be effectively prevented, ensuring the safe operation of industrial equipment. Currently, automatic fire alarm systems in cable tunnels suffer from low reliability or poor adaptability to harsh environments, resulting in short service life, significant difficulties in maintenance, and frequent false alarms. This causes monitoring personnel to lose awareness of the alarms, making it difficult to detect and effectively control fires in their early stages. Summary of the Invention
[0004] The purpose of this invention is to address the problems existing in the prior art by providing a cable tunnel fire extinguishing decision-making method that combines multi-sensor data fusion with the FWA-BP neural network algorithm. This cable tunnel fire extinguishing decision-making method solves the problems of outliers, incompleteness, and even inconsistent observation targets in the measurement data of different sensors due to the complexity and uncertainty of the observation environment by using multi-parameter, multi-point, and two-dimensional data fusion. This greatly reduces the false alarm rate of fire alarms and can effectively ensure the safe and stable operation of high-voltage cables.
[0005] The objective of this invention is achieved through the following technical solution:
[0006] A fire extinguishing decision-making method for cable tunnels, combining multi-sensor data fusion with an FWA-BP neural network algorithm, is characterized by the following: the multi-parameter sensors used in this fire extinguishing decision-making method are arranged at multiple locations, and each location's multi-parameter sensors include at least one temperature sensor, one CO sensor, and one smoke sensor. The steps of this fire extinguishing decision-making method are as follows:
[0007] A. An improved fuzzy support function is used to fuse multi-point monitoring data with the same parameters to obtain fused temperature data, fused CO concentration data, and fused smoke concentration data.
[0008] B. Perform principal component analysis to reduce noise in the obtained multi-parameter fusion data, and create and train a multi-parameter data fusion model based on the FWA-BP neural network algorithm, namely the cable fire FWA-BP neural network model.
[0009] C. Input the test data into the FWA-BP neural network model for cable fire verification, and output the fire extinguishing decision and fire extinguishing range.
[0010] The temperature sensor is attached to the metal support of the cable in the cable tunnel using thermally conductive silicone to output a temperature signal to monitor the cable condition. The CO sensor and smoke sensor are installed according to the sensing radiation range.
[0011] The improved fuzzy support function in step A replaces the absolute distance in the DTW method with dynamic bending distance to calculate the support between monitoring data from the same type of sensor.
[0012] The specific steps of the improved fuzzy support function are as follows:
[0013] A1. The formula for calculating the support between monitoring data from sensors of the same type is as follows:
[0014] (1)
[0015] In equation (1), For time series data and The dynamic bending distance, ≥0, The larger the value, the higher the degree of differentiation in mutual support;
[0016] A2. Multiple similar sensors respectively collect temperature signals, CO concentration signals, and smoke concentration signals to obtain a dataset. , The number of sensors of the same type can be obtained from the time period according to formula (1). Within, the support between data from different locations of similar sensors Then, a fuzzy support matrix is constructed. ,
[0017] (2);
[0018] A3, Time Period Inside, sensors other than themselves are for the first The support of each sensor is shown in formula (3), and the support value range is [0,1].
[0019] (3)
[0020] In equation (3), For sensors other than themselves, for the first Support for individual sensors;
[0021] A4, if the first Each sensor in a time period The data collected internally is , If is the number of data points, then the th The mean and variance of the data from each sensor node are shown in Equations (4) and (5), respectively:
[0022] (4)
[0023] (5)
[0024] In equations (4) and (5), For the first The average value of data from each sensor node. For the first The variance of data from each sensor node;
[0025] A5. Data fusion between similar sensors is achieved through a weighted fusion algorithm. The final weighted fusion expression is:
[0026] (6)
[0027] In equation (6), Indicates the first Each sensor in a time period The weighted values corresponding to the time series data collected internally are calculated as shown in formula (7):
[0028] (7).
[0029] The number of similar sensors in step A2 is at least 3.
[0030] The principal component analysis (PCA) noise reduction method used in step B can extract feature information from high-dimensional data, reduce dimensionality, and retain the maximum amount of information in the original high-dimensional data. Specifically, in the cable trench fire monitoring system… Total per unit time The fused data is composed of data collected simultaneously from different sensors. OK Column sample matrix For the sample matrix By transforming the matrix using formulas (8), (9), and (10), the standardized matrix can be obtained. ,
[0031] (8)
[0032] (9)
[0033] (10)
[0034] The standardized matrix is obtained by formula (11). Find the correlation coefficient matrix And solve the characteristic equation shown in formula (12).
