Substation equipment health state monitoring system and method based on electric field intensity detection

Abstract

The invention discloses a substation equipment health state monitoring system and method based on electric field intensity detection, and the system consists of an electric field intensity sensor, an image recognition module, an infrared imaging module, a data processing and analysis module, a wireless transmission module, an inspection robot, and a health state detection platform. The inspection robot collects electric field distribution, a visible light image and an infrared thermogram according to a preset line, appearance defects and temperature abnormity are analyzed through a convolutional neural network, and the data processing and analysis module can generate an electric field thermodynamic diagram and evaluate the equipment fault risk in real time through a deep learning algorithm. Abnormal information can be wirelessly transmitted to the health state detection platform, and an alarm is automatically given. Through multi-source data fusion and intelligent diagnosis, the system can effectively improve the inspection efficiency and accuracy, reduce manual errors, realize early detection of equipment cracks, corrosion, looseness and high-temperature overheating defects, and greatly improve the operation and maintenance efficiency and safety of a transformer substation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment inspection, and in particular relates to a substation equipment health status monitoring system and method based on electric field strength detection. Background Art

[0002] Key equipment in substations, such as transformers, circuit breakers, disconnectors, current transformers, voltage transformers, lightning arresters, busbars, switch cabinets, grounding devices, capacitor banks, load switches and fuses, have multiple parameters that can be used for equipment health status analysis in addition to temperature and appearance information during operation. These parameters include equipment location, adjacent equipment points, and electrical parameters (such as voltage, current, power factor, etc.).

[0003] Traditional remote inspection systems mainly use thermal imaging cameras to collect equipment appearance photos and infrared heat maps to monitor equipment status. However, this method has the following problems:

[0004] 1. Lag: When the problem is discovered, the equipment is already in a "damaged operation" state;

[0005] 2. The burden reduction effect is not obvious: operation and maintenance personnel still need to frequently deal with emergencies;

[0006] 3. Insufficient subdivision of abnormality types: Failure to classify equipment abnormalities in detail;

[0007] 4. Lack of research and analysis means: lack of effective tools and technical support.

[0008] In addition, the collection of electrical parameters requires the modification of the operating equipment and the installation of an external monitoring unit, which has two main problems:

[0009] 1. High voltage risk: Primary equipment carries high voltage, and newly mounted equipment may affect the normal operation of existing equipment, bringing risks to power production.

[0010] 2. Impact of shutdown: The transformation process requires the shutdown of equipment, which will lead to interruption of substation production and affect people’s production and life. Summary of the invention

[0011] The technical problem to be solved by the present invention is to provide a substation equipment health status monitoring system and method based on electric field strength detection, which can effectively improve inspection efficiency and accuracy, reduce human errors, realize early detection of equipment cracks, rust, looseness and high-temperature overheating defects, and greatly improve the substation operation and maintenance efficiency and safety.

[0012] The technical solution of the present invention is:

[0013] A substation equipment health status monitoring system based on electric field strength detection comprises: an electric field strength sensor, an image recognition module, an infrared imaging module, a data processing and analysis module, a wireless transmission module, an inspection robot and a health status detection platform, wherein the electric field strength sensor is arranged at the front end of the inspection robot, the image recognition module is arranged at the top of the inspection robot, the infrared imaging module is arranged at the side or bottom of the inspection robot, the electric field strength sensor, the image recognition module and the infrared imaging module are wirelessly connected to the data processing and analysis module via the wireless transmission module, the data processing and analysis module is wirelessly connected to the health status detection platform via the wireless transmission module, and the health status detection platform is wirelessly connected to the inspection robot.

[0014] Furthermore, the image recognition module includes a high-definition camera, which supports at least 1080p resolution, and the image recognition module has a built-in defect recognition algorithm based on deep learning.

[0015] Furthermore, the data processing and analysis module includes a microprocessor and has a built-in fault diagnosis algorithm based on machine learning.

[0016] Furthermore, the wireless transmission module supports the following communication modes: WAPI, WiFi, 4G and 5G communications.

[0017] Furthermore, the inspection robot is a humanoid robot or a wheeled robot or a quadruped robot dog, has a battery life of no less than 8 hours, supports remote control, and navigation methods include: visual navigation, lidar navigation, GPS positioning, inertial navigation and ultrasonic obstacle avoidance.

