Method for predicting steady-state temperature rise of multi-loop tunnel cable

Abstract

The invention relates to a method for predicting steady-state temperature rise of a multi-loop tunnel cable. The method comprises the following steps: acquiring a sample data set of the multi-loop tunnel cable; performing data preprocessing on the acquired sample data set, and constructing a BP neural network model; an HIPSO algorithm of hybrid improved particle swarm optimization is constructed, and the HIPSO algorithm is formed by fusing Sinodal chaotic mapping, dynamic adjustment of inertia weight, T distribution disturbance and Gaussian-Cauchy hybrid variation; optimizing the BP neural network model by using the constructed HIPSO algorithm to obtain an optimal BP neural network model, and establishing an HIPSO-BPNN cable core temperature prediction model; and utilizing the HIPSO-BPNN cable core temperature prediction model to predict the steady-state temperature rise of the multi-loop tunnel cable to obtain a predicted value of the highest temperature of the cable core. Compared with the prior art, the method has the advantages of being more accurate, more efficient, more stable and the like when the HIPSO-BPNN is used for predicting the steady-state temperature rise of the multi-loop tunnel cable.

Description

Technical Field

[0001] The invention relates to the field of cable steady-state temperature rise prediction, and in particular to a method for predicting the steady-state temperature rise of a multi-loop tunnel cable. Background Art

[0002] With the acceleration of urban development, the corresponding power supply problem has gradually become prominent. Compared with other laying forms, tunnel cables have been increasingly widely used in urban power grid construction in recent years due to their larger internal space, convenient personnel management and maintenance, and the ability to reduce the waste of land resources. In order to ensure the safe operation of tunnel cables and avoid accidents, it is necessary to predict the cable core temperature of the cable during steady-state operation. Currently, the commonly used cable core temperature calculation methods are analytical method and numerical calculation method, but both have certain limitations: the analytical method can be better applied to simple cable structures, but it is difficult to extend to complex laying methods; the numerical calculation method has better calculation accuracy, but the calculation takes a long time and requires higher calculation conditions.

[0003] In view of the shortcomings of the above-mentioned methods, the present invention selects to use BP neural network to predict the steady-state temperature of cable core. This method can improve the accuracy of temperature prediction, and the implementation method is relatively quick and simple. The neural network input and output variables involved in multi-loop cables are relatively large, and there are a large number of weights and thresholds between each layer that need to be determined. For this kind of multi-input and multi-output neural network optimization problem, the traditional learning method has poor convergence effect and the optimal solution is difficult to obtain. Therefore, the swarm optimization algorithm is often combined to determine the undetermined parameters in the neural network. Although the single particle swarm algorithm (PSO) shows certain advantages in dealing with multi-objective optimization problems, it is easy to fall into local optimality and has poor search efficiency. Therefore, it is necessary to adopt appropriate optimization strategies to improve it in order to improve its search efficiency and optimization effect.

[0004] After searching, Chinese invention patent publication number CN116956744A discloses a method for predicting steady-state temperature rise of multi-loop grooved cables based on improved particle swarm algorithm, including the following steps: step S1, constructing a two-dimensional physical model of a four-loop grooved cable, using the finite element method to calculate the maximum temperature of the four loop cable cores under different ambient temperatures and constant load currents, and obtaining the required sample data set; step S2, dividing the data set obtained in step S1 into a training set and a test set, and normalizing the input and output data; step S3, using the ambient temperature and the load current of the four loops as input variables, and the maximum temperature of the four loop cable cores as output variables, to construct a BP neural network topology; step S4, constructing an optimized IPSO algorithm model; step S5, using the IPSO algorithm model of step S4 on the training set to determine the weights and thresholds of the BP neural network, and obtaining the BP neural network under the optimal parameters; step S6, using the optimal BP neural network obtained in step S5 to predict the maximum temperature of the four loop cable cores on the test set, and comparing the predicted value with the finite element calculation result to evaluate the accuracy of the prediction result. This existing patent has the problem of inaccurate and inaccurate prediction.

