Overhead conductor fatigue strength prediction method and device based on neural network, terminal equipment and computer readable storage medium

Abstract

The invention discloses an overhead conductor fatigue strength prediction method and device based on a neural network, terminal equipment and a computer readable storage medium. The method comprises the steps that the average stress and the target fatigue life of a to-be-predicted conductor are acquired; inputting the average stress and the target fatigue life of the to-be-predicted wire into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the to-be-predicted wire according to the average stress and the target fatigue life of the to-be-predicted wire; wherein the fatigue strength prediction model takes the average stress and the fatigue life of the training conductor sample as input and takes the predicted stress amplitude of the training conductor sample as output for training, and network parameters of the fatigue strength prediction model are adjusted in the training process according to the predicted stress amplitude and the actual stress amplitude of the training conductor sample. By implementing the method, the stress amplitude of the to-be-predicted wire can be quickly obtained, the prediction period is greatly shortened, and the working efficiency is improved.

Description

Technical Field

[0001] The invention relates to the field of fatigue strength analysis, and in particular to a method for predicting fatigue strength of an overhead conductor based on a neural network. Background Art

[0002] Overhead transmission lines are a type of power transmission device used to transmit electric energy. They are usually made of metal materials, such as aluminum or copper. They are suspended on power towers and transmit electric energy from power plants to users through transmission lines. In transmission lines, the conductors are subject to wind loads, which is one of the main causes of conductor fatigue damage. The force exerted by the wind on the conductors will cause the conductors to vibrate and swing, resulting in uneven stress distribution of the conductors. Under the action of wind loads for a long time, the conductors may suffer fatigue cracking, fracture or even break. Fatigue damage has a serious impact on the safe operation of transmission lines. When fatigue damage occurs to the conductors, it may cause the conductors to fall off or form a short circuit between the conductors, which will cause failures in the power system, affect the stability of power supply, and may have an adverse effect on electricity consumption in nearby areas.

[0003] In order to ensure the safe operation of transmission lines and effectively prevent and control fatigue damage of conductors, prestress is introduced into the conductors to effectively reduce the vibration and stress changes of the conductors, thereby reducing the risk of fatigue damage and improving the fatigue strength of the conductors. The commonly used method is to perform fatigue tests on conductors under different average stresses through experimental means. However, this experimental method has a long cycle and requires a lot of time and resources to conduct fatigue tests. From designing conductor samples to conducting experiments, and analyzing and verifying the results, the entire process requires a lot of time and money. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, terminal equipment and computer-readable storage medium for predicting fatigue strength of overhead conductors based on neural networks, which can quickly obtain the stress amplitude of the conductor to be predicted, greatly shorten the prediction cycle and improve work efficiency.

[0005] An embodiment of the present invention provides a method for predicting fatigue strength of an overhead conductor based on a neural network, comprising:

[0006] Obtain the average stress and target fatigue life of the conductor to be predicted;

[0007] Inputting the average stress and target fatigue life of the conductor to be predicted into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the conductor to be predicted according to the average stress and target fatigue life of the conductor to be predicted;

[0008] Among them, the fatigue strength prediction model takes the average stress and fatigue life of the training wire samples as input and the predicted stress amplitude of the training wire samples as output for training. During the training process, the network parameters of the fatigue strength prediction model are adjusted according to the predicted stress amplitude and actual stress amplitude of the training wire samples.

[0009] Furthermore, the average stress and target fatigue life of the conductor to be predicted are obtained, including:

[0010] Obtain the average tension of the conductor to be predicted and the total cross-sectional area of ​​the conductor;

[0011] The average tension of the conductor to be predicted is divided by the total cross-sectional area of ​​the conductor to be predicted to obtain the average stress of the conductor to be predicted;

[0012] Obtain the target fatigue life of the conductor to be predicted.

[0013] Furthermore, the preset fatigue strength prediction model is trained in the following way:

[0014] Obtain the average stress, different stress amplitudes and fatigue life corresponding to each stress amplitude of several groups of training wire samples;

[0015] Normalizing all training conductor sample data to obtain the average stress, different stress amplitudes, and fatigue life corresponding to each stress amplitude of the normalized training conductor samples;

[0016] The average stress of each group of normalized training wire samples, different stress amplitudes, and fatigue life corresponding to each stress amplitude are divided into a training set, a validation set, and a test set according to a preset ratio;

[0017] The training set and the validation set are input into the fatigue strength prediction model to be trained for iterative training until the model error converges to obtain the preset fatigue strength prediction model.