[0018] Furthermore, the deep learning-based defect recognition algorithm includes the following steps:

[0019] Step 1: Data preparation, including the following steps:

[0020] Step 1.1, data collection: collect image or signal data containing defective and non-defective samples;

[0021] Step 1.2: Data preprocessing, including:

[0022] Normalization: Normalize the input data. The formula is as follows:

[0023]

[0024] Among them, x represents the original data, μ represents the mean of the data, and σ represents the standard deviation of the data;

[0025] Step 1.3, data enhancement: use translation, rotation, and cropping to increase data diversity;

[0026] Step 2: Feature extraction: Automatically extract data features through deep learning convolutional neural networks, including:

[0027] Step 2.1, convolution operation:

[0028]

[0029] Among them, y i，j，k Represents the value of the output feature map, x i+m，j+n，c Represents the value of the input feature map, w m，n，c，k represents the convolution kernel, b k represents the bias term;

[0030] Step 2.2, ReLU activation function:

[0031] f(x)=max(0,x),

[0032] Among them, f(x) represents the output value of the activation function, and x represents the input value;

[0033] Step 2.3, pooling operation:

[0034]

[0035] Among them, y i，j represents the output value at position (i, j) after the pooling operation, x i+m，j+n Represents the pixel value or feature value at the position (i+m,j+n) of the pooling window in the input feature map, kernel represents the size of the pooling window, and max represents the maximum value of all pixel values ​​in the pooling window;

[0036] Step 3: Model construction and training, including:

[0037] Step 3.1, forward propagation:

[0038]

[0039] in, represents the model output, i.e., the predicted probability of each category, W l represents the weight matrix of the current layer, f l-1 represents the output features of the previous layer, b l represents the bias term of the current layer, softmax(z i ) indicates that the Softmax function converts the original output value into a probability distribution, z i represents the original output value of the i-th category, represents the normalization factor, C represents the number of categories;

[0040] Step 3.2, loss function:

[0041]

[0042] Among them, L represents the loss function, N represents the number of samples, C represents the number of categories, and y i，c represents the actual label, represents the predicted probability;

[0043] Step 3.3, back propagation: calculate the gradient through the chain rule and update the weights:

[0044] Among them, w (t) represents the weight at the tth iteration, w (t+1) represents the updated weight at the t+1th iteration, η represents the learning rate, Represents the gradient of the loss function L with respect to the weight w. The gradient indicates the direction of change of the loss function in the parameter space.

[0045] Step 4: Defect classification and prediction: Input the test data into the model to obtain the prediction results:

[0046] in, Represents the final predicted category of the model, z represents the original output value of the network output, and argmax represents the category with the highest return probability;

[0047] Step 5: Model evaluation, measure model performance through the following indicators:

[0048] Accuracy:

[0049]

[0050] Recall:

[0051]

[0052] F1 score:

[0053]

[0054] Among them, TP represents the number of positive classes correctly classified as positive classes by the model, TN represents the number of negative classes correctly classified as negative classes by the model, FP represents the number of negative classes misclassified as positive classes by the model, FN represents the number of positive classes misclassified as negative classes by the model, Precision represents the accuracy, that is, the proportion of samples correctly predicted as positive classes to all samples predicted as positive classes, F1 represents the harmonic mean between precision and recall, ranging from 0 to 1, and the closer the value is to 1, the better the model performance.

[0055] A monitoring method of a substation equipment health status monitoring system based on electric field strength detection comprises the following steps:

[0056] Step 1: Calibrate the equipment location and inspection route planning in the substation: According to the geographical layout and equipment distribution information of the substation, calibrate the specific coordinates of each key equipment and the walking path of the inspection route, and record the results in the inspection management system;

[0057] Step 2: Deploy an electric field strength sensor, an image recognition module, an infrared thermal imaging module, and a data processing and analysis module on the inspection robot: Integrate the electric field strength sensor, the visible light image recognition camera, and the infrared thermal imager into the inspection robot, and install the data processing and analysis module on the robot in advance, so as to complete preliminary data collection and processing on site;

[0058] Step 3: The inspection robot moves to the working point according to the set program: According to the route planned in step 1, the inspection robot moves to each target equipment point in turn; in the process, the odometer, laser radar or visual navigation is used to achieve autonomous positioning and obstacle avoidance;

[0059] Step 4: Collect the electric field strength data around the device through the electric field strength sensor: After reaching the vicinity of the specified device, the electric field strength sensor starts working to obtain the electric field strength value E(x, y) at each sampling point in the current space. Different points (x, y) correspond to different electric field strength measurement results, and finally a set of discrete electric field values ​​is obtained;

[0060] Step 5: Collect the appearance of the device through the image recognition module: Use a visible light camera to take pictures of the front and side views of the device for subsequent appearance defect identification, where the appearance defects include cracks, rust, and looseness;

[0061] Step 6: Collect the infrared thermal image of the device through the infrared thermal imaging module: Use the infrared camera to capture the thermal distribution map of the device surface for temperature anomaly detection, where the temperature anomaly includes: temperature rise caused by high temperature overheating and poor contact;