[0005] How to achieve accurate and rapid prediction of the steady-state temperature rise of multi-circuit tunnel cables has become a technical problem that needs to be solved. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for predicting the steady-state temperature rise of a multi-loop tunnel cable.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] According to one aspect of the present invention, a method for predicting the steady-state temperature rise of a multi-loop tunnel cable is provided, the method comprising the following steps:

[0009] Step S1: obtaining a sample data set of a multi-loop tunnel cable;

[0010] Step S2: performing data preprocessing on the sample data set obtained in step S1;

[0011] Step S3: constructing a BP neural network model;

[0012] Step S4: constructing a hybrid improved particle swarm optimization HIPSO algorithm, wherein the HIPSO algorithm is formed by fusing Sinusoidal chaotic mapping, dynamically adjusting inertia weight, T distribution perturbation and Gauss-Cauchy mixed mutation;

[0013] Step S5: Use the HIPSO algorithm constructed in step S4 to optimize the BP neural network model, obtain the optimal BP neural network model, and establish a HIPSO-BPNN cable core temperature prediction model;

[0014] Step S6: using the HIPSO-BPNN cable core temperature prediction model of step S5 to predict the steady-state temperature rise of the multi-loop tunnel cable, and obtain a predicted value of the maximum temperature of the cable core.

[0015] Preferably, the HIPSO algorithm uses Sinusoidal chaotic mapping to initialize the particle population, uses dynamic adjustment of inertia weight to update the particle population velocity, uses T distribution perturbation to update the particle population position, and uses Gaussian-Cauchy mixed mutation to perturb a particle near its historical optimal position.

[0016] More preferably, the method of initializing the particle population using the Sinusoidal chaotic map includes:

[0017] Set the number of particles in the particle group to J, and the range of particle position change to [a i ,b i ], the particle dimension is selected as the total number L of inter-layer weights and node thresholds in the BP neural network, and the maximum number of iterations for iterative updating of particle positions is set, and the Sinusoidal chaotic mapping is used to initialize the particle population:

[0018] X j =(X1,X2,...,X L ),j=1,2...J

[0019] X i+1 =Sinusoidal·X i ·sin(πX i )

[0020] In the formula, X j (j=1,2,…,J) is the position of the jth particle; X i (i=1,2,…,L) is a random number between [0,1], representing the coordinates of the particle in each dimension; L is the dimension of the particle; Sinusoidal is the chaos coefficient.

[0021] More preferably, the method of updating the particle swarm velocity by dynamically adjusting the inertia weight is as follows:

[0022] The formula for updating the particle swarm velocity is:

[0023]

[0024] In the formula, is the velocity of the jth particle at the tth iteration; is the individual historical optimal position of the particle, is the global optimal particle position at the tth iteration; ω t is the dynamic adjustment inertia weight of each particle velocity at the tth iteration, ω max is the maximum inertia weight, ω min is the minimum inertia weight, σ is the inertia adjustment factor, betarnd is a random number that obeys the Beta distribution, p and q are the parameters of the Beta distribution; r1 and r2 are random numbers between [0,1]; c1 and c2 are the individual learning factor improved by the linear decreasing strategy and the social learning factor improved by the linear increasing strategy, respectively, and their expressions are:

[0025]

[0026] Where N is the maximum number of iterations, and t represents the tth iteration.

[0027] More preferably, the updating of particle swarm positions by using T-distribution disturbance includes:

[0028] If the particle position stagnates, then:

[0029]

[0030] otherwise:

[0031]

[0032] Where a is the speed weight factor; θ is the walking step length scaling factor; trnd t is a random number generated at the tth iteration that follows T distribution, is the position of the jth particle at the tth iteration, is the updated position of the jth particle at the t+1th iteration.

[0033] More preferably, the perturbation of a particle near its historical optimal position by using Gaussian-Cauchy mixed variation includes:

[0034] The mutation method is as follows:

[0035]

[0036] Where ω1 and ω2 are the inertia weights of Gaussian distribution and Cauchy distribution respectively; is the individual optimal historical position of the particle; gauss~N(0,1) is an L-dimensional random array that obeys Gaussian distribution; cauchy is an L-dimensional random array that obeys Cauchy distribution, and the distribution function is as follows:

[0037] cauchy=tan[(ξ-0.5)π],ξ~N(0,1)

[0038] If the particle position after mutation Corresponding fitness value Less than the historical optimal position of the particle before mutation The corresponding fitness value Then Update the optimal position and optimal fitness value, otherwise keep them unchanged.

[0039] Preferably, the BP neural network model uses the ambient temperature and the load currents of multiple loops as input variables, and uses the maximum cable core temperature of the multiple loops as output variable.

[0040] Preferably, the optimal BP neural network can minimize the fitness function value obtained on the training set.