[0018] Furthermore, the training set and the validation set are input into the fatigue strength prediction model to be trained for iterative training until the model error converges, and a preset fatigue strength prediction model is obtained, including:

[0019] Initialize the network parameters, verification cycle number and training parameters of the fatigue strength prediction model; the training parameters include the number of training rounds;

[0020] In each iterative training, a batch of data is sequentially selected from the training set and input into the input layer of the fatigue strength prediction model;

[0021] Perform forward calculation on the neurons of each layer according to each input data in the selected batch to generate the final predicted stress amplitude of each input data;

[0022] Calculating a first loss function according to a final predicted stress amplitude of each input data and a true stress amplitude corresponding to each input data;

[0023] Adjusting the network parameters of the fatigue strength prediction model to be trained according to the first loss function;

[0024] Count the number of training rounds;

[0025] Determine whether the number of training rounds meets the number of verification cycles. If so, input the data of the verification set into the current fatigue strength prediction model to calculate the second loss function; if not, select the next batch of data from the training set and input it into the input layer for iterative training;

[0026] Determine whether the second loss function converges. If so, use the current fatigue strength prediction model as the preset fatigue strength prediction model; if not, select the next batch of data from the training set and input it into the input layer for iterative training.

[0027] Further, for each input data, forward calculation is performed on the neurons of each layer according to each input data in the selected batch to generate the final predicted stress amplitude of each input data, including:

[0028] For each layer, the input and output of each neuron are calculated by the following formula:

[0029] z_j^l=Σw_jk^l·a_k^(l-1)+b_j^l;

[0030] a_j^l=f^l(z_j^l);

[0031] Wherein, z_j^l represents the input of the jth neuron in the lth layer, a_k^(l-1) represents the output of the kth neuron in the previous layer of the lth layer, w_jk^l represents the connection weight between two neurons, b_j^l represents the bias term of the jth neuron in the lth layer, a_j^l represents the output of the jth neuron in the lth layer, f^l(·) represents the activation function of the lth layer, the activation function of the hidden layer in the fatigue strength prediction model is f(x)=1 / (1+e^(-x)), the activation function of the output layer in the fatigue strength prediction model is g(x)=x, and x represents the input corresponding to the activation function;

[0032] The output of the output layer is taken as the final predicted stress amplitude of the network.

[0033] Furthermore, the network parameters include: weight matrix and bias vector;

[0034] The network parameters of the fatigue strength prediction model to be trained are adjusted according to the first loss function, including:

[0035] Reversely calculate the partial derivatives of each neuron in the output layer and the hidden layer according to the first loss function;

[0036] Calculate the gradient of the weight matrix and bias vector based on the partial derivatives of each neuron in the output layer and hidden layer;

[0037] Update the weight matrix and bias vector according to their gradients.

[0038] Furthermore, the neural network-based overhead conductor fatigue strength prediction method further includes:

[0039] The average stress of the conductor to be predicted, the target fatigue life, and the stress amplitude predicted by the model are plotted as a three-dimensional surface graph.

[0040] Based on the above method embodiment, the present invention provides a corresponding device embodiment, including: a module for acquiring data to be predicted, and a module for predicting stress amplitude;

[0041] A module for obtaining data to be predicted, used for obtaining the average stress and target fatigue life of the conductor to be predicted;

[0042] A stress amplitude prediction module, used for inputting the average stress and target fatigue life of the conductor to be predicted into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the conductor to be predicted according to the average stress and target fatigue life of the conductor to be predicted;

[0043] Among them, the fatigue strength prediction model takes the average stress and fatigue life of the training wire samples as input and the predicted stress amplitude of the training wire samples as output for training. During the training process, the network parameters of the fatigue strength prediction model are adjusted according to the predicted stress amplitude and actual stress amplitude of the training wire samples.

[0044] Based on the above-mentioned method embodiment, the present invention provides a corresponding terminal equipment embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the overhead wire fatigue strength prediction method based on neural network as described in the present invention are implemented.

[0045] Based on the above-mentioned method embodiment, the present invention provides a corresponding computer-readable storage medium embodiment, including: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the overhead wire fatigue strength prediction method based on neural network as described in the present invention when the computer program is running.

