Pressing plate state identification method based on heuristic manta ray foraging optimization and adaptive extreme learning machine

Abstract

The invention relates to a pressing plate state identification method based on heuristic manta ray foraging optimization and an adaptive extreme learning machine, and the method comprises the steps: firstly obtaining a pressing plate image from an on-site pressing plate data set, carrying out the processing of the pressing plate image, and carrying out the segmentation of the pressing plate image, so as to extract an area related to the pressing plate image; in the process, the accurate area of the pressing plate is segmented, and the state and color recognition operation of the pressing plate is carried out. Thereafter, the segmented portion about the platen image is used for feature extraction for the purpose of extracting a plurality of features about the platen state to improve the subsequent identification process. According to the invention, the heuristic manta ray foraging optimization method is used to optimize the number of the selected features, so that the accuracy of the recognition model is improved, the complexity is reduced, and the training and recognition time of the model can be shortened. And finally, identifying the state of the pressing plate by using a self-adaptive extreme learning machine method and the determined feature number. The method provided by the invention can improve the accuracy of model identification.

Description

Technical Field

[0001] The invention relates to the technical field of electric power maintenance, and in particular to a pressure plate state recognition method based on heuristic manta ray foraging optimization and adaptive extreme learning machine. Background Art

[0002] The main function of the pressure plate in the power system is to provide a reliable electrical connection and transmit the current or voltage signal of the electrical equipment in the power system to the protection device. Therefore, the detection of the pressure plate state is of great significance to ensure the safe and reliable operation of the power system. At present, on-site personnel often manually check, read and identify the state of the substation pressure plate, which has problems such as low efficiency and low accuracy, which seriously limits the development of substations in the intelligent direction. In order to realize the online monitoring of the pressure plate, researchers have proposed several pressure plate state detection methods, such as frequency analysis method and artificial neural network method. The frequency analysis method usually uses a vibration transducer to capture the vibration signal of the device and extract the typical characteristics of the vibration signal, such as frequency and amplitude, to determine the state of the device. The artificial neural network method can learn from a large amount of data and build a model to detect the state of the equipment, which has advantages in processing nonlinear and unsteady signals. Common artificial intelligence methods include support vector machines, artificial neural networks and decision trees. However, due to the complexity of the on-site environment, the pressure plate signal contains a lot of noise, and the recognition effect of the above methods is not ideal. Summary of the invention

[0003] In view of this, the object of the present invention is to provide a pressure plate state recognition method based on heuristic manta ray foraging optimization and adaptive extreme learning machine to improve the accuracy of model recognition.

[0004] To achieve the above object, the present invention adopts the following technical solution: a pressure plate state recognition method based on heuristic manta ray foraging optimization and adaptive extreme learning machine, comprising the following steps:

[0005] Step 1: Use the hue saturation value color space model to detect the color of the pressure plate, and then calculate the threshold through the histogram model to distinguish the pressure plate and non-pressure plate parts to achieve target segmentation;

[0006] Step 2: Extract the image features of the platen state from the segmented image using a multi-faceted feature extraction model;

[0007] Step 3: After extracting the features, the optimal number of features is selected according to the optimal fitness value using the heuristic manta ray foraging optimization method;

[0008] Step 4: Based on the optimal feature number of the pressure plate image and the adaptive extreme learning machine method, a pressure plate state recognition model is constructed.

[0009] In a preferred embodiment, the step 1 is specifically as follows: first, the graphic segmentation part aims to accurately segment the pressure plate area from a given input image based on the pressure plate color detection and graphic background suppression part, and use the hue saturation value (HSV) color space model to detect the pressure plate color, and obtain illumination invariance by eliminating the value of V and using the values ​​of H and S; then, the histogram model is used to calculate the threshold to distinguish between the pressure plate and non-pressure plate parts to achieve target segmentation.

[0010] In a preferred embodiment, the step 2 is specifically as follows: the aspect feature extraction model mainly includes the features of the major axis-minor axis, centroid, range, direction, solidity, eccentricity, and salient poles, and its related feature indicators for the image are defined as follows:

[0011]

[0012] Among them, M A and M B They represent the lengths of the major and minor axes, respectively, and L(a,b) and S(a,b) represent the values ​​of the major and minor axes, respectively;

[0013]

[0014]

[0015] Where a1 and a2 represent the horizontal coordinates of any two points in the image, b1 and b2 represent the vertical coordinates of any two points, and Cen S represents the center of mass, Ex S Indicates the range, T P Defines the total number of pixels in the segmented image, BT P Represents the total number of pixels in the bounded area;

[0016] Con S =∑Co I (a,b) (4)

[0017]

[0018] Among them, Con S represents a convex region, Co I is a convex image, So S Define solidity, H A is the hull area of ​​the shape, Eccen S Eccentricity, Fo D is the focal distance, se m It is a semi-spindle.