[0035] (11)
[0036] (12)
[0037] Further obtain Each feature value, based on Sure Value, i.e., the current value When the information contribution rate of each principal component is greater than 85%, the preceding... Using the principal components as sample features, the feature values are sorted in descending order, and the top ones are taken. The transformation matrix is composed of the eigenvectors corresponding to the eigenvalues. :
[0038] (13)
[0039] calculate Before Principal Components This can achieve the goal of dimensionality reduction of fire monitoring fusion data; It is a sample matrix with n1 rows and m1 columns.
[0040] The BP neural network in the FWA-BP neural network algorithm in step B is a multi-layer feedforward network, which includes an input layer, a hidden layer, and an output layer. The input layer has 3 nodes, which are temperature fusion data, CO concentration fusion data, and smoke concentration fusion data, respectively. The output layer has 2 nodes, which are the fire status risk assessment value and the trigger value for whether to trigger the fire extinguishing system. The trigger value only includes two states: 0 and 1, where 0 indicates no trigger and 1 indicates trigger. The hidden layer of the BP neural network uses a non-linear Sigmoid function, while the output layer often uses a linear Purelin function. The Sigmoid function is divided into logsig types, and the expressions are shown in equations (14) and (15), respectively.
[0041] (14)
[0042] (15).
[0043] Before the FWA optimization begins in step B of the FWA-BP neural network algorithm, it is necessary to determine the population size, the upper and lower limits of the firework dimension, the explosion spark radius adjustment constant, the explosion spark number adjustment constant, the Gaussian spark number, and the maximum number of iterations. Then, the fitness value of each firework is calculated, using the squared error and SSE of the neural network as the fitness function.
[0044] (16)
[0045] In equation (16), The expected output of the BP neural network is... This represents the number of neurons in the output layer of the BP neural network. This represents the actual output value of the BP neural network.
[0046] Finally, the Levenberg-Marquardt algorithm is used to train and optimize the weights and thresholds of the BP neural network with higher precision. The training objective error function is set as squared error and SSE. When the training reaches the target error or the maximum number of iterations, the ideal cable fire FWA-BP neural network model is established. Otherwise, return to the previous step to continue training.
[0047] The present invention has the following advantages over the prior art:
[0048] This invention provides a cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm. It uses multi-parameter, multi-point sensors to monitor fire characteristic parameters and performs multi-parameter data fusion based on the FWA-BP neural network algorithm. By using multi-parameter, multi-point, two-dimensional data fusion, it solves the problems of outliers, incompleteness, and even inconsistencies in the measurement data of different sensors due to the complexity and uncertainty of the observation environment. After identifying obvious abnormal faults, the fire is displayed on the terminal in a timely manner, triggering the fire extinguishing device. This greatly reduces the false alarm rate of fire alarms, avoids the occurrence of serious accidents, and effectively ensures the safe and stable operation of high-voltage cables. Attached Figure Description
[0049] Appendix Figure 1 This is a flowchart illustrating the overall algorithm of the cable tunnel fire extinguishing decision-making method of the present invention.
[0050] Appendix Figure 2 This is a flowchart illustrating the establishment of the FWA-BP neural network model for cable fires according to the present invention.
[0051] Appendix Figure 3 The figure shows the prediction results after inputting test data into the FWA-BP neural network model for cable fires of the present invention. Detailed Implementation
[0052] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0053] like Figure 1 As shown, this invention provides a cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm. This method includes a multi-parameter, multi-point sensing device, a multi-parameter data fusion method, and a multi-parameter data fusion method based on the FWA-BP neural network algorithm. In the multi-parameter, multi-point sensing device, each point's multi-parameter sensor includes at least one temperature sensor, one CO sensor, and one smoke sensor. The temperature sensor is attached to the metal support of the cable in the cable tunnel using thermally conductive silicone to output a temperature signal monitoring the cable's condition. The CO sensor and smoke sensor are installed according to their sensing radiation range.
