Improved electric connector contact resistance prediction method based on ant lion algorithm

Abstract

The invention discloses an improved electric connector contact resistance prediction method based on an ant lion algorithm, and the prediction method is based on a classic BP neural network, and introduces an optimization algorithm based on the ant lion predation behavior in the nature. According to the contact resistance prediction method under the condition of fretting wear of the electric connector, a BP neural network model (ALO-BP neural network model for short) is optimized by utilizing the global search capability of an ant lion algorithm. Comprising the following steps: initializing a population, introducing random walk of ants to realize global search, and ensuring the diversity of the population and the optimization performance of an algorithm through a roulette strategy and an elitist strategy; comparing fitness and elite ant lion positions; the fitness is calculated again, and the positions of the ants and the ant lions are updated; judging whether a constraint condition or the maximum number of iterations is met or not, and giving a weight and a threshold value to the BP neural network by the global optimal value under the condition that the constraint is met; and optimizing the BP neural network to obtain an optimal model, and performing reliability prediction on the contact resistance value. According to the invention, the optimization performance, the operation efficiency and the prediction precision of the BP neural network can be improved.

Description

Technical Field

[0001] The present invention designs a method for predicting the contact resistance of an electric connector, and specifically relates to an optimization algorithm based on the predation behavior of ant lions in nature, which utilizes the global search capability of the ant lion algorithm to optimize the BP neural network model (abbreviated as ALO-BP neural network model) to predict the contact resistance of an electric connector under micro-wear conditions. Background Art

[0002] Electrical connectors are basic electromechanical components that realize signal transmission and electrical connection between devices. They are widely used in military equipment, aerospace and other fields, and play the role of connecting circuits and transmitting electrical signals. These functions are achieved through the mutual contact between the pins and sockets (i.e., contacts) of the electrical connector. With the continuous development of technology, the requirements for the working environment of electrical connectors are getting higher and higher, and the factors affecting their life include vibration, salt spray, ultraviolet rays, extreme temperatures, etc. Among them, temperature is one of the main environmental factors. Under temperature cycling stress, the contact surface of the electrical connector produces a small relative displacement, and a micro-motion wear process occurs, which in turn causes the degradation of the electrical performance of the contact surface and the hazard of reduced reliability. Contact resistance is a key indicator for measuring the contact performance and expected life of electrical connectors. Therefore, it is particularly important to construct a model by analyzing the characteristic values ​​of wear debris to predict the contact resistance of the electrical connector for the reliable operation, safety and stability of power systems and electrical equipment.

[0003] A large number of experiments, data and theories show that in the process of micro-wear of electrical connectors, there is a significant correlation and complex nonlinear relationship between the generation and accumulation of wear debris and the change of contact resistance. The existing technology mostly uses BP neural network to establish a prediction model for the wear debris characteristic value and contact resistance under micro-wear of electrical connectors. However, BP neural network has problems such as slow convergence speed and easy to fall into local optimal solution.

[0004] In order to solve the above technical problems and further improve the performance prediction level of electrical equipment, a contact resistance prediction method based on the improved ant lion algorithm is proposed. Summary of the invention

[0005] In view of the problems in the prior art, the present invention proposes a contact resistance prediction method for an electrical connector under micro-wear conditions based on the ALO-BP algorithm. The method is used to solve the problem that the contact resistance of the electrical connector cannot be accurately predicted when the amount of data is small.

[0006] The present invention is implemented through the following technical solutions:

[0007] A contact resistance prediction method based on an improved antlion algorithm comprises the following steps:

[0008] S1: Formulate a temperature cycle test plan, adopt a pre-experimental test method, use a high-precision contact resistance meter to measure the contact resistance data, and use a wear characteristic value sensing device designed based on the capacitance tomography non-destructive testing mechanism to obtain the wear characteristic value of the contact surface;

[0009] S2: Build an electrical connector fretting wear test platform, conduct fretting wear tests under temperature cycles, and obtain the wear debris characteristic values ​​and contact resistance of the electrical connector under fretting wear as a database.

[0010] S3: Establish a prediction model based on ALO-BP neural network.

[0011] S4: Bring the training samples into the ALO-BP neural network prediction model for training.

[0012] S5: Assign the global optimal value to the weight and threshold of the BP neural network.

[0013] S6: BP neural network performs local optimization to obtain the optimal prediction model.