[0041] Preferably, the method of obtaining a sample data set of a multi-loop tunnel cable comprises: establishing a physical model of the multi-loop tunnel cable, calculating the maximum temperature of the cable core under different working conditions using a finite element model, and obtaining a sample data set.

[0042] Preferably, the method further comprises comparing the predicted value with the result calculated by the finite element model to evaluate the accuracy of the predicted result, and the evaluation indicators comprise mean absolute error MAE, mean square error MSE and absolute percentage error APE.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1) The present invention integrates Sinusoidal chaotic mapping, dynamic adjustment of inertia weight, T distribution perturbation and Gauss-Cauchy mixed variation, and innovatively proposes a HIPSO algorithm. The BP neural network optimized by the HIPSO algorithm is used to accurately predict the steady-state temperature rise of multi-loop tunnel cables, which provides an effective way to achieve efficient thermal management of tunnel cables. It has certain practical significance in engineering applications, including applications in temperature monitoring, hazard warning and thermal management.

[0045] 2) The HIPSO algorithm proposed in the present invention effectively solves the defects of a single particle swarm algorithm, such as difficulty in parameter selection, poor convergence, and easy to fall into local optimum. Compared with a single improved particle swarm algorithm, the HIPSO algorithm integrates multiple strategies. Compared with other PSO optimization BPNN algorithms, HIPSO has good optimization effects in the entire iteration process. Under the condition of the same accuracy, the sample size required for HIPSO-BPNN is greatly reduced, and the prediction efficiency is greatly improved. At the same time, the number of iterations under the same fitness is the least, and it can quickly stabilize at the beginning of the iteration. The algorithm has great advantages in terms of rapidity. Therefore, the steady-state temperature rise prediction of multi-loop tunnel cables using the HIPSO-BPNN algorithm has higher accuracy, efficiency and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic flow chart of the steady-state temperature rise prediction method in the present invention;

[0047] Figure 2 A schematic diagram of a physical model of an example in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of the BP neural network structure constructed in the present invention;

[0049] FIG4( a ) is a schematic diagram showing a comparison of iteration curves of the present invention and other algorithms at the early stage of iteration in the training process of a data set;

[0050] FIG4( b ) is a schematic diagram of a comparison curve between the present invention and other algorithms in the early stage of the iteration of the data set training process;

[0051] FIG4( c ) is a schematic diagram of a comparison curve between the present invention and other algorithms at the end of the iteration of the data set training process;

[0052] FIG4(d) is a partial enlarged schematic diagram of portion A in FIG4(c) of the present invention;

[0053] Figure 5 It is a schematic diagram of the variation curve of MSE of the present invention and other algorithms with the number of samples;

[0054] Figure 6 It is a schematic flow chart of the steady-state temperature rise prediction method of the present invention;

[0055] In the attached figure, T sur is the surface temperature, T soil is the soil temperature, and h is the convective heat transfer coefficient. DETAILED DESCRIPTION

[0056] 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 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 should fall within the scope of protection of the present invention.

[0057] The present invention integrates multiple optimization strategies, such as Sinusoidal chaotic mapping, dynamic adjustment of inertia weight, T distribution perturbation, Gaussian-Cauchy mixed mutation, etc., to form a new HIPSO algorithm.

[0058] Professional term: Hybrid Improved PSO, referred to as HIPSO, is a hybrid improved particle swarm optimization algorithm that integrates Sinusoidal chaotic mapping, dynamic adjustment of inertia weight, T distribution perturbation, and Gaussian-Cauchy mixed mutation.

[0059] Example 1

[0060] This embodiment relates to a method for predicting the steady-state temperature rise of a multi-circuit tunnel cable. Figure 6 As shown, including:

[0061] Step S1: A finite element model of a multi-circuit tunnel cable is established based on known conditions such as the cable's structural parameters, material properties, laying conditions, and boundary conditions. The maximum temperature of the multi-circuit cable core under different ambient temperatures and load currents is calculated using the finite element model to obtain a sample data set.

[0062] Step S2: preprocess the sample data set obtained in step S1, divide it into a training set and a test set, and normalize the data.

[0063] Step S3: Establish a BP neural network (BPNN) model, select the ambient temperature and the load currents of multiple loops as input variables, and select the maximum temperature of the cable cores of multiple loops as output variables.

[0064] Step S4: Construct the HIPSO algorithm, set the particle dimension, population size, and position change range, use Sinusoidal chaotic mapping to initialize the particle population, use dynamic adjustment of inertia weight to update the particle population velocity, use T distribution perturbation to update the particle population position, and use Gaussian-Cauchy mixed mutation to perturb a particle near its historical optimal position.