[0046] Compared with the prior art, the beneficial effects of the embodiment of this solution are:

[0047] The present invention can quickly predict the stress amplitude of the conductor under given average stress and target life conditions through a preset fatigue strength prediction model, thereby reflecting the fatigue strength of the conductor. The model learns the relationship between the stress amplitude of the training conductor sample and the average stress and fatigue life of the training conductor sample. Compared with traditional experiments that require long-term experiments and tests to obtain data with a long stress amplitude cycle, the fatigue strength prediction model can quickly obtain the stress amplitude of the conductor to be predicted. When predicting different working conditions, the prediction cycle is greatly shortened and work efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of a method for predicting fatigue strength of overhead conductors based on a neural network provided by an embodiment of the present invention;

[0049] Figure 2 is a curve diagram showing the relationship between the average stress and the stress amplitude of the conductor under conditions of different target fatigue lives provided by an embodiment of the present invention;

[0050] Figure 3 It is a flowchart of a training process of a fatigue strength prediction model provided by an embodiment of the present invention;

[0051] Figure 4 is a three-dimensional surface diagram of partial average stress, target fatigue life and stress amplitude provided by an embodiment of the present invention;

[0052] Figure 5 It is a structural schematic diagram of an overhead conductor fatigue strength prediction device based on a neural network provided by an embodiment of the present invention;

[0053] Figure 6 It is a schematic structural diagram of another neural network-based overhead conductor fatigue strength prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0056] like Figure 1As shown, an embodiment of the present invention provides a method for predicting fatigue strength of overhead wires based on a neural network, the method comprising at least the following steps:

[0057] Step S1: obtaining the average stress and target fatigue life of the conductor to be predicted;

[0058] In a preferred embodiment, obtaining the average stress and target fatigue life of the conductor to be predicted includes:

[0059] Obtain the average tension of the conductor to be predicted and the total cross-sectional area of ​​the conductor;

[0060] The average tension of the conductor to be predicted is divided by the total cross-sectional area of ​​the conductor to be predicted to obtain the average stress of the conductor to be predicted;

[0061] Obtain the target fatigue life of the conductor to be predicted.

[0062] This embodiment takes the aluminum stranded steel core (ACSR) conductor as an example for explanation:

[0063] For step S1, the average tension and total cross-sectional area of ​​the steel core aluminum stranded wire conductor to be predicted under different conditions are obtained by monitoring equipment, the average tension of the steel core aluminum stranded wire conductor to be predicted is divided by the total cross-sectional area, and the values ​​of different average stresses of the steel core aluminum stranded wire conductor to be predicted (unit: MPa) are calculated. In this embodiment, the prediction conditions are that the average stress is 20MPa, 40MPa, 60MPa, 80MPa and 100MPa; and the target fatigue life of the steel core aluminum stranded wire conductor to be predicted is determined by the design standard. For example, this embodiment needs to predict the target fatigue life within the range of 1e5 (cycles), 1e6 (cycles), 1e7 (cycles) and 1e8 (cycles).

[0064] It should be noted that the fatigue life of the wire is 1e5 (cycles), which means the number of cycles the wire can safely withstand when subjected to alternating loading (alternating tension or compression) of the same stress level, i.e., it can safely withstand 100,000 cycles with a low risk of fatigue failure.

[0065] Step S2: inputting the average stress and target fatigue life of the conductor to be predicted into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the conductor to be predicted according to the average stress and target fatigue life of the conductor to be predicted;

[0066] Among them, the fatigue strength prediction model takes the average stress and fatigue life of the training wire samples as input and the predicted stress amplitude of the training wire samples as output for training. During the training process, the network parameters of the fatigue strength prediction model are adjusted according to the predicted stress amplitude and actual stress amplitude of the training wire samples.

[0067] For step S2, a preset fatigue strength prediction model is used to predict whether the steel core aluminum stranded wire conductor meets the fatigue requirements of 1e5 (cycles), 1e6 (cycles), 1e7 (cycles) and 1e8 (cycles) respectively when the average stress is 20MPa, 40MPa, 60MPa, 80MPa and 100MPa. By inputting different average stresses and target fatigue life, the fatigue strength prediction model can output the corresponding stress amplitude. The results are as follows: Figure 2 As shown in the figure, the relationship curve between the average stress and stress amplitude of the conductor under different target fatigue life conditions.

[0068] Through the prediction results, engineers can make corresponding design decisions based on actual needs and target requirements. For example, when designing a conductor, they can choose the appropriate material, size, and structure based on the predicted stress amplitude to ensure the reliability and safety of the conductor under a given fatigue life.