[0014] S7: Use the optimal prediction model to predict the contact resistance, denormalize the output data, and obtain the contact resistance prediction result of the electrical connector under micro-wear conditions.

[0015] Furthermore, the specific method of formulating the test plan described in S1 is: the process of the temperature cycle experiment is slow, and the temperature cycle accelerated degradation test is designed in combination with the Arrhenius model to study the micro-motion wear problem. The specific implementation method of the test is: build a micro-motion wear test platform for electrical connectors under temperature cycles, conduct temperature cycle accelerated tests, measure the wear debris characteristic value of the electrical connector contact parts of the micro-motion wear test using a sensor device based on the principle of electrical capacitance tomography (ECT), and use a high-precision contact resistance meter to measure the contact resistance value data.

[0016] Furthermore, the basic Arrhenius model equation for the experimental scheme is:

[0017] L = A exp(E a / KT)

[0018] Furthermore, the database construction described in S2 is to reduce the redundancy of the wear debris characteristic value input by combining the whitening method and PCA; and to construct a database by screening some of the wear debris characteristic values ​​after dimensionality reduction and the contact resistance.

[0019] Furthermore, the data preprocessing described in S2 is to divide the wear debris characteristic values ​​and contact resistance data obtained from the micro-wear test into a training set and a test set, with 80% of the data being set as the training set and 20% of the data being set as the test set.

[0020] Furthermore, the ALO-BP neural network prediction model described in S3 is based on the classic BP neural network model, adopts an optimization algorithm of the ant lion predation behavior in nature, and uses the global search capability of the ant lion algorithm to optimize the BP neural network model.

[0021] Furthermore, the BP neural network is a nonlinear classical model, and the BP neural network algorithm is as follows:

[0022] Calculate the number of hidden layer nodes:

[0023]

[0024] Where n and m are the number of input layer nodes and output layer nodes respectively, p is the number of hidden layer nodes, and α is an integer between 1 and 10.

[0025] Calculation of hidden layer input and output:

[0026] The input of each neuron in the hidden layer is:

[0027]

[0028] In the formula, i and j are the number of input layers and hidden layers respectively; the hidden layer input vector obtained by the lth training data is hi(l)=(hi 1 (l),hi 2 (l), ...hi p (l));θ j is the threshold between the hidden layer and the input and output layers; W ij is the weight between the input layer and the hidden layer.

[0029] The output of each neuron in the hidden layer is:

[0030] ho j (l) = f(hi j (l)) j=1,2...,p

[0031] In the formula, i and j are the number of input layers and hidden layers respectively; the hidden layer output vector obtained by the lth training data is ho(l)=(ho 1 (l),ho 2 (l), ...ho p (l));

[0032] Calculation of output layer input and output:

[0033] The input of each neuron in the output layer is:

[0034]

[0035] Where f(·) is the activation function, and the hidden layer input vector obtained from the lth training data is hi(l)=(hi 1 (l),hi 2 (l), ...hi p (l));θ k is the threshold of each neuron in the output layer, W jk is the weight between the hidden layer and the output layer.

[0036] The output of each neuron in the output layer is:

[0037] ho k (l) = f(hi k (l)) k=1,2...,m

[0038] In the formula, the hidden layer output vector obtained from the lth training data is Yo(l)=(Yo 1 (l), Yo 2 (l), ...Yo m (l)).

[0039] The partial derivative of the error function is calculated:

[0040] The partial derivative of the error function η k The calculation formula of (l) is:

[0041] η k (l)=(D k (l)-Yo k (l))Yo k (l)(1-Yo k (l)) k=1,2...,m

[0042] Where D k (l) is the expected output.

[0043] The partial derivative of the error function η j The calculation formula of (l) is:

[0044]

[0045] In the formula, ho j (l) is the output of each neuron in the hidden layer.

[0046] Update network weights and thresholds:

[0047] Update the weights between the neurons in the hidden layer and output layer of the network:

[0048] w jk (t+1)=w jk (t)+λη k ho j

[0049] In the formula, λ is the learning efficiency, which is set to 0.01, and η k is the partial derivative of the error function, ho j is the output of each neuron in the hidden layer.