[0019] In a preferred embodiment, in step 3, the behavior and position of the manta ray can be changed according to its food source, as shown below:

[0020]

[0021] Among them, Po i,j is the position of the manta ray, i represents the horizontal coordinate of the bat's position, j represents the vertical coordinate, and x(:) represents outputting the data in the matrix as a column. is a random position, B L，j represents the best position of j, K is the number of iterations, Nu v Define the number of variables, Nu p Represents the population size; after finding the food source, the manta ray's new position is updated and adjusted to the optimal position, as shown below:

[0022]

[0023]

[0024] Where CP represents the control system of heuristic manta ray foraging optimization, R H Represents a random value; on this basis, the best optimal solution is estimated to improve the overall accuracy of classification, and the fitness of the i-th solution is calculated as follows:

[0025]

[0026] Fit i represents the ith solution, π1 and π2 are equilibrium parameters, Acc i Define the precision of the ith solution, OS i represents the size of the ith solution, S F is a feature set.

[0027] In a preferred embodiment, in step 4, the adaptive extreme learning machine includes an input layer, a hidden layer and an output layer, wherein the input weight is randomly defined and obtained by the input layer, the hidden layer processes the information with deviation according to the input weight, and the output layer generates the final classification result; the adaptive extreme learning machine randomly selects the deviation and the input weight, constructs the output matrix of the hidden layer, and estimates the weight of the output layer using the Moore-Penrose generalization model;

[0028] The numerical model of the adaptive extreme learning machine is:

[0029] Hβ=O (12)

[0030]

[0031] Among them, H is the hidden layer output matrix; β is the output layer weight; O is the model calculation output; w i =[w i1 ,w i2 ,…,w in](i=1,2,…,l) is the connection weight between the input layer and the hidden layer neurons, l is the number of hidden layer neurons, and n is the length of the sample attribute; x i =[x i1 , x i2 , …, x in ]T(i=1,2,…,N) is the input sample, there are N groups in total; b i (i=1,2,…,l) is the bias value of the hidden layer neuron; g(·) is the activation function with infinite differentiability; given the activation function g(·) and the training sample set (x i ,y i )(i=1,2,…,N), randomly generate a set of weights w and bias values ​​b, and by solving the least squares solution of equation (14), we can uniquely determine a β=H + Y satisfies the condition ‖Hβ-Y‖≤ε, where ε is a given threshold;

[0032]

[0033] Based on the output matrix H of the hidden layer and combined with the input of the adaptive extreme learning machine, the state recognition label of the pressing plate is obtained, thereby completing the state recognition of the pressing plate.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. Using a multi-faceted feature extraction model to extract image features of the platen state from the segmented image helps improve the recognition accuracy of subsequent models;

[0036] 2. After extracting the features, the heuristic manta ray foraging optimization method is used to select the optimal number of features according to the optimal fitness value, which can effectively reduce the number of model iterations and improve the convergence speed to find the global optimal solution.

[0037] 3. An adaptive extreme learning machine method is used to identify the state of the pressure plate. This method has high recognition efficiency in high-dimensional space, short training and recognition time, and high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0040] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0041] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0042] A platen state recognition method based on heuristic manta ray foraging optimization and adaptive extreme learning machine, reference Figure 1 ,First, the graphic segmentation part aims to accurately segment the pressure plate area from the given input picture based on the pressure plate color detection and the ,graphic background suppression part. The hue saturation value (HSV) color space model is used for pressure plate color ,detection. The illumination invariance is obtained by eliminating the value of V and using the values ​​of H and S. Then, the threshold is ,calculated using the histogram model to distinguish the pressure plate and the non-pressure plate parts to ,achieve the target segmentation.

[0043] Secondly, the image features of the platen state are extracted from the segmented image using a multi-aspect feature extraction model, which helps to improve the recognition accuracy of the subsequent model. The aspect feature extraction model mainly includes the features of the major axis-minor axis, center of mass, range, direction, solidity, eccentricity, and salient poles. The relevant feature indicators of the image are defined as follows:

[0044]

[0045] Among them, M A and M B They represent the length of the major axis and the minor axis respectively, L(a,b) and S(a,b) are the values ​​of the major axis and the minor axis respectively.