[0054] The method for fusing multi-point data with the same parameters fuses sensor data with the same parameters by improving the data fusion method of fuzzy support. Specifically, this invention improves the fuzzy support function by replacing the absolute distance in the DTW method with dynamic bending distance, and calculates the support between monitoring data from the same type of sensor, as shown in equation (1):
[0055] (1)
[0056] In equation (1), For time series data and The dynamic bending distance, ≥0, The larger the value, the higher the degree of differentiation in mutual support;
[0057] This invention uses multiple similar sensors to collect temperature signals, CO concentration signals, and smoke concentration signals respectively, to obtain a dataset. , The number of sensors of the same type A larger numerical value results in better data fusion, but it also excessively wastes sensor resources and increases the cost of fire extinguishing decision-making methods. Therefore, this application recommends a minimum of three similar sensors. The time period can be obtained from formula (1). Within, the support between data from different locations of similar sensors Then, a fuzzy support matrix is constructed. ,
[0058] (2);
[0059] Time period Inside, sensors other than themselves are for the first The support of each sensor is shown in formula (3), and the support value range is [0,1].
[0060] (3)
[0061] In equation (3), For sensors other than themselves, for the first Support for individual sensors;
[0062] If the first Each sensor in a time period The data collected internally is , If is the number of data points, then the th The mean and variance of the data from each sensor node are shown in Equations (4) and (5), respectively:
[0063] (4)
[0064] (5)
[0065] In equations (4) and (5), For the first The average value of data from each sensor node. For the first The variance of data from each sensor node;
[0066] Data fusion among similar sensors is achieved using a weighted fusion algorithm. The final weighted fusion expression is as follows:
[0067] (6)
[0068] In equation (6), Indicates the first Each sensor in a time period The weighted values corresponding to the time series data collected internally are calculated as shown in formula (7):
[0069] (7).
[0070] The multi-parameter data fusion method based on the FWA-BP neural network algorithm includes data acquisition, principal component analysis and dimensionality reduction, creation and training of a multi-parameter data fusion model based on the FWA-BP neural network algorithm, prediction of fire results based on detection data, and decision on whether to trigger the fire extinguishing system and the fire extinguishing range.
[0071] Principal component analysis (PCA), a widely used data processing technique, can extract feature information from high-dimensional data, reduce dimensionality, and retain the maximum amount of information from the original high-dimensional data. Specifically, in a cable trench fire monitoring system… Total per unit time The fused data is composed of data collected simultaneously from different sensors. OK Column sample matrix For the sample matrix By transforming the matrix using formulas (8), (9), and (10), the standardized matrix can be obtained. .
[0072] (8)
[0073] (9)
[0074] (10)
[0075] The standardized matrix is obtained by formula (11). Find the correlation coefficient matrix And solve the characteristic equation shown in formula (12).
[0076] (11)
[0077] (12)
[0078] Further obtain Each feature value, based on Sure Value, i.e., the current value When the information contribution rate of each principal component is greater than 85%, the preceding... Using the principal components as sample features, the feature values are sorted in descending order, and the top ones are taken. The transformation matrix is composed of the eigenvectors corresponding to the eigenvalues. :
[0079] (13)
[0080] calculate Before Principal Components This can achieve the goal of dimensionality reduction of fire monitoring fusion data; It is a sample matrix with n1 rows and m1 columns.
[0081] like Figure 2 The flowchart shown is for establishing the FWA-BP neural network model for cable fires. The FWA-BP neural network algorithm first establishes a BP neural network, which is a multi-layer feedforward network. The multi-layer network includes an input layer, a hidden layer, and an output layer. The input layer has 3 nodes, which are temperature fusion data, CO concentration fusion data, and smoke concentration fusion data, respectively. The output layer has 2 nodes, which are the fire status risk assessment value and the trigger value for whether to trigger the fire extinguishing system. The trigger value only includes two states: 0 and 1, where 0 indicates no trigger and 1 indicates trigger. The hidden layer of the BP neural network uses the non-linear Sigmoid function, while the output layer often uses the linear Purelin function. The Sigmoid function is divided into logsig types, and the expressions are shown in equations (14) and (15), respectively.
[0082] (14)
[0083] (15).
[0084] Before starting the FWA optimization, it is necessary to determine the population size, the upper and lower limits of the fireworks dimension, the explosion spark radius adjustment constant, the explosion spark number adjustment constant, the Gaussian spark number, and the maximum number of iterations. Then, the fitness value of each fireworks is calculated, using the squared error and SSE of the neural network as the fitness function.