[0062] Step 7: Generate an electric field thermal map, including the following steps:

[0063] Step 7.1, establish the image coordinate system: take the upper left corner of the thermal image as the origin O(0,0), the right as the u axis, and the downward as the v axis; the corresponding sizes of unit pixels in the physical coordinate system are Δx and Δy respectively; let the center pixel of the location of the device under test on the thermal image be O 1 (u 0 ,v 0 );

[0064] Step 7.2, Pixel coordinates and physical distance conversion: The distance between the device and a point is L, and the calculation formula corresponding to the pixel coordinates (u, v) on the heat map is:

[0065]

[0066] When the electric field intensity sensor collects the electric field value E(x,y), it maps it to the pixel point corresponding to the distance;

[0067] Step 7.3, range normalization and pixel assignment: The range of the electric field sensor is [0,E max ], in order to display the electric field strength in an 8-bit image, linear normalization is used:

[0068]

[0069] Where V represents the pixel grayscale value, E(x,y) represents the measured electric field strength, and E max Indicates the upper limit of the electric field strength sensor range;

[0070] If the distance O in the image 1 (u 0 ,v 0 ) The same pixels correspond to the same measured electric field value, and the grayscale values ​​V of these pixels are also the same;

[0071] Step 7.4, Lagrange interpolation completion: For pixels that have not been measured, that is, areas where the electric field value is unknown, one-dimensional Lagrange interpolation is used for estimation. The interpolation formula is:

[0072]

[0073] Where P(x) represents the interpolation function, n represents the interpolation point index, x represents the interpolation variable, and x i Indicates the horizontal coordinate of the i-th known interpolation node, y i represents the ordinate of the i-th known interpolation node, L i (x) represents the i-th Lagrangian basis function, x j Represents the horizontal coordinate of the jth known interpolation node;

[0074] Step 7.5, forming an electric field thermal map: after the above mapping, normalization and interpolation, an electric field thermal map of size 65×65 is obtained, which uses grayscale or pseudo-color to display the electric field intensity distribution around the device;

[0075] Step 8: Convolutional neural network analysis of infrared images, including the following steps:

[0076] Step 8.1, forward propagation of convolutional neural network: take infrared thermal imaging image as input, and extract features through multi-layer convolution:

[0077] f (l) =σ(W (l) *f (l-1) +b (l) ),

[0078] Among them, f (l) represents the output feature of the lth layer, σ(·) represents the activation function, and W (l) represents the convolution kernel of layer l, b (l) represents the bias of the lth layer, f (l-1) Represents the output features of the l-1th layer, that is, the input of the lth layer;

[0079] Finally, the prediction results of temperature anomaly detection or defect level are given in the output layer;

[0080] Step 8.2, temperature comparison standard and defect level: classify the temperature into normal, slightly overheated and severely overheated, and record the network output as: z = [z 1 , z 2 , ..., z C ], z represents the prediction vector of the defect category, C represents the number of categories, and the probability of each category is obtained by Softmax:

[0081]

[0082] in, represents the model's predicted probability for the cth class, z c represents the original output of the model for the cth class, z k Represents the original output of the model for the kth class;

[0083] Define the cross entropy loss function for parameter optimization during training:

[0084]

[0085] in, Represents the classification loss value, y c represents the true label, Represents the predicted probability, predicted probability The closer to the true label y c , the corresponding cross entropy loss On the contrary, the greater the deviation, the higher the loss;

[0086] Or divide the intervals according to the actual temperature threshold:

[0087]

[0088] Among them, T 1 , T 2 represents the reference temperature threshold, T represents the actual temperature. If the actual temperature exceeds T 2 It is considered a serious defect;

[0089] Step 9. Transmit the analysis results to the health status detection platform through the wireless transmission module: upload the electric field thermogram, visible light image defect detection results and infrared temperature anomaly analysis result data completed on the inspection robot to the background health status detection platform through WAPI, 4G, 5G or Wi-Fi for comprehensive evaluation and recording;

[0090] Step 10. When an abnormal situation is found, an early warning message is issued to notify the operation and maintenance personnel to carry out maintenance: If the background health status detection platform detects that the temperature at a certain point exceeds the limit, the electric field distribution is abnormal, or the appearance defect is serious, an alarm is automatically triggered, and the abnormal equipment information and fault type information are pushed to the operation and maintenance personnel's mobile terminal or duty center to guide subsequent inspection and maintenance work.