[0046]

[0047] Where a1 and a2 represent the horizontal coordinates of any two points in the image, b1 and b2 represent the vertical coordinates of any two points, and Cen S represents the center of mass, Ex S Indicates the range, T P Defines the total number of pixels in the segmented image, BT P Represents the total number of pixels within the bounded area.

[0048] Con S =∑Co I (a,b) (4)

[0049]

[0050] Among them, Con S represents a convex region, CoI is a convex image, So S Defines solidity, H A is the hull area of ​​the shape, Eccen S Eccentricity, Fo D is the focal distance, se m It is a semi-spindle.

[0051] Furthermore, after extracting the features, the heuristic manta foraging optimization method is used to select the optimal number of features according to the optimal fitness value, which can effectively reduce the number of model iterations while improving the convergence speed to find the global optimal solution. Since the performance of the pressure plate state recognition classifier is highly dependent on the number of suitable features used to train the model, the heuristic manta foraging optimization mechanism is used to select the best features to train the classifier model. In this process, the behavior and position of the manta ray can change according to its food source, as shown below:

[0052]

[0053] Among them, Po i,j is the position of the manta ray, i represents the horizontal coordinate of the bat's position, j represents the vertical coordinate, and x(:) represents outputting the data in the matrix as a column. is a random position, B L represents the best position of j, K is the number of iterations, Nu v Defines the number of variables, Nu p Represents the number of populations. After finding the food source, the manta ray's new position is updated and adjusted to the optimal position, as shown below:

[0054]

[0055]

[0056] Where CP represents the control system of heuristic manta ray foraging optimization, R H Represents a random value. On this basis, the best optimal solution is estimated to improve the overall accuracy of classification, and the fitness of the i-th solution is calculated as follows:

[0057]

[0058] Fit i represents the ith solution, π1 and π2 are equilibrium parameters, Acc i Defines the accuracy of the ith solution, OS i represents the size of the ith solution, S F is a feature set.

[0059] Furthermore, an adaptive extreme learning machine method is used to identify the state of the pressure plate. The advantages of this method in pressure plate identification are: high recognition efficiency in high-dimensional space, short training and recognition time, and high recognition accuracy. This method includes an input layer, a hidden layer, and an output layer, wherein the input weights are randomly defined and obtained by the input layer, the hidden layer processes the biased information according to the input weights, and the output layer produces the final classification result. In the patent of this invention, the adaptive extreme learning machine performs the following processing: 1) randomly selects biases and input weights; 2) constructs the output matrix of the hidden layer; 3) uses the Moore-Penrose generalization model to estimate the weights of the output layer.

[0060] The numerical model of the adaptive extreme learning machine is:

[0061] Hβ=O (12)

[0062]

[0063] Among them, H is the hidden layer output matrix; β is the output layer weight; O is the model calculation output; w i =[w i1 ,w i2 ,…,w in ](i=1,2,…,l) is the connection weight between the input layer and the hidden layer neurons, l is the number of hidden layer neurons, and n is the length of the sample attribute; x i =[x i1 , x i2 , …, x in ] T (i=1,2,…,N) is the input sample, there are N groups in total; b i (i=1,2,…,l) is the bias value of the hidden layer neuron; g(·) is the activation function with infinite differentiability. Given the activation function g(·) and the training sample set (x i ,y i )(i=1,2,…,N), randomly generate a set of weights w and bias values ​​b, and by solving the least squares solution of equation (14), we can uniquely determine a β=H + Y satisfies the condition ‖Hβ-Y‖≤ε, where ε is a given threshold.

[0064]

[0065] Based on the output matrix H of the hidden layer and combined with the input of the adaptive extreme learning machine, the state recognition label of the pressing plate is obtained, thereby completing the state recognition of the pressing plate.

Claims

1. A method for identifying a pressure plate state based on heuristic manta ray foraging optimization and adaptive extreme learning machine, characterized in that: The following steps are involved: Step 1: Use the hue saturation value color space model to detect the color of the pressure plate, and then calculate the threshold through the histogram model to distinguish the pressure plate and non-pressure plate parts to achieve target segmentation; Step 2: Extract the image features of the platen state from the segmented image using a multi-faceted feature extraction model; Step 3: After extracting the features, the optimal number of features is selected according to the optimal fitness value using the heuristic manta ray foraging optimization method; Step 4: Based on the optimal feature number of the pressure plate image and the adaptive extreme learning machine method, a pressure plate state recognition model is constructed.