[0069] Compared with traditional experimental methods, which require a large number of experiments and tests, involving the preparation, loading and measurement of actual wires, which is time-consuming and costly; the use of fatigue strength prediction models can predict different working conditions in a short time, saving experimental time and cost.

[0070] In addition, the fatigue strength prediction model can be applied to different conductor materials and structures and make predictions under various working conditions, while traditional experimental methods may be limited by experimental conditions and the availability of conductor samples.

[0071] The following is a detailed description of the training process of the fatigue strength prediction model:

[0072] like Figure 3 As shown in Figure 1, the training process of the fatigue strength prediction model includes the following steps:

[0073] Step S201: obtaining the average stress, different stress amplitudes and fatigue life corresponding to each stress amplitude of several groups of training wire samples;

[0074] For step S201, M groups of training wire samples under different average stress conditions are obtained, wherein each group of training wire samples includes N different stress amplitudes and fatigue life corresponding to each stress amplitude under the same average stress condition, and M×N “average stress-stress amplitude-fatigue life” data points are obtained, i.e., M×N (σ_m, σ_a, N_f);

[0075] Among them, the average stress σ_m of each data point in the training wire sample is calculated by the following formula:

[0076] σ_m=F_m / A_Al;

[0077] The stress amplitude σ_a is calculated by the following formula:

[0078] σ_a=Δσ / 2=(σ_max-σ_min) / 2;

[0079] The fatigue life N_f is obtained by the number of vibration cycles of the specified number of wire fractures:

[0080] N_f=N_b;

[0081] F_m represents the average tension of the training wire, A_Al represents the total cross-sectional area of ​​the wire, σ_max represents the maximum stress of the wire in one vibration cycle, σ_min represents the minimum stress of the wire in one vibration cycle, and N_b represents the number of vibration cycles that causes a specified number of wires to break.

[0082] Step S202: normalizing all training conductor sample data to obtain the normalized average stress, different stress amplitudes, and fatigue life corresponding to each stress amplitude of the training conductor samples;

[0083] For step S202, in order to improve the efficiency and stability of the neural network training in the fatigue strength prediction model, it is necessary to normalize the data points, map each variable in the data points to the interval [0,1], find the maximum and minimum values ​​of each variable in each data point, and map the value of each data point to the interval [0,1] by the following normalization formula:

[0084] x'=(x-x_min) / (x_max-x_min);

[0085] Among them, x represents the initial value of the variable, x' represents the normalized value of the variable, x_max represents the maximum value of the variable in this category, and x_min represents the minimum value of the variable in this category.

[0086] Step S203: dividing the average stress, different stress amplitudes and fatigue life corresponding to each stress amplitude of each group of normalized training wire samples into a training set, a validation set and a test set according to a preset ratio;

[0087] For step S203, the normalized data points "average stress-stress amplitude-fatigue life" are divided into a training set, a validation set and a test set according to a preset ratio (6:2:2). The training set is used to train the fatigue strength prediction model, the validation set is used to verify whether the model converges, and the test set is used to evaluate the performance of the final trained model.

[0088] Step S204: input the training set and the validation set into the fatigue strength prediction model to be trained for iterative training until the model error converges to obtain a preset fatigue strength prediction model.

[0089] In a preferred embodiment, the training set and the validation set are input into the fatigue strength prediction model to be trained for iterative training until the model error converges, thereby obtaining a preset fatigue strength prediction model, including:

[0090] Initialize the network parameters, verification cycle number and training parameters of the fatigue strength prediction model; the training parameters include the number of training rounds;

[0091] In each iterative training, a batch of data is sequentially selected from the training set and input into the input layer of the fatigue strength prediction model;

[0092] Perform forward calculation on the neurons of each layer according to each input data in the selected batch to generate the final predicted stress amplitude of each input data;

[0093] Calculating a first loss function according to a final predicted stress amplitude of each input data and a true stress amplitude corresponding to each input data;

[0094] Adjusting the network parameters of the fatigue strength prediction model to be trained according to the first loss function;

[0095] Count the number of training rounds;

[0096] Determine whether the number of training rounds meets the number of verification cycles. If so, input the data of the verification set into the current fatigue strength prediction model to calculate the second loss function; if not, select the next batch of data from the training set and input it into the input layer for iterative training;

[0097] Determine whether the second loss function converges. If so, use the current fatigue strength prediction model as the preset fatigue strength prediction model; if not, select the next batch of data from the training set and input it into the input layer for iterative training.