[0091] Beneficial effects of the present invention:

[0092] 1. Improved inspection efficiency: The present invention integrates an electric field strength sensor, a visible light camera and an infrared thermal imager on the inspection robot to collect multiple data at one time, significantly shorten the inspection time, and reduce the frequency of manual inspections;

[0093] 2. Improved fault identification accuracy: The present invention uses multi-source data fusion of electric field distribution information, visible light images and infrared thermal images, combined with deep learning algorithms (convolutional neural networks), to accurately detect appearance and temperature anomalies, reducing missed reports and false alarm rates;

[0094] 3. Enhanced early warning capability: The present invention generates electric field thermal maps and analyzes equipment surface defects and temperature anomalies in real time, which can timely identify potential signs of failures, such as cracks, rust, looseness, and high temperature overheating, to prevent accidents from expanding;

[0095] 4. Reduction of operation and maintenance costs and manual workload: The inspection robot of the present invention can automatically perform inspections according to the planned route and transmit the results to the health status detection platform in real time, saving a lot of manpower and travel costs, and can also replace manual operations in harsh or dangerous environments;

[0096] 5. Improved safety and reliability: Through remote monitoring and early warning mechanisms, the present invention enables operation and maintenance personnel to be informed of abnormalities in the first place and take corresponding measures, thus avoiding personnel from working directly in high voltage and high temperature environments, thereby improving the overall safety of the system;

[0097] 6. Flexible deployment and scalability: The monitoring system of the present invention adopts wireless transmission design, supports multiple communication methods (WAPI, Wi-Fi, 4G, 5G), can be flexibly configured according to different substation scales and environments, and the algorithm module can be upgraded at any time. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1It is a structural schematic diagram of a substation equipment health status monitoring system based on electric field strength detection of the present invention.

[0099] Figure 2 It is a flow chart of a monitoring method of a substation equipment health status monitoring system based on electric field strength detection according to the present invention.

[0100] Figure 3 The present invention is a schematic diagram of the arrangement of electric field strength sensors of a substation equipment health status monitoring system based on electric field strength detection.

[0101] Figure 4 This is an electric field thermal example diagram of a substation equipment health status monitoring system based on electric field strength detection of the present invention.

[0102] Figure 5 This is an infrared thermal imaging example diagram of a substation equipment health status monitoring system based on electric field strength detection of the present invention. DETAILED DESCRIPTION

[0103] like Figure 1-5 As shown, a substation equipment health status monitoring system based on electric field strength detection includes: an electric field strength sensor, an image recognition module, an infrared imaging module, a data processing and analysis module, a wireless transmission module, an inspection robot and a health status detection platform, the electric field strength sensor is arranged at the front end of the inspection robot, the image recognition module is arranged at the top of the inspection robot, the infrared imaging module is arranged at the side or bottom of the inspection robot, the electric field strength sensor, the image recognition module and the infrared imaging module are wirelessly connected to the data processing and analysis module through the wireless transmission module, the data processing and analysis module is wirelessly connected to the health status detection platform through the wireless transmission module, and the health status detection platform is wirelessly connected to the inspection robot.

[0104] Preferably, the image recognition module includes a high-definition camera, which supports at least 1080p resolution, and the image recognition module has a built-in defect recognition algorithm based on deep learning.

[0105] Preferably, the data processing and analysis module includes a microprocessor and has a built-in fault diagnosis algorithm based on machine learning.

[0106] Preferably, the wireless transmission module supports the following communication modes: WAPI, WiFi, 4G and 5G communications.

[0107] Preferably, the inspection robot is a humanoid robot or a wheeled robot or a quadruped robot dog, has a battery life of not less than 8 hours, supports remote control, and navigation methods include: visual navigation, lidar navigation, GPS positioning, inertial navigation and ultrasonic obstacle avoidance.

[0108] Preferably, the deep learning-based defect recognition algorithm comprises the following steps:

[0109] Step 1: Data preparation, including the following steps:

[0110] Step 1.1, data collection: collect image or signal data containing defective and non-defective samples;

[0111] Step 1.2: Data preprocessing, including:

[0112] Normalization: Normalize the input data. The formula is as follows:

[0113]

[0114] Among them, x represents the original data, μ represents the mean of the data, and σ represents the standard deviation of the data;

[0115] Step 1.3, data enhancement: use translation, rotation, and cropping to increase data diversity;

[0116] Step 2: Feature extraction: Automatically extract data features through deep learning convolutional neural networks, including:

[0117] Step 2.1, convolution operation:

[0118]

[0119] Among them, y i，j，k Represents the value of the output feature map, x i+m，j+n，c Represents the value of the input feature map, w m，n，c，k represents the convolution kernel, b k represents the bias term;

[0120] Step 2.2, ReLU activation function:

[0121] f(x)=max(0,x),

[0122] Among them, f(x) represents the output value of the activation function, and x represents the input value;

[0123] Step 2.3, pooling operation:

[0124]