[0050] The threshold for updating the output layer neurons is:

[0051] θ k (t+1)=θ k (t)+λη k

[0052] Update the weights between the neurons in the hidden layer and input layer of the network as follows:

[0053] w ij (t+1)=w ij (t)+λη j X i

[0054] The threshold for updating the output layer neurons is:

[0055] θ j (t+1)=θ j (t)+λη j

[0056] In the formula, λ is the learning efficiency and is taken as 0.01, η j is the partial derivative of the error function, X i The input layer

[0057] The error calculation is expressed as:

[0058]

[0059] Where D k (l) is the expected output, Yo k (l) is the output of the output layer.

[0060] The ALO algorithm is a biologically inspired algorithm that simulates the hunting behavior of ant lions in nature. It has the advantages of few adjustment parameters, high convergence accuracy, and good robustness. The specific steps of introducing the ALO algorithm to optimize the weights and thresholds in the constructed model are as follows:

[0061] Antlion builds a trap:

[0062] Antlions are randomly selected to build traps through a roulette operation based on their fitness values.

[0063] Ants walk randomly:

[0064] X(t) = [0, cums(2r(t 1)-1)...cums(2r(t n )-1)]

[0065] Where cums is used to calculate the cumulative sum; t is the step length of the random walk; r(t) is the random function, which is defined as:

[0066]

[0067] In the formula, rand is a random number between [0, 1]

[0068] Record and save the position and fitness of ants during optimization, which can be expressed as:

[0069]

[0070]

[0071] Where M AL is the position of the ant; A nd is the value of d of the nth ant; M OA is the fitness of the ant; F(·) is the objective function;

[0072] Antlion's location M AL and fitness M OAL They can be:

[0073]

[0074]

[0075] In the formula, AL nd is the value of the d-dimension of the n-th antlion.

[0076] In order to ensure that the ants walk randomly in the search space, this paper initializes the positions of the ants, namely:

[0077]

[0078] In the formula, a i and d i are the maximum and minimum values ​​of the position of the i-th random walking ant respectively; and are the maximum and minimum values ​​of the position of the i-th random walking ant after t iterations.

[0079]

[0080]

[0081] Where t is the current iteration number; T is the maximum iteration number; I is the ratio.

[0082] Rebuilding the trap: After the antlion captures the prey, it evaluates the fitness value of the ant’s new position and updates the antlion’s position.

[0083] Elite: The average of the wandering positions of the randomly operated ant lion and the elite ant lion is used as the updated position of the ant.

[0084] Furthermore, the training samples described in S4 are brought into the ALO-BP neural network prediction model for training; the specific steps of the training process are as follows:

[0085] S4: Create a BP neural network model and normalize the sample data of the training neural network: the normalization formula is:

[0086]

[0087] In the formula, is the wear debris characteristic value after normalization; x i is the test data of wear debris characteristic value; x min and x max are the minimum and maximum values ​​respectively.

[0088] S5.1: Initialize the weights and thresholds of the antlion population and BP neural network, and set the algorithm parameters, including the number of ants and antlion populations, the dimension and range of the search space, and the maximum number of iterations.

[0089] In the formula, is the wear debris characteristic value after normalization; x i is the test data of wear debris characteristic value; x min and x max are the minimum and maximum values ​​respectively.

[0090] S5.2: Update the position and speed of the antlion population according to the search strategy of the antlion algorithm.

[0091] S5.3: Update the weights and thresholds of the BP neural network according to the location of the ant lion population.

[0092] S5.4 Set fitness function: Use the mean square error between the actual number of temperature cycles and the predicted value to suggest a fitness function.

[0093] Furthermore, the contact resistance prediction is performed using the optimal prediction model described in S7, and the output data is denormalized to obtain the contact resistance prediction result of the electrical connector under micro-wear. The denormalization formula is as follows:

[0094]

[0095] In the formula, is the normalized output prediction value; yi is the denormalized actual predicted value. Beneficial effects:

[0096] This paper proposes an algorithm based on the predation behavior of ant lions in nature to optimize the BP neural network model to predict the contact resistance of electrical connectors under micro-wear conditions. Compared with the traditional BP neural network model, it has the following beneficial effects:

[0097] (1) The BP neural network model optimized by the improved ant lion predation algorithm surpasses the traditional BP neural network model in terms of prediction accuracy. The traditional BP neural network is widely used in the prediction field due to its feedforward and feedback structure. This structure gives it a stronger learning ability than ordinary neural networks, making it particularly suitable for building nonlinear prediction models. The selection of initial parameters has a significant impact on the performance of the BP neural network. By using the improved ant lion predation algorithm to adjust the initial weights and thresholds of the neural network, the error of the network output can be effectively reduced, thereby improving the prediction accuracy of the BP neural network prediction model.