[0098] In a preferred embodiment, for each input data, forward calculation is performed on neurons in each layer according to each input data in the selected batch to generate a final predicted stress amplitude for each input data, including:

[0099] For each layer, the input and output of each neuron are calculated by the following formula:

[0100] z_j^l=Σw_jk^l·a_k^(l-1)+b_j^l;

[0101] a_j^l=f^l(z_j^l);

[0102] Wherein, z_j^l represents the input of the jth neuron in the lth layer, a_k^(l-1) represents the output of the kth neuron in the previous layer of the lth layer, w_jk^l represents the connection weight between two neurons, b_j^l represents the bias term of the jth neuron in the lth layer, a_j^l represents the output of the jth neuron in the lth layer, f^l(·) represents the activation function of the lth layer, the activation function of the hidden layer in the fatigue strength prediction model is f(x)=1 / (1+e^(-x)), the activation function of the output layer in the fatigue strength prediction model is g(x)=x, and x represents the input corresponding to the activation function;

[0103] The output of the output layer is taken as the final predicted stress amplitude of the network.

[0104] In a preferred embodiment, the network parameters include: a weight matrix and a bias vector;

[0105] The network parameters of the fatigue strength prediction model to be trained are adjusted according to the first loss function, including:

[0106] Reversely calculate the partial derivatives of each neuron in the output layer and the hidden layer according to the first loss function;

[0107] Calculate the gradient of the weight matrix and bias vector based on the partial derivatives of each neuron in the output layer and hidden layer;

[0108] Update the weight matrix and bias vector according to their gradients.

[0109] For step S204, the training set and the validation set obtained in step S203 are input into the fatigue strength prediction model to be trained for iterative training until the model error converges to obtain a preset fatigue strength prediction model.

[0110] First, before iterative training, the network parameters, verification cycle number and training parameters of the fatigue strength prediction model need to be initialized. The network parameters include the weight matrix W and the bias vector b, and the training parameters include: the number of training rounds epoch, the number of batches batch, the batch size m, the learning rate η and the momentum factor α.

[0111] The dimensions of the weight matrix W are:

[0112] W_1:(n_in,n_h);

[0113] W_2:(n_h,n_out);

[0114] The dimension of the bias vector b is:

[0115] b_1:(n_h,1);

[0116] b_2: (n_out, 1);

[0117] Among them, n_in represents the number of neurons in the input layer (2 in this embodiment), n_h represents the number of neurons in the hidden layer (14 in this embodiment), and n_out represents the number of neurons in the output layer (1 in this embodiment);

[0118] The number of training epochs represents the number of iterative trainings, initialized to 0; the number of batches represents the number of small batches into which the training set is divided; the batch size m represents the number of data points in each small batch; the learning rate represents the step size for updating parameters each time; the momentum factor represents the weight for considering historical gradients when updating parameters.

[0119] After initialization is completed, the BP algorithm is used to perform iterative training of the neural network on the fatigue strength prediction model to be trained. For each iterative training, a batch of data points is selected from the training set and input into the input layer of the fatigue strength prediction model for forward propagation calculation. Assuming the input of the i-th data point is x_i and the expected output is d_i, then for each layer of the neural network in the fatigue strength prediction model, the input and output of each neuron are calculated through the following formulas:

[0120] z_j^l = Σw_jk^l·a_k^(l - 1) + b_j^l;

[0121] a_j^l = f^l(z_j^l);

[0122] Among them, z_j^l represents the input of the j-th neuron in the l-th layer, a_k^(l - 1) represents the output of the k-th neuron in the layer above the l-th layer, w_jk^l represents the connection weight between two neurons, b_j^l represents the bias term of the j-th neuron in the l-th layer, a_j^l represents the output of the j-th neuron in the l-th layer, f^l(·) represents the activation function of the l-th layer. The activation function of the hidden layer in the fatigue strength prediction model is f(x) = 1 / (1 + e^(-x)), and the activation function of the output layer in the fatigue strength prediction model is g(x) = x. x represents the input corresponding to the activation function, l = 0 is the input layer, l = L is the output layer, and 0 < l < L is the hidden layer;

[0123] Finally, the output of the output layer is used as the output of the fatigue strength prediction model to obtain the final predicted stress amplitude for each input data point:

[0124] y_i = a_1^L;

[0125] Among them, y_i represents the output of the fatigue strength prediction model, and a_1^L represents the output of the first neuron in the output layer of the fatigue strength prediction model.