[0085] (16)
[0086] In equation (16), The expected output of the BP neural network is... This represents the number of neurons in the output layer of the BP neural network. This represents the actual output value of the BP neural network.
[0087] Finally, the Levenberg-Marquardt algorithm is used to train and optimize the weights and thresholds of the BP neural network with higher precision. The training objective error function is set as squared error and SSE. When the training reaches the target error or the maximum number of iterations, the ideal cable fire FWA-BP neural network model is established. Otherwise, return to the previous step to continue training.
[0088] like Figure 3 As shown, the prediction results of the FWA-BP neural network model for cable fires are input as test data. Based on published papers, field surveys, and the cable fire test involved in this invention, 380 sets of data were selected as training data, from which 9 data points were randomly selected as test data for testing. From... Figure 3 As can be seen, the predicted results are basically consistent with the expected results, and the three fire sample data in the elliptical area of the figure can be successfully identified.
[0089] The above model enables the prediction and assessment of cable tunnel fires. When a fire is identified based on monitored parameters, an ultrafine dry powder fire extinguishing system is triggered to extinguish the fire. Furthermore, the fire extinguishing range is determined through the arrangement of multiple sensors.
[0090] In this invention, the ground fire monitoring terminal can view temperature, CO concentration and smoke concentration monitoring data in real time, and can also view the operating status of each temperature sensor, CO concentration sensor and smoke concentration sensor. When a data disconnection is detected, the system fault can be troubleshooted based on the prompt information data, which is convenient for later maintenance.
[0091] It should be added that: Typically, fuzzy support functions are measured by absolute distance. If the absolute distance between two sensor data points at any given time is small, their fuzzy support is large; conversely, if the absolute distance between two data points is large, their fuzzy support is small. However, when processing time-series sensor data, this method ignores the correlation information between data points before and after a given time, and estimating the mutual support provided by two sensors solely based on data proximity is impractical. Therefore, to address the above problems, this invention employs an improved fuzzy support function. This improved fuzzy support function replaces the absolute distance in the DTW method with dynamic bending distance to calculate the support between monitoring data from similar sensors.
[0092] This invention provides a cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm. It uses multi-parameter, multi-point sensors to monitor fire characteristic parameters and performs multi-parameter data fusion based on the FWA-BP neural network algorithm. By using multi-parameter, multi-point, two-dimensional data fusion, it solves the problems of outliers, incompleteness, and even inconsistencies in the measurement data of different sensors due to the complexity and uncertainty of the observation environment. After identifying obvious abnormal faults, the fire is displayed on the terminal in a timely manner, triggering the fire extinguishing device. This greatly reduces the false alarm rate of fire alarms, avoids the occurrence of serious accidents, and effectively ensures the safe and stable operation of high-voltage cables.
[0093] The cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm provided by this invention is suitable for complex on-site implementation environments in cable tunnels, has low cost, and can serve as an effective supplement to traditional temperature measurement fiber optic methods to ensure the safe and stable operation of high-voltage cables.
[0094] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention. Technologies not covered in this invention can be implemented using existing technologies.