[0098] (2) In this technical solution, we use an improved ant lion predation algorithm to adjust the initial weights and thresholds of the neural network to prevent the BP neural network from falling into a local optimal solution during the training process. This helps to further improve the generalization ability and adaptability of the BP neural network, and can shorten the time required for the network to determine the weights and thresholds, thereby speeding up the prediction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 It is a schematic diagram of the structure of the BP performance degradation model;

[0100] Figure 2 It is the BP neural network training flow chart;

[0101] Figure 3 It is the performance prediction flow chart of ALO-BP neural network;

[0102] Figure 4 It is a comparison chart between the prediction results of the BP neural network model and the actual value;

[0103] Figure 5 It is the relative deviation graph predicted by the ALO-BP neural network model optimized by the ant lion predation algorithm; DETAILED DESCRIPTION

[0104] The technical solution is described below in conjunction with the accompanying drawings to make the objects and advantages of the present invention clearer and easier to understand for those skilled in the art. However, the present invention is not limited to the scope of the specific implementation methods. For those of ordinary skill in the art, as long as the changes are obvious within the scope and spirit of the claims, they are all protected by the present invention.

[0105] A method for predicting the contact resistance of an electrical connector, specifically, a method for predicting the contact resistance of an electrical connector under micro-wear conditions, based on an optimization algorithm based on the predation behavior of ant lions in nature, using the global search capability of the ant lion algorithm to optimize a BP neural network model (abbreviated as ALO-BP neural network model). The method specifically includes the following steps:

[0106] S1: Formulate a temperature cycle test plan, build a micro-motion wear test platform for electrical connectors under temperature cycle, adopt a pre-experimental test method, use a high-precision contact resistance meter to measure the contact resistance value data, and use a wear characteristic value sensing device designed based on the capacitance tomography non-destructive testing mechanism to obtain the wear characteristic value of the contact surface;

[0107] S2: Conduct temperature cycle test, use the sensing device based on capacitance tomography technology to measure the wear characteristic value of the electrical connector contact, use the high-precision contact resistance meter to measure the contact resistance of the electrical connector, select 24 electrical connector samples for test, the test includes low temperature 10℃ and four high temperature stress levels (50℃, 59℃, 73℃, 95℃) for high temperature (1h) and low temperature (1h) cycle test, the number of cycles is 100. The specific test plan is shown in Table 1: Table 1 Stress levels of temperature and humidity accelerated life test for electrical connectors

[0108] S3: Obtain 150 sets of data to form a database. Divide the 150 sets of data into training samples and test samples at a ratio of 4:1. Establish an ALO-BP neural network prediction model, which is improved by introducing the ant lion algorithm into the classic BP neural network, and further optimizes the training process of the BP neural network.

[0109] Figure 1 This is the structure diagram of the BP neural network. The BP neural network is a model widely used in the prediction field. Its internal structure includes feedforward and feedback mechanisms, which makes it have more powerful learning ability than traditional neural networks and is more suitable for building nonlinear prediction models. Figure 2 The training process of BP neural network is shown, and the algorithm steps of BP neural network are as follows:

[0110] Calculate the number of hidden layer nodes:

[0111]

[0112] Where n and m are the number of input layer nodes and output layer nodes respectively, p is the number of hidden layer nodes, and α is an integer between 1 and 10.

[0113] Calculation of hidden layer input and output:

[0114] The input of each neuron in the hidden layer is:

[0115]

[0116] In the formula, i and j are the number of input layers and hidden layers respectively; the hidden layer input vector obtained by the lth training data is hi(l)=(hi 1 (l),hi 2 (l), ...hi p (l));θ j is the threshold between the hidden layer and the input layer; W ij is the weight between the input layer and the hidden layer.