[0126] According to the final predicted stress amplitude \(y_i\) of each input data point and the expected stress amplitude \(d_i\) of each data point, calculate the mean squared error (MSE) of this iterative training and use it as the first loss function \(L\):

[0127] \(L=\frac{1}{2m}\cdot\sum_{i = 1}^{m}(d_i - y_i)^2\);

[0128] Then, through the chain rule, starting from the output layer, calculate the partial derivative of the loss function \(L\) with respect to each layer of neurons in the reverse direction, that is, the error term \(\delta\);

[0129] Calculate the error term \(\delta_j^L\) of the \(j\)-th neuron in the output layer (\(I = L\)) through the following formula:

[0130] \(\delta_j^L=(d_j - a_j^L)\cdot f'^L(z_j^L)\);

[0131] where \(d_j\) represents the expected output of the \(j\)-th data point, \(a_j^L\) represents the output of the output layer, and \(f'^L(\cdot)\) represents the derivative of the activation function of the output layer;

[0132] After obtaining the error terms of the output layer neurons, continue to calculate the error term \(\delta_j^l\) of the \(j\)-th neuron in the hidden layer (\(0 < I < L\)) in the reverse direction based on the error terms of the output layer neurons through the following formula:

[0133] \(\delta_j^l=(\sum_{k}\delta_k^{(l + 1)}\cdot w_{kj}^{(l + 1)})\cdot f'^l(z_j^l)\);

[0134] where \(\delta_k^{(l + 1)}\) represents the error term of the \(k\)-th neuron in the next layer of the \(I\)-th layer, and \(w_{kj}^{(l + 1)}\) represents the connection weight between two neurons;

[0135] According to the partial derivatives of each neuron in the output layer and the hidden layer, calculate the gradients of the loss function \(L\) with respect to the weight matrix \(W\) and the bias vector \(b\):

[0136]

[0137] Finally, update the gradients of the weight matrix \(W\) and the bias vector \(b\) of the fatigue strength prediction model through the gradient descent method. When using the gradient descent method, improve the performance of the algorithm by adding the learning rate \(\eta\) and the momentum factor \(\alpha\):

[0138]

[0139] \(w_{jk}^l=w_{jk}^l+\Delta w_{jk}^l\);

[0140]

[0141] \(b_j^l=b_j^l+\Delta b_j^l\);

[0142] At this point, one iteration training of the fatigue strength prediction model is completed, and the current number of training rounds, that is, the number of iterative training of the model, is counted. Assuming that the preset number of verification cycles is 3, it is determined whether the number of training rounds meets the cycle of 3. If the number of iterative training does not meet the cycle of 3, the next batch of data is selected from the training set and input into the model input layer for iterative training;

[0143] If the number of iterative training times meets the cycle of 3, the data of the validation set is input into the fatigue strength prediction model obtained in this training, and the predicted stress amplitude y'_i of each data point in the validation set is obtained. The root mean square error (RMSE) is calculated and used as the second loss function:

[0144] RMSE=√(1 / m·Σ(d_i-y'_i)^2);

[0145] After calculating the second loss function, it is also necessary to determine whether the second loss function converges. If the second loss function has not converged, continue to select the next batch of data from the training set and input it into the model input layer for iterative training; if the second loss function converges, it means that the value of the loss function has stabilized, which means that the current fatigue strength prediction model has learned the characteristics of the data and has achieved good performance. In this case, the current fatigue strength prediction model is used as the preset fatigue strength prediction model in this scheme, thereby completing the training of the entire fatigue strength prediction model.

[0146] In a preferred embodiment, the method for predicting fatigue strength of overhead conductors based on neural networks further includes:

[0147] The average stress of the conductor to be predicted, the target fatigue life, and the stress amplitude predicted by the model are plotted as a three-dimensional surface graph.

[0148] In one embodiment of the present invention, the average stress σ_m of some wires to be predicted, the target fatigue life N_f, and the stress amplitude σ_a predicted by the model are plotted as follows: Figure 4 The three-dimensional surface graph shown in the figure intuitively displays the quantitative relationship between the average stress, stress amplitude and fatigue life. It can also quantitatively reveal the inherent laws between the fatigue characteristics of the wire and the average stress and stress amplitude, which can provide new theoretical basis and practical tools for the fatigue design, life prediction and status evaluation of the wire.