[0125] Among them, y i，j represents the output value at position (i, j) after the pooling operation, x i+m，j+n Represents the pixel value or feature value at the position (i+m,j+n) of the pooling window in the input feature map, kernel represents the size of the pooling window, and max represents the maximum value of all pixel values ​​in the pooling window;

[0126] Step 3: Model construction and training, including:

[0127] Step 3.1, forward propagation:

[0128]

[0129] in, represents the model output, i.e., the predicted probability of each category, W l represents the weight matrix of the current layer, f l-1 represents the output features of the previous layer, b l represents the bias term of the current layer, softmax(z i ) indicates that the Softmax function converts the original output value into a probability distribution, z i represents the original output value of the i-th category, represents the normalization factor, C represents the number of categories;

[0130] Step 3.2, loss function:

[0131]

[0132] Among them, L represents the loss function, N represents the number of samples, C represents the number of categories, and y i，c represents the actual label, represents the predicted probability;

[0133] Step 3.3, back propagation: calculate the gradient through the chain rule and update the weights:

[0134]

[0135] Among them, w (t) represents the weight at the tth iteration, w (t+1) represents the updated weight at the t+1th iteration, η represents the learning rate, Represents the gradient of the loss function L with respect to the weight w. The gradient indicates the direction of change of the loss function in the parameter space.

[0136] Step 4: Defect classification and prediction: Input the test data into the model to obtain the prediction results:

[0137] in, Represents the final predicted category of the model, z represents the original output value of the network output, and argmax represents the category with the highest return probability;

[0138] Step 5: Model evaluation, measure model performance through the following indicators:

[0139] Accuracy:

[0140]

[0141] Recall:

[0142]

[0143] F1 score:

[0144]

[0145] Among them, TP represents the number of positive classes correctly classified as positive classes by the model, TN represents the number of negative classes correctly classified as negative classes by the model, FP represents the number of negative classes misclassified as positive classes by the model, FN represents the number of positive classes misclassified as negative classes by the model, Precision represents the accuracy, that is, the proportion of samples correctly predicted as positive classes to all samples predicted as positive classes, F1 represents the harmonic mean between precision and recall, ranging from 0 to 1, and the closer the value is to 1, the better the model performance.

[0146] A monitoring method of a substation equipment health status monitoring system based on electric field strength detection comprises the following steps:

[0147] Step 1: Calibrate the equipment location and inspection route planning in the substation: According to the geographical layout and equipment distribution information of the substation, calibrate the specific coordinates of each key equipment and the walking path of the inspection route, and record the results in the inspection management system;

[0148] Step 2: Deploy an electric field strength sensor, an image recognition module, an infrared thermal imaging module, and a data processing and analysis module on the inspection robot: Integrate the electric field strength sensor, the visible light image recognition camera, and the infrared thermal imager into the inspection robot, and install the data processing and analysis module on the robot in advance, so as to complete preliminary data collection and processing on site;

[0149] Step 3: The inspection robot moves to the working point according to the set program: According to the route planned in step 1, the inspection robot moves to each target equipment point in turn; in the process, the odometer, laser radar or visual navigation is used to achieve autonomous positioning and obstacle avoidance (the inspection robot moves from the farthest point to the nearest point according to the set inspection route, one point at a time, and performs electric field strength detection and infrared thermal imaging at each point);

[0150] Step 4: Collect the electric field strength data around the device through the electric field strength sensor: After reaching the vicinity of the specified device, the electric field strength sensor starts working to obtain the electric field strength value E(x, y) at each sampling point in the current space. Different points (x, y) correspond to different electric field strength measurement results, and finally a set of discrete electric field values ​​is obtained;

[0151] Step 5: Collect the appearance of the device through the image recognition module: Use a visible light camera to take pictures of the front and side views of the device for subsequent appearance defect identification, where the appearance defects include cracks, rust, and looseness;

[0152] Step 6: Collect the infrared thermal image of the device through the infrared thermal imaging module: Use the infrared camera to capture the thermal distribution map of the device surface for temperature anomaly detection, where the temperature anomaly includes: temperature rise caused by high temperature overheating and poor contact;

[0153] Step 7: Generate an electric field thermal map, including the following steps:

[0154] Step 7.1, establish the image coordinate system: take the upper left corner of the heat map image as the origin O(0,0), the right is the u axis, and the downward is the v axis; the corresponding sizes of unit pixels in the physical coordinate system (meters / pixel) are Δx and Δy respectively; let the center pixel of the location of the device under test on the heat map be O 1 (u 0 ,v 0 );

[0155] Step 7.2, Pixel coordinates and physical distance conversion: The distance between the device and a point is L, and the calculation formula corresponding to the pixel coordinates (u, v) on the heat map is:

[0156]

[0157] When the electric field intensity sensor collects the electric field value E(x,y), it maps it to the pixel point corresponding to the distance;