[0117] The output of each neuron in the hidden layer is:

[0118] ho j (l) = f(hi j (l)) j=1,2...,p

[0119] In the formula, i and j are the number of input layers and hidden layers respectively; the hidden layer output vector obtained by the lth training data is ho(l)=(ho 1 (l),ho 2 (l), ...ho p (l));

[0120] Calculation of output layer input and output:

[0121] The input of each neuron in the output layer is:

[0122]

[0123] Where f(·) is the activation function, and the hidden layer input vector obtained from the lth training data is hi(l)=(hi 1 (l),hi 2 (l), ...hi p (l));θ k is the threshold of each neuron in the output layer, W jk is the weight between the hidden layer and the output layer

[0124] The output of each neuron in the output layer is:

[0125] ho k (l) = f(hi k (l)) k=1,2...,m

[0126] In the formula, the hidden layer output vector obtained from the lth training data is Yo(l)=(Yo 1 (l), Yo2 (l), ...Yo m (l)).

[0127] The partial derivative of the error function is calculated:

[0128] The partial derivative of the error function η k The calculation formula of (l) is:

[0129] η k (l)=(D k (l)-Yo k (l))Yo k (l)(1-Yo k (l)) k=1,2...,m

[0130] Where D k (l) is the expected output.

[0131] The partial derivative of the error function η j The calculation formula of (l) is:

[0132]

[0133] In the formula, ho j (l) is the output of each neuron in the hidden layer.

[0134] Update network weights and thresholds:

[0135] Update the weights between the neurons in the hidden layer and output layer of the network:

[0136] w jk (t+1)=w jk (t)+λη k ho j

[0137] In the formula, λ is the learning efficiency, which is set to 0.01, and η k is the partial derivative of the error function, ho j is the output of each neuron in the hidden layer.

[0138] The threshold for updating the output layer neurons is:

[0139] θ k (t+1)=θ k (t)+λη k

[0140] Update the weights between the neurons in the hidden layer and input layer of the network as follows:

[0141] w ij (t+1)=w ij (t)+λη j Xi

[0142] The threshold for updating the output layer neurons is:

[0143] θ j (t+1)=θ j (t)+λη j

[0144] In the formula, λ is the learning efficiency and is taken as 0.01, η j is the partial derivative of the error function, X i is the input of the input layer.

[0145] The error calculation is expressed as:

[0146]

[0147] Where D k (l) is the expected output, Yo k (l) is the output of the output layer.

[0148] The ALO algorithm is a biologically inspired algorithm that simulates the hunting behavior of ant lions in nature. It has the advantages of few adjustment parameters, high convergence accuracy, and good robustness. The specific steps of introducing the ALO algorithm to optimize the weights and thresholds in the constructed model are as follows:

[0149] Ants walk randomly:

[0150] X(t) = [0, cums(2r(t 1 )-1)...cums(2r(t n )-1)]

[0151] In the formula, cums is used to accumulate the step length of the walk; n represents the number of ants; t is the step length of the random walk; r(t) is a random function, which is defined as:

[0152]

[0153] In the formula, rand is a random number between [0, 1]

[0154] Antlion builds a trap:

[0155] Antlions are randomly selected to build traps through a roulette wheel operation based on their fitness values;

[0156] Record and save the position and fitness of ants during optimization, which can be expressed as:

[0157]

[0158]

[0159] Where M AL is the position of the ant; A nd is the value of d of the nth ant; M OA is the fitness of the ant; F(·) is the objective function.

[0160] Antlion's location M AL and fitness M OAL They can be:

[0161]

[0162] In the formula, AL nd is the value of the d-dimension of the n-th antlion.

[0163] S5: Use the optimized model to predict the performance of electrical equipment.

[0164] By comparing fitness, ant lions prey on ants and update the position of the ant lion population to select new elite ant lions;

[0165] In order to ensure that the ants walk randomly in the search space, this paper initializes the positions of the ants, namely

[0166]

[0167] In the formula, a i and d i are the maximum and minimum values ​​of the position of the i-th random walking ant respectively; and are the maximum and minimum values ​​of the position of the i-th random walking ant after t iterations.

[0168] Ant Slide Trap Center:

[0169] The traps made by ant lions will affect the routes of ants' random walks. To mathematically model this hypothesis, we propose:

[0170]

[0171] c t is the minimum value of all variables at the tth iteration, d t is the maximum value of all variables at the tth iteration.

[0172] in:

[0173]

[0174] Where t is the current iteration number; T is the maximum iteration number; I is the ratio.

[0175] When the fitness value of an ant is smaller than that of an antlion, it is considered that the antlion has captured it, and the antlion will update its position based on the position of the ant.