[0149] like Figure 5 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0150] An embodiment of the present invention provides a fatigue strength prediction device for overhead conductors based on a neural network, comprising:

[0151] A module for acquiring data to be predicted and a module for predicting stress amplitude;

[0152] A module for obtaining data to be predicted, used for obtaining the average stress and target fatigue life of the conductor to be predicted;

[0153] A stress amplitude prediction module, used for inputting the average stress and target fatigue life of the conductor to be predicted into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the conductor to be predicted according to the average stress and target fatigue life of the conductor to be predicted;

[0154] Among them, the fatigue strength prediction model takes the average stress and fatigue life of the training wire samples as input and the predicted stress amplitude of the training wire samples as output for training. During the training process, the network parameters of the fatigue strength prediction model are adjusted according to the predicted stress amplitude and actual stress amplitude of the training wire samples.

[0155] like Figure 6 As shown, an embodiment of the present invention provides another device for predicting fatigue strength of overhead conductors based on neural network;

[0156] In a preferred embodiment, the neural network-based overhead conductor fatigue strength prediction device further includes: a fatigue strength prediction model training module;

[0157] A fatigue strength prediction model training module includes: a training data acquisition submodule, a normalization processing submodule, a training data division submodule, and an iterative training submodule;

[0158] A training data acquisition submodule is used to obtain the average stress of several groups of training wire samples, different stress amplitudes and the fatigue life corresponding to each stress amplitude;

[0159] A normalization processing submodule is used to perform normalization processing on all training wire sample data to obtain the average stress, different stress amplitudes and fatigue life corresponding to each stress amplitude of the normalized training wire samples;

[0160] The training data division submodule is used to divide the average stress of each group of normalized training wire samples, different stress amplitudes and the fatigue life corresponding to each stress amplitude into a training set, a validation set and a test set according to a preset ratio;

[0161] The iterative training submodule is used to input the training set and the validation set into the fatigue strength prediction model to be trained for iterative training until the model error converges to obtain a preset fatigue strength prediction model.

[0162] In a preferred embodiment, the neural network-based overhead conductor fatigue strength prediction device further includes: a three-dimensional curved surface display module;

[0163] The three-dimensional surface display module is used to draw the average stress of the conductor to be predicted, the target fatigue life and the stress amplitude predicted by the model into a three-dimensional surface graph.

[0164] It can be understood that the above-mentioned device item embodiments correspond to the method item embodiments of the present invention, and can implement the neural network-based overhead wire fatigue strength prediction method provided by any one of the above-mentioned method item embodiments of the present invention.

[0165] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art may understand and implement the present invention without creative work.

[0166] Based on the above-mentioned embodiment of the method for predicting fatigue strength of overhead wires based on neural network, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for predicting fatigue strength of overhead wires based on neural network according to any embodiment of the present invention is implemented.

[0167] Exemplarily, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0168] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0169] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0170] Based on the above method embodiment, another embodiment is provided: another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the neural network-based overhead wire fatigue strength prediction method described in any one of the above method embodiments of the present invention.

[0171] Wherein, the module / unit integrated in the neural network-based overhead conductor fatigue strength prediction device / terminal equipment can be stored in a computer-readable storage medium if it is implemented in the form of a software functional unit and sold or used as an independent product. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0172] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for predicting fatigue strength of overhead wires based on neural network, characterized in that: include: Obtain the average stress and target fatigue life of the conductor to be predicted; Inputting the average stress and target fatigue life of the conductor to be predicted into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the conductor to be predicted according to the average stress and target fatigue life of the conductor to be predicted; The fatigue strength prediction model is trained with the average stress and fatigue life of the training wire samples as input and the predicted stress amplitude of the training wire samples as output, and the network parameters of the fatigue strength prediction model are adjusted according to the predicted stress amplitude and actual stress amplitude of the training wire samples during the training process.

2. The method for predicting fatigue strength of overhead wires based on neural network according to claim 1, characterized in that: Get the average stress and target fatigue life of the conductor to be predicted, including: Obtain the average tension of the conductor to be predicted and the total cross-sectional area of ​​the conductor; Dividing the average tension of the conductor to be predicted by the total cross-sectional area of ​​the conductor to be predicted to obtain the average stress of the conductor to be predicted; Obtain the target fatigue life of the conductor to be predicted.