Claims
1. A decision-making method for fire suppression in cable tunnels based on multi-sensor data fusion combined with the FWA-BP neural network algorithm, characterized in that: The fire extinguishing decision-making method employs multi-parameter sensors deployed at multiple locations. Each location's multi-parameter sensors include at least one temperature sensor, one CO sensor, and one smoke sensor. The steps of this fire extinguishing decision-making method are as follows: A. An improved fuzzy support function is used to fuse multi-point monitoring data with the same parameters to obtain fused temperature data, fused CO concentration data, and fused smoke concentration data. B. Perform principal component analysis to reduce noise in the obtained multi-parameter fusion data, and create and train a multi-parameter data fusion model based on the FWA-BP neural network algorithm, namely the cable fire FWA-BP neural network model. C. Input the test data into the FWA-BP neural network model for cable fire verification, and output the fire extinguishing decision and fire extinguishing range; The specific steps of the improved fuzzy support function are as follows: A1. The formula for calculating the support between monitoring data from sensors of the same type is as follows: （1） In equation (1), For time series data and The dynamic bending distance, ≥0, The larger the value, the higher the degree of differentiation in mutual support; A2. Multiple similar sensors respectively collect temperature signals, CO concentration signals, and smoke concentration signals to obtain a dataset. , The number of sensors of the same type can be obtained from the time period according to formula (1). Within, the support between data from different locations of similar sensors Then, a fuzzy support matrix is constructed. , （2）； A3, Time Period Inside, sensors other than themselves are for the first The support of each sensor is shown in formula (3), and the support value range is [0,1]. （3） In equation (3), For sensors other than themselves, for the first Support for individual sensors; A4, if the first Each sensor in a time period The data collected internally is , If is the number of data points, then the th The mean and variance of the data from each sensor node are shown in Equations (4) and (5), respectively: （4） （5） In equations (4) and (5), For the first The average value of data from each sensor node. For the first The variance of data from each sensor node; A5. Data fusion between similar sensors is achieved through a weighted fusion algorithm. The final weighted fusion expression is: （6） In equation (6), Indicates the first Each sensor in a time period The weighted values corresponding to the time series data collected internally are calculated as shown in formula (7): （7）； The BP neural network in the FWA-BP neural network algorithm in step B is a multi-layer feedforward network, which includes an input layer, a hidden layer, and an output layer. The input layer has 3 nodes, which are temperature fusion data, CO concentration fusion data, and smoke concentration fusion data, respectively. The output layer has 2 nodes, which are the fire status risk assessment value and the trigger value for whether to trigger the fire extinguishing system. The trigger value only includes two states: 0 and 1, where 0 indicates no trigger and 1 indicates trigger. The hidden layer of the BP neural network uses a non-linear Sigmoid function, while the output layer often uses a linear Purelin function. The Sigmoid function is divided into logsig types, and the expressions are shown in equations (14) and (15), respectively. （14） （15）。 2. The cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm according to claim 1, characterized in that: The temperature sensor is attached to the metal support of the cable in the cable tunnel using thermally conductive silicone to output a temperature signal to monitor the cable condition. The CO sensor and smoke sensor are installed according to the sensing radiation range.
3. The cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm according to claim 1, characterized in that: The number of similar sensors in step A2 is at least 3.
4. The cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm according to claim 1, characterized in that: The principal component analysis (PCA) noise reduction method used in step B can extract feature information from high-dimensional data, reduce dimensionality, and retain the maximum amount of information in the original high-dimensional data. Specifically, in the cable trench fire monitoring system, n1 represents the fused data collected simultaneously by m1 different sensors per unit time, forming an n1-row, m1-column sample matrix. For the sample matrix By transforming the matrix using formulas (8), (9), and (10), the standardized matrix can be obtained. , （8） （9） （10） The standardized matrix is obtained by formula (11). Find the correlation coefficient matrix And solve the characteristic equation shown in formula (12). (11) (12) Further obtain Each feature value, based on Sure Value, i.e., the current value When the information contribution rate of each principal component is greater than 85%, the preceding... Using the principal components as sample features, the feature values are sorted in descending order, and the top ones are taken. The transformation matrix is composed of the eigenvectors corresponding to the eigenvalues. : (13) calculate Before Principal Components This can achieve the goal of dimensionality reduction of fire monitoring fusion data; It is a sample matrix with n1 rows and m1 columns.
5. The cable tunnel fire extinguishing decision-making method based on multi-sensor data fusion combined with the FWA-BP neural network algorithm according to claim 1, characterized in that: Before the FWA optimization begins in step B of the FWA-BP neural network algorithm, it is necessary to determine the population size, the upper and lower limits of the firework dimension, the explosion spark radius adjustment constant, the explosion spark number adjustment constant, the Gaussian spark number, and the maximum number of iterations. Then, the fitness value of each firework is calculated, using the squared error and SSE of the neural network as the fitness function. (16) In equation (16), The expected output of the BP neural network is... This represents the number of neurons in the output layer of the BP neural network. This represents the actual output value of the BP neural network. Finally, the Levenberg-Marquardt algorithm is used to train and optimize the weights and thresholds of the BP neural network with higher precision. The training objective error function is set as squared error and SSE. When the training reaches the target error or the maximum number of iterations, the ideal cable fire FWA-BP neural network model is established. Otherwise, return to the previous step to continue training.