[0176] Furthermore, the training samples are introduced into the ALO-BP neural network prediction model for training as described in S4;

[0177] Figure 3 The performance prediction flow chart of the ALO-BP neural network. The specific steps of the training process are as follows:

[0178] S3: Create a BP neural network model and normalize the sample data for training the neural network; the normalization formula is:

[0179]

[0180] Where, X i is the wear debris characteristic value after normalization; x i is the test data of wear debris characteristic value; x min and x max are the minimum and maximum values ​​respectively.

[0181] Data processing: First, the data set is processed for prediction, and the processed data is divided into a training set and a prediction set. Since each group of tests has 6 samples and each group of tests has 150 sets of data, 120 sets of data are selected as training sets and 30 sets of data are selected as prediction sets.

[0182] S4.1: Initialize the weights and thresholds of the ant lion population and BP neural network

[0183] Initialize the weights and thresholds of the antlion population and BP neural network, and set the algorithm parameters, including the number of ants and antlion populations, the dimension and range of the search space, and the maximum number of iterations.

[0184] In this experiment, the number of populations is SearchAgents_no = 10, and the maximum number of iterations is Max_iteration = 15. The upper and lower limits of the independent variable are 1 and -1.

[0185] S4.2 Setting the fitness function: The fitness function is suggested using the mean square error between the actual number of temperature cycles and the predicted value.

[0186] S4.3: According to the search strategy of the antlion algorithm, calculate the fitness and update the position and speed of the antlion population.

[0187] Rebuilding the trap: After the ant lion captures the prey, it evaluates the fitness value of the ant's new position. If an individual with a higher fitness value than the ant lion appears in the wandering ant population, the ant lion will be replaced and the ant lion's position needs to be updated.

[0188] Elite: Both the randomly operated ant lion and the elite ant lion will affect the random walk of the ant. Therefore, the average of the wandering positions of the randomly operated ant lion and the elite ant lion is used as the updated position of the ant.

[0189] After each iteration, the antlion with the best fitness is selected as the elite antlion.

[0190] S4.4: Update the weights and thresholds of the neural network based on the location of the ant lion population

[0191] The ant lion algorithm is used to optimize the model parameters of the BP neural network. Through continuous iterations, the optimal parameters are output until the extreme value obtained is less than the set threshold or the maximum number of iterations is reached. The neural network is locally optimized and finally the performance degradation of the electrical connector contacts is predicted through the trained model to obtain the optimal prediction model.

[0192] The normalized sample values, the obtained weights and the thresholds are calculated by the neural network to determine whether the global error meets the requirements. If not, the above steps are continued to update the threshold weights. If the requirements are met, the optimal prediction model is obtained.

[0193] Furthermore, the contact resistance prediction is performed using the optimal prediction model described in S7, and the output data is denormalized to obtain the contact resistance prediction result of the electrical connector under micro-wear. The denormalization formula is as follows:

[0194]

[0195] In the formula, is the normalized output prediction value; y i is the denormalized actual predicted value.

[0196] In order to reflect the advantages of the present invention, the temperature cycle test data were input into the traditional BP neural network and the ALO-BP neural network model improved based on the ant lion algorithm for training. Figure 4 It is a comparison chart between the prediction results of the BP neural network model and the actual value; Figure 5 This is the relative deviation diagram of the prediction of the ALO-BP neural network model optimized by the ant lion predation algorithm. According to the output prediction results, the mean absolute error (MAE) is calculated, ((a) 10℃~50℃; (b) 10℃~59℃; (c) 10℃~73℃; (d) 10℃~95℃), the mean absolute percentage error (MAPE), the root mean square error (RMSE), the calculation results of the classic BP model and the model using the ALO-BP algorithm are shown in Table 2: Table 2 Analysis of error calculation results of classic BP and ALO-BP under different temperature cyclic stress

[0197] The BP neural network prediction model has a large error, and the average absolute percentage error (MAPE) is between 7.6% and 13.2%. The model prediction average absolute percentage error (MAPE) is in the range of 4.1% to 10.1%. Under high temperature difference conditions, the prediction error is within 5%. Compared with the BP degradation model, the percentage of relative deviations greater than 0.1 in the ALO-BP degradation model decreased from 66.7% to 36.7% (10℃~50℃), 43% to 26.7% (10℃~59℃), 40% to 23.3% (10℃~73℃), and 46.7% to 20% (10℃~95℃), indicating that the ALO-BP performance degradation model has greatly improved the model prediction accuracy.