3. The method for predicting fatigue strength of overhead wires based on neural network according to claim 1, characterized in that: The preset fatigue strength prediction model is trained in the following manner: Obtain the average stress, different stress amplitudes and fatigue life corresponding to each stress amplitude of several groups of training wire samples; Normalizing all training conductor sample data to obtain the average stress, different stress amplitudes, and fatigue life corresponding to each stress amplitude of the normalized training conductor samples; The average stress of each group of normalized training wire samples, different stress amplitudes, and fatigue life corresponding to each stress amplitude are divided into a training set, a validation set, and a test set according to a preset ratio; The training set and the validation set are input into the fatigue strength prediction model to be trained for iterative training until the model error converges, thereby obtaining the preset fatigue strength prediction model.

4. The method for predicting fatigue strength of overhead wires based on neural network according to claim 3, characterized in that: The training set and the validation set are input into the fatigue strength prediction model to be trained for iterative training until the model error converges, thereby obtaining the preset fatigue strength prediction model, including: Initialize the network parameters, verification cycle number and training parameters of the fatigue strength prediction model; wherein the training parameters include the number of training rounds; In each iterative training, a batch of data is sequentially selected from the training set and input into the input layer of the fatigue strength prediction model; Perform forward calculation on the neurons of each layer according to each input data in the selected batch to generate the final predicted stress amplitude of each input data; Calculating a first loss function according to a final predicted stress amplitude of each input data and a true stress amplitude corresponding to each input data; Adjusting the network parameters of the fatigue strength prediction model to be trained according to the first loss function; Count the number of training rounds; Determine whether the number of training rounds meets the number of verification cycles. If so, input the data of the verification set into the current fatigue strength prediction model to calculate the second loss function; if not, select the next batch of data from the training set and input it into the input layer for iterative training; Determine whether the second loss function converges. If so, use the current fatigue strength prediction model as the preset fatigue strength prediction model; if not, select the next batch of data from the training set and input it into the input layer for iterative training.

5. The method for predicting fatigue strength of overhead conductors based on neural network according to claim 4, characterized in that: For each input data, forward calculation is performed on the neurons of each layer according to each input data in the selected batch to generate the final predicted stress amplitude of each input data, including: For each layer, the input and output of each neuron are calculated by the following formula: z_j^l=Σw_jk^l·a_k^(l-1)+b_j^l; a_j^l=f^l(z_j^l); Wherein, z_j^l represents the input of the jth neuron in the lth layer, a_k^(l-1) represents the output of the kth neuron in the previous layer of the lth layer, w_jk^l represents the connection weight between two neurons, b_j^l represents the bias term of the jth neuron in the lth layer, a_j^l represents the output of the jth neuron in the lth layer, f^l(·) represents the activation function of the lth layer, the activation function of the hidden layer in the fatigue strength prediction model is f(x)=1 / (1+e^(-x)), the activation function of the output layer in the fatigue strength prediction model is g(x)=x, and x represents the input corresponding to the activation function; The output of the output layer is taken as the final predicted stress amplitude of the network.

6. The method for predicting fatigue strength of overhead conductors based on neural network according to claim 5, characterized in that: The network parameters include: a weight matrix and a bias vector; The adjusting the network parameters of the fatigue strength prediction model to be trained according to the first loss function includes: Reversely calculate the partial derivative of each neuron in the output layer and the hidden layer according to the first loss function; Calculating the gradient of the weight matrix and the bias vector according to the partial derivatives of each neuron of the output layer and the hidden layer; The weight matrix and the bias vector are updated according to the gradients of the weight matrix and the bias vector.

7. The method for predicting fatigue strength of overhead conductors based on neural network according to claim 1, characterized in that: Also includes: The average stress of the conductor to be predicted, the target fatigue life, and the stress amplitude predicted by the model are plotted as a three-dimensional surface graph.

8. A neural network-based overhead conductor fatigue strength prediction device, characterized in that: include: A module for acquiring data to be predicted and a module for predicting stress amplitude; The to-be-predicted data acquisition module is used to acquire the average stress and target fatigue life of the conductor to be predicted; The stress amplitude prediction module is used to input the average stress and target fatigue life of the conductor to be predicted into a preset fatigue strength prediction model, so that the fatigue strength prediction model determines the stress amplitude of the conductor to be predicted according to the average stress and target fatigue life of the conductor to be predicted; The fatigue strength prediction model is trained with the average stress and fatigue life of the training wire samples as input and the predicted stress amplitude of the training wire samples as output, and the network parameters of the fatigue strength prediction model are adjusted according to the predicted stress amplitude and actual stress amplitude of the training wire samples during the training process.

9. A terminal device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the neural network-based overhead wire fatigue strength prediction method as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: include: A stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the neural network-based overhead wire fatigue strength prediction method as described in any one of claims 1-7.

Images