[0158] Step 7.3, range normalization and pixel assignment: The range of the electric field sensor is [0,E max ], in order to display the electric field strength in an 8-bit image, linear normalization is used:

[0159]

[0160] Where V represents the pixel grayscale value (0 to 255), E(x, y) represents the measured electric field strength, and E max Indicates the upper limit of the electric field strength sensor range;

[0161] If the distance O in the image 1 (u 0 ,v 0 ) The same pixels correspond to the same measured electric field value, and the grayscale values ​​V of these pixels are also the same;

[0162] Step 7.4, Lagrange interpolation completion: For pixels that have not been measured, that is, areas where the electric field value is unknown, one-dimensional Lagrange interpolation is used for estimation. The interpolation formula is:

[0163]

[0164] Where P(x) represents the interpolation function, n represents the interpolation point index, x represents the interpolation variable, and x i Indicates the horizontal coordinate of the i-th known interpolation node, y i represents the ordinate of the i-th known interpolation node, L i (x) represents the i-th Lagrangian basis function, x j Represents the horizontal coordinate of the jth known interpolation node;

[0165] Step 7.5, forming an electric field thermal map: after the above mapping, normalization and interpolation, an electric field thermal map of size 65×65 is obtained, which uses grayscale or pseudo-color to display the electric field intensity distribution around the device;

[0166] Step 8: Convolutional neural network analysis of infrared images, including the following steps:

[0167] Step 8.1, forward propagation of convolutional neural network: take infrared thermal imaging image as input, and extract features through multi-layer convolution:

[0168] f (l) =σ(W (l) *f (l-1) +b (l) ),

[0169] Among them, f (l) represents the output feature of the lth layer, σ(·) represents the activation function (such as ReLU, Sigmoid, etc.), W (l) represents the convolution kernel of layer l, b (l) represents the bias of the lth layer, f (l-1) Represents the output features of the l-1th layer, that is, the input of the lth layer;

[0170] Finally, the prediction results of temperature anomaly detection or defect level are given in the output layer;

[0171] Step 8.2, temperature comparison standard and defect level: classify the temperature into normal, slightly overheated and severely overheated, and record the network output as: z = [z 1 , z 2 , ..., z C ], z represents the prediction vector of the defect category, C represents the number of categories, and the probability of each category is obtained by Softmax:

[0172]

[0173] in, represents the model's predicted probability for the cth class, z crepresents the original output of the model for the cth class, z k Represents the original output of the model for the kth class;

[0174] Define the cross entropy loss function for parameter optimization during training:

[0175]

[0176] in, Represents the classification loss value, y c represents the true label, Represents the predicted probability, predicted probability The closer to the true label y c , the corresponding cross entropy loss On the contrary, the greater the deviation, the higher the loss;

[0177] Or divide the intervals according to the actual temperature threshold:

[0178]

[0179] Among them, T 1 , T 2 represents the reference temperature threshold, T represents the actual temperature. If the actual temperature exceeds T 2 It is considered a serious defect;

[0180] Step 9. Transmit the analysis results to the health status detection platform through the wireless transmission module: upload the electric field thermogram, visible light image defect detection results and infrared temperature anomaly analysis result data completed on the inspection robot to the background health status detection platform through WAPI, 4G, 5G or Wi-Fi for comprehensive evaluation and recording;

[0181] Step 10. When an abnormal situation is found, an early warning message is issued to notify the operation and maintenance personnel to carry out maintenance: If the background health status detection platform detects that the temperature at a certain point exceeds the limit, the electric field distribution is abnormal, or the appearance defect is serious, an alarm is automatically triggered, and the abnormal equipment information and fault type information are pushed to the operation and maintenance personnel's mobile terminal or duty center to guide subsequent inspection and maintenance work.

[0182] The basic principles, main features and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which shall fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A substation equipment health status monitoring system based on electric field strength detection, characterized in that: include: An electric field strength sensor, an image recognition module, an infrared imaging module, a data processing and analysis module, a wireless transmission module, an inspection robot and a health status detection platform, wherein the electric field strength sensor is arranged at the front end of the inspection robot, the image recognition module is arranged at the top of the inspection robot, the infrared imaging module is arranged at the side or bottom of the inspection robot, the electric field strength sensor, the image recognition module and the infrared imaging module are wirelessly connected to the data processing and analysis module via the wireless transmission module, the data processing and analysis module is wirelessly connected to the health status detection platform via the wireless transmission module, and the health status detection platform is wirelessly connected to the inspection robot.

2. A substation equipment health status monitoring system based on electric field strength detection according to claim 1, characterized in that: The image recognition module includes a high-definition camera that supports at least 1080p resolution, and the image recognition module has a built-in defect recognition algorithm based on deep learning.