[0198] In terms of algorithm model construction, this application uses the ALO-BP neural network with few adjustment parameters, high convergence accuracy and good robustness as prediction. Compared with the BP neural network, it improves the stability of prediction accuracy and reduces prediction errors. Traditional BP neural networks are prone to fall into local optimality. The ant lion algorithm is introduced to optimize the initial weights and thresholds of the BP neural network to avoid falling into local optimality and improve calculation accuracy and speed.

Claims

1. A method for predicting the contact resistance of an electrical connector, specifically, an optimization algorithm based on the predation behavior of ant lions in nature, wherein the algorithm is characterized in that: the method combines test data to establish a prediction model based on wear debris characteristic values ​​as input and electrical connector contact resistance as output; the specific implementation method is: using the global search capability of the ant lion algorithm to optimize the BP neural network prediction model (abbreviated as ALO-BP neural network model) of the contact resistance prediction method of the electrical connector under micro-wear conditions. By training the ALO-BP neural network model, the optimal prediction model for contact resistance is obtained, and the contact resistance is reliably predicted. The method specifically has the following steps: S1: Formulate a temperature cycle test plan, adopt a pre-experimental test method, use a high-precision contact resistance meter to measure the contact resistance data, and use a wear characteristic value sensing device designed based on the capacitance tomography non-destructive testing mechanism to obtain the wear characteristic value of the contact surface; S2: Build a fretting wear test platform for electrical connectors, conduct fretting wear tests under temperature cycles, and obtain the wear debris characteristic values ​​and contact resistance of electrical connectors under fretting wear as a database; S3: Establish a prediction model based on ALO-BP neural network; S4: Bring the training samples into the ALO-BP neural network prediction model for training; S5: Assign the global optimal value to the weights and thresholds of the BP neural network S6: BP neural network performs local optimization to obtain the optimal prediction model; S7: Use the optimal prediction model to predict the contact resistance, denormalize the output data, and obtain the contact resistance prediction result of the electrical connector under micro-wear conditions.

2. The method for predicting contact resistance of an electrical connector based on the predation behavior of ant lions in nature according to claim 1, characterized in that: In the steps S1 and S2, the method for detecting the wear debris characteristic value and contact resistance value generated by the fretting wear of the electrical connector is: using a wear debris characteristic value sensing device designed based on the capacitance tomography non-destructive testing mechanism to obtain the wear debris characteristic quantity and distribution information of the contact surface; The contact resistance data is measured using a high-precision contact resistance meter. The acquired wear debris characteristic values ​​and contact resistance values ​​are processed and analyzed, and divided into training data and test samples in a certain proportion and input into the established neural network prediction model.

3. According to claim 1, a method for predicting contact resistance of an electrical connector based on the predation behavior of ant lions in nature, characterized in that: In step S3, a prediction model based on the ALO-BP neural network is established. The ALO algorithm that simulates the predation behavior of ant lions in nature is introduced into the traditional BP neural network model, the global search capability of the traditional BP neural network is improved, and the data training process and speed of the traditional BP neural network are further optimized.