2. The method for identifying a pressure plate state based on heuristic manta ray foraging optimization and adaptive extreme learning machine according to claim 1, characterized in that: The step 1 is specifically as follows: first, the graphic segmentation part aims to accurately segment the pressure plate area from a given input image based on the pressure plate color detection and graphic background suppression part, and use the hue saturation value (HSV) color space model to detect the pressure plate color, and obtain illumination invariance by eliminating the value of V and using the values ​​of H and S; Then, the histogram model is used to calculate the threshold to distinguish the pressed plate and non-pressed plate parts to achieve target segmentation.

3. The method for identifying a pressure plate state based on heuristic manta ray foraging optimization and adaptive extreme learning machine according to claim 1, characterized in that: The step 2 is specifically as follows: the aspect feature extraction model mainly includes the features of the major axis-minor axis, centroid, range, direction, solidity, eccentricity, and salient poles, and its related feature indicators for the image are defined as follows: Among them, M A and M B They represent the lengths of the major and minor axes, respectively, and L(a,b) and S(a,b) represent the values ​​of the major and minor axes, respectively; Where a1 and a2 represent the horizontal coordinates of any two points in the image, b1 and b2 represent the vertical coordinates of any two points, and Cen S represents the center of mass, Ex S Indicates the range, T P Defines the total number of pixels in the segmented image, BT P Represents the total number of pixels in the bounded area; Con S =∑Co I (a,b) (4) Among them, Con S represents a convex region, Co I is a convex image, So S Define solidity, H A is the hull area of ​​the shape, Eccen S Eccentricity, Fo D is the focal distance, se m It is a semi-spindle.

4. The method for identifying a pressure plate state based on heuristic manta ray foraging optimization and adaptive extreme learning machine according to claim 1, characterized in that: In step 3, the manta ray's behavior and position can change depending on its food source, as follows: Among them, Po i,j is the position of the manta ray, i represents the horizontal coordinate of the bat's position, j represents the vertical coordinate, and x(:) represents outputting the data in the matrix as a column. is a random position, B L，j represents the best position of j, K is the number of iterations, Nu v Define the number of variables, Nu p Represents the population size; after finding the food source, the manta ray's new position is updated and adjusted to the optimal position, as shown below: Where CP represents the control system of heuristic manta ray foraging optimization, R H Represents a random value; on this basis, the best optimal solution is estimated to improve the overall accuracy of classification, and the fitness of the i-th solution is calculated as follows: Fit i represents the ith solution, π1 and π2 are equilibrium parameters, Acc i Define the precision of the ith solution, OS i represents the size of the i-th solution, S F is a feature set.

5. The method for identifying a pressure plate state based on heuristic manta ray foraging optimization and adaptive extreme learning machine according to claim 1, characterized in that: In step 4, the adaptive extreme learning machine includes an input layer, a hidden layer and an output layer, wherein the input weight is randomly defined and obtained by the input layer, the hidden layer processes the information with deviation according to the input weight, and the output layer generates the final classification result; The adaptive extreme learning machine randomly selects biases and input weights, constructs the output matrix of the hidden layer, and estimates the weights of the output layer using the Moore-Penrose generalization model; The numerical model of the adaptive extreme learning machine is: Hβ=O (12) Among them, H is the hidden layer output matrix; β is the output layer weight; O is the model calculation output; w i =[w i1 ,w i2 ,…,w in ](i=1,2,…,l) is the connection weight between the input layer and the hidden layer neurons, l is the number of hidden layer neurons, and n is the length of the sample attribute; x i =[x i1 , x i2 , …, x in ]T(i=1,2,…,N) is the input sample, there are N groups in total; b i (i=1,2,…,l) is the bias value of the hidden layer neuron; g(·) is the activation function with infinite differentiability; given the activation function g(·) and the training sample set (x i ,y i )(i=1,2,…,N), randomly generate a set of weights w and bias values ​​b, and by solving the least squares solution of equation (14), we can uniquely determine a β=H + Y satisfies the condition ‖Hβ-Y‖≤ε, where ε is a given threshold; Based on the output matrix H of the hidden layer and combined with the input of the adaptive extreme learning machine, the state recognition label of the pressing plate is obtained, thereby completing the state recognition of the pressing plate.

Images