3. A substation equipment health status monitoring system based on electric field strength detection according to claim 1, characterized in that: The data processing and analysis module includes a microprocessor and a built-in fault diagnosis algorithm based on machine learning.

4. A substation equipment health status monitoring system based on electric field strength detection according to claim 1, characterized in that: The wireless transmission module supports the following communication modes: WAPI, WiFi, 4G and 5G communications.

5. A substation equipment health status monitoring system based on electric field strength detection according to claim 1, characterized in that: The inspection robot is a humanoid robot or a wheeled robot or a quadruped robot dog, has a battery life of no less than 8 hours, supports remote control, and navigation methods include: visual navigation, lidar navigation, GPS positioning, inertial navigation and ultrasonic obstacle avoidance.

6. A substation equipment health status monitoring system based on electric field strength detection according to claim 2, characterized in that: The deep learning-based defect recognition algorithm includes the following steps: Step 1: Data preparation, including the following steps: Step 1.1, data collection: collect image or signal data containing defective and non-defective samples; Step 1.2: Data preprocessing, including: Normalization: Normalize the input data. The formula is as follows: Among them, x represents the original data, μ represents the mean of the data, and σ represents the standard deviation of the data; Step 1.3, data enhancement: use translation, rotation, and cropping to increase data diversity; Step 2: Feature extraction: Automatically extract data features through deep learning convolutional neural networks, including: Step 2.1, convolution operation: Among them, y i，j，k Represents the value of the output feature map, x i+m，j+n，c Represents the value of the input feature map, w m，n，c，k represents the convolution kernel, b k represents the bias term; Step 2.2, ReLU activation function: f(x)=max(0,x), Among them, f(x) represents the output value of the activation function, and x represents the input value; Step 2.3, pooling operation: Among them, y i，j represents the output value at position (i, j) after the pooling operation, x i+m，j+n Represents the pixel value or feature value at the position (i+m,j+n) of the pooling window in the input feature map, kernel represents the size of the pooling window, and max represents the maximum value of all pixel values ​​in the pooling window; Step 3: Model construction and training, including: Step 3.1, forward propagation: in, represents the model output, i.e., the predicted probability of each category, W l represents the weight matrix of the current layer, f l-1 represents the output features of the previous layer, b l represents the bias term of the current layer, softmax(z i ) indicates that the Softmax function converts the original output value into a probability distribution, z i represents the original output value of the i-th category, represents the normalization factor, C represents the number of categories; Step 3.2, loss function: Among them, L represents the loss function, N represents the number of samples, C represents the number of categories, and y i，c represents the actual label, represents the predicted probability; Step 3.3, back propagation: calculate the gradient through the chain rule and update the weights: Among them, w (t) represents the weight at the tth iteration, w (t+1) represents the updated weight at the t+1th iteration, η represents the learning rate, Represents the gradient of the loss function L with respect to the weight w. The gradient indicates the direction of change of the loss function in the parameter space. Step 4: Defect classification and prediction: Input the test data into the model to obtain the prediction results: in, Represents the final predicted category of the model, z represents the original output value of the network output, and argmax represents the category with the highest return probability; Step 5: Model evaluation, measure model performance through the following indicators: Accuracy: Recall: F1 score: Among them, TP represents the number of positive classes correctly classified as positive classes by the model, TN represents the number of negative classes correctly classified as negative classes by the model, FP represents the number of negative classes misclassified as positive classes by the model, FN represents the number of positive classes misclassified as negative classes by the model, Precision represents the accuracy, that is, the proportion of samples correctly predicted as positive classes to all samples predicted as positive classes, F1 represents the harmonic mean between precision and recall, ranging from 0 to 1, and the closer the value is to 1, the better the model performance.