4. According to claim 3, a method for predicting contact resistance of an electrical connector based on the predation behavior of ant lions in nature, characterized in that: The traditional BP neural network is a classic nonlinear model, and its establishment method is as follows: Calculate the number of hidden layer nodes: Where n and m are the number of input layer nodes and output layer nodes respectively, p is the number of hidden layer nodes, and α is an integer between 1 and 10. Calculation of hidden layer input and output: The input of each neuron in the hidden layer is: Where i and j are the number of input layers and hidden layers respectively; the hidden layer input vector obtained from the lth training data is hi(l) = (hi1(l), hi2(l), …hi p (l)); θ j is the threshold between the hidden layer and the input and output layers; W ij is the weight between the input layer and the hidden layer. The output of each neuron in the hidden layer is: it j (l)=f(hi j (l)) j=1,2…,p In the formula, i and j are the number of input layers and hidden layers respectively; the hidden layer output vector obtained from the lth training data is ho(l) = (ho1(l), ho2(l), ...ho p (l)). Calculation of output layer input and output: The input of each neuron in the output layer is: Where f(·) is the activation function, and the hidden layer input vector obtained from the lth training data is hi(l)=(hi1(l),hi2(l),...hi p (l))θ k is the threshold of each neuron in the output layer, W jk is the weight between the hidden layer and the output layer. The output of each neuron in the output layer is: it k (l)=f(hi k (l)) k=1,2…,m In the formula, the hidden layer output vector obtained from the lth training data is Yo(l) = (Yo1(l), Yo2(l), ...Yo m (l)). The partial derivative of the error function is calculated: The partial derivative of the error function η k The calculation formula of (l) is: η k (l)=(D k (l)-They k (l))They k (l)(1-Yo) k (l)) k=1,2...,m Where θ k (l) is the expected output. The partial derivative of the error function η j The calculation formula of (l) is: In the formula, ho j (l) is the output of each neuron in the hidden layer. Update network weights and thresholds: Update the weights between the neurons in the hidden layer and output layer of the network: w jk (t+1)=w jk (t)+λη k ho j In the formula, λ is the learning efficiency, which is set to 0.01, and η k is the partial derivative of the error function, ho j is the output of each neuron in the hidden layer. The threshold for updating the output layer neurons is: i k (t+1)=θ k (t)+λη k Update the weights between the neurons in the hidden layer and input layer of the network as follows: w ij (t+1)=w ij (t)+λη j X i The threshold for updating the output layer neurons is: i j (t+1)=θ j (t)+λη j In the formula, λ is the learning efficiency and is taken as 0.01, η j is the partial derivative of the error function, X i is the input of the input layer. The error calculation is expressed as: Where D k (l) is the expected output, Yo k (l) is the output of the output layer. The ALO algorithm is a biologically inspired algorithm that simulates the hunting behavior of ant lions in nature. It has the advantages of few adjustment parameters, high convergence accuracy, and good robustness. The specific steps of introducing the ALO algorithm to optimize the weights and thresholds in the constructed model are as follows: Antlion builds a trap: Antlions are randomly selected to build traps through a roulette operation based on their fitness values. Ants walk randomly: X(t)=[0,cums(2r(t1)-1)…cums(2r(t n )-1)] Where cums is used to calculate the cumulative sum; t is the step length of the random walk; r(t) is the random function, which is defined as: In the formula, rand is a random number between [0, 1] Record and save the position and fitness of ants during optimization, which can be expressed as: Where M AL is the position of the ant; A nd is the value of d of the nth ant; M OA is the fitness of the ant; F(·) is the objective function. Antlion's location M AL and fitness M OAL They can be: In the formula, AL nd is the value of the d-dimension of the n-th antlion. In order to ensure that the ants walk randomly in the search space, this paper initializes the positions of the ants, namely: In the formula, a i and d i are the maximum and minimum values ​​of the position of the i-th random walking ant respectively; and is the maximum and minimum value of the position of the i-th random walking ant after t iterations Ant Slide Trap Center: in: Where t is the current iteration number; T is the maximum iteration number; I is the ratio. Rebuilding the trap: After the antlion captures the prey, it evaluates the fitness value of the ant’s new position and updates the antlion’s position. Elite: The average of the wandering positions of the randomly operated ant lion and the elite ant lion is used as the updated position of the ant. Furthermore, the training samples described in S4 are brought into the ALO-BP neural network prediction model for training; the specific steps of the training process are as follows: S4: Create a BP neural network model and normalize the sample data of the training neural network; the normalization formula is: In the formula, is the wear debris characteristic value after normalization; x i is the test data of wear debris characteristic value; x min and x max are the minimum and maximum values ​​respectively. S5.1: Initialize the weights and thresholds of the antlion population and BP neural network, and set the algorithm parameters, including the number of ants and antlion populations, the dimension and range of the search space, and the maximum number of iterations. S5.2: Update the position and speed of the antlion population according to the search strategy of the antlion algorithm S5.3: Update the weights and thresholds of the BP neural network based on the location of the ant lion population S5.4 Set the fitness function: Use the mean square error between the actual number of temperature cycles and the predicted value to suggest the fitness function. Furthermore, the contact resistance prediction is performed using the optimal prediction model described in S7, and the output data is denormalized to obtain the contact resistance prediction result of the electrical connector under micro-wear. The denormalization formula is as follows: In the formula, is the normalized output prediction value; y i is the denormalized actual prediction.

Images