7. A monitoring method for a substation equipment health status monitoring system based on electric field strength detection according to claim 1, characterized in that: The following steps are involved: Step 1: Calibrate the equipment location and inspection route planning in the substation: According to the geographical layout and equipment distribution information of the substation, calibrate the specific coordinates of each key equipment and the walking path of the inspection route, and record the results in the inspection management system; Step 2: Deploy an electric field strength sensor, an image recognition module, an infrared thermal imaging module, and a data processing and analysis module on the inspection robot: Integrate the electric field strength sensor, the visible light image recognition camera, and the infrared thermal imager into the inspection robot, and install the data processing and analysis module on the robot in advance, so as to complete preliminary data collection and processing on site; Step 3: The inspection robot moves to the working point according to the set program: According to the route planned in step 1, the inspection robot moves to each target equipment point in turn; in the process, the odometer, laser radar or visual navigation is used to achieve autonomous positioning and obstacle avoidance; Step 4: Collect the electric field strength data around the device through the electric field strength sensor: After reaching the vicinity of the specified device, the electric field strength sensor starts working to obtain the electric field strength value E(x, y) at each sampling point in the current space. Different points (x, y) correspond to different electric field strength measurement results, and finally a set of discrete electric field values ​​is obtained; Step 5: Collect the appearance of the device through the image recognition module: Use a visible light camera to take pictures of the front and side views of the device for subsequent appearance defect identification, where the appearance defects include cracks, rust, and looseness; Step 6: Collect the infrared thermal image of the device through the infrared thermal imaging module: Use the infrared camera to capture the thermal distribution map of the device surface for temperature anomaly detection, where the temperature anomaly includes: temperature rise caused by high temperature overheating and poor contact; Step 7: Generate an electric field thermal map, including the following steps: Step 7.1, establish the image coordinate system: take the upper left corner of the thermal image as the origin O(0,0), the right as the u axis, and the downward as the v axis; the corresponding sizes of unit pixels in the physical coordinate system are Δx and Δy respectively; let the center pixel of the position of the device under test on the thermal image be O1(u0,v0); Step 7.2, Pixel coordinates and physical distance conversion: The distance between the device and a point is L, and the calculation formula corresponding to the pixel coordinates (u, v) on the heat map is: When the electric field intensity sensor collects the electric field value E(x,y), it maps it to the pixel point corresponding to the distance; Step 7.3, range normalization and pixel assignment: The range of the electric field sensor is [0,E max ], in order to display the electric field strength in an 8-bit image, linear normalization is used: Where V represents the pixel grayscale value, E(x,y) represents the measured electric field strength, and E max Indicates the upper limit of the range of the electric field strength sensor; If some pixels in the image with the same distance O1(u0,v0) correspond to the same measured electric field value, then the grayscale values ​​V of these pixels are also the same; Step 7.4, Lagrange interpolation completion: For pixels that have not been measured, that is, areas where the electric field value is unknown, one-dimensional Lagrange interpolation is used for estimation. The interpolation formula is: Where P(x) represents the interpolation function, n represents the interpolation point index, x represents the interpolation variable, and x i Indicates the horizontal coordinate of the i-th known interpolation node, y i represents the ordinate of the i-th known interpolation node, L i (x) represents the i-th Lagrangian basis function, x j Represents the horizontal coordinate of the jth known interpolation node; Step 7.5, forming an electric field thermal map: after the above mapping, normalization and interpolation, an electric field thermal map of size 65×65 is obtained, which uses grayscale or pseudo-color to display the electric field intensity distribution around the device; Step 8: Convolutional neural network analysis of infrared images, including the following steps: Step 8.1, forward propagation of convolutional neural network: take infrared thermal imaging image as input, and extract features through multi-layer convolution: f (l) =σ(W (l) *f (l-1) +b (l) ), Among them, f (l) represents the output feature of the lth layer, σ(·) represents the activation function, and W (l) represents the convolution kernel of layer l, b (l) represents the bias of the lth layer, f (l-1) Represents the output features of the l-1th layer, that is, the input of the lth layer; Finally, the prediction results of temperature anomaly detection or defect level are given in the output layer; Step 8.2, temperature comparison standard and defect level: classify the temperature into normal, slightly overheated and severely overheated, and record the network output as: z = [z1, z2, ..., z C ], z represents the prediction vector of the defect category, C represents the number of categories, and the probability of each category is obtained by Softmax: in, represents the model's predicted probability for the cth class, z c represents the original output of the model for the cth class, z k Represents the original output of the model for the kth class; Define the cross entropy loss function for parameter optimization during training: in, Represents the classification loss value, y c represents the true label, Represents the predicted probability, predicted probability The closer to the true label y c , the corresponding cross entropy loss On the contrary, the greater the deviation, the higher the loss; Or divide the intervals according to the actual temperature threshold: Wherein, T1 and T2 represent reference temperature thresholds, T represents the actual temperature, and if the actual temperature exceeds T2, it is judged as a serious defect; Step 9. Transmit the analysis results to the health status detection platform through the wireless transmission module: upload the electric field thermogram, visible light image defect detection results and infrared temperature anomaly analysis result data completed on the inspection robot to the background health status detection platform through WAPI, 4G, 5G or Wi-Fi for comprehensive evaluation and recording; Step 10. When an abnormal situation is found, an early warning message is issued to notify the operation and maintenance personnel to carry out maintenance: If the background health status detection platform detects that the temperature at a certain point exceeds the limit, the electric field distribution is abnormal, or the appearance defect is serious, an alarm is automatically triggered, and the abnormal equipment information and fault type information are pushed to the operation and maintenance personnel's mobile terminal or duty center to guide subsequent inspection and maintenance work.

Images