[0065] Step S5: Using the HIPSO algorithm constructed in step S4, the weights and thresholds of the BP neural network are optimized on the training set, the BP neural network under the optimal parameters is obtained, and a HIPSO-BPNN cable core temperature prediction model is established.

[0066] Step S6: Using the HIPSO-BPNN cable core temperature prediction model obtained in step S5, the maximum temperatures of multiple loop cable cores are predicted on the test set, and the accuracy of the prediction results is evaluated by comparing the predicted values ​​with the results of finite element calculations.

[0067] In step S1, a constant load current is applied independently to each loop of the multi-loop tunnel cable, and the ambient temperature is variable. The data set consists of a number of data samples, each of which takes the ambient temperature and the load current of multiple loops as input variables, and the maximum cable core temperature of multiple loops as output variables. The heat transfer modes involved in the finite element model include heat conduction, convection, and radiation between surfaces.

[0068] In step S2, the data set is divided into a training set and a test set according to a certain ratio, and the training set data and the test set data are normalized respectively. The normalization formula is as follows:

[0069]

[0070] In the formula, x' is the normalized value; x is the original value in the data set; x min and x max are the minimum and maximum values ​​of a variable in the training set or test set, respectively.

[0071] In step S3, the constructed BP neural network includes an input layer, a hidden layer and an output layer. The parameters to be optimized include the threshold θ of each node. i and the weights W between each layer i .

[0072] In step S4, the number of particles in the particle group is set to J, and the particle position variation range is [a i ,b i ], the particle dimension is selected as the total number L of inter-layer weights and node thresholds in the BP neural network, and the maximum number of iterations for iterative updating of particle positions.

[0073] Use Sinusoidal chaotic mapping to initialize the particle population:

[0074] X j =(X1,X2,...,X L ),j=1,2...J

[0075] X i+1 =Sinusoidal·X i ·sin(πX i )

[0076] Where, X j (j=1,2,…,J) is the position of the jth particle; Xi (i=1,2,…,L) is a random number between [0,1], representing the coordinates of the particle in each dimension; L is the dimension of the particle; Sinusoidal is the chaos coefficient.

[0077] The particle swarm velocity is updated by dynamically adjusting the inertia weight. The update method is as follows:

[0078]

[0079] In the formula, is the velocity of the jth particle at the tth iteration; is the individual historical optimal position of the particle, is the global optimal particle position at the tth iteration; ω t is the dynamic adjustment inertia weight of each particle velocity at the tth iteration, ω max is the maximum inertia weight, ω min is the minimum inertia weight, σ is the inertia adjustment factor, betarnd is a random number that obeys the Beta distribution, N is the maximum number of iterations; p and q are the parameters of the Beta distribution; r1 and r2 are random numbers between [0,1]; c1 and c2 are the individual learning factors improved by the linear decreasing strategy and the social learning factors improved by the linear increasing strategy, respectively, and their expressions are:

[0080]

[0081] The method of updating the particle swarm position using T distribution perturbation is as follows:

[0082] If the particle position stagnates:

[0083]

[0084] otherwise:

[0085]

[0086] Where a is the speed weight factor; θ is the walking step length scaling factor; trnd t is a random number generated at the tth iteration that follows T distribution, is the position of the jth particle at the tth iteration.

[0087] Using Gauss-Cauchy hybrid mutation to find a particle at its historical optimal position The disturbance is carried out nearby, and the mutation is as follows:

[0088]

[0089] Where ω1 and ω2 are the inertia weights of Gaussian distribution and Cauchy distribution respectively; gauss~N(0,1) is an L-dimensional random array that obeys Gaussian distribution; cauchy is an L-dimensional random array that obeys Cauchy distribution, and the distribution function is as follows:

[0090] cauchy=tan[(ξ-0.5)π],ξ~N(0,1)

[0091] If the particle position after mutation Corresponding fitness value Less than the historical optimal fitness value of the particle before mutation Then Update the optimal position and optimal fitness value, otherwise keep them unchanged.

[0092] In step S5, the optimal BP neural network obtained by HIPSO algorithm optimization can minimize the fitness function value obtained on the training set, where the fitness function selects the mean square error MSE, and its expression is:

[0093]

[0094] In the formula, y i,k is the finite element calculated value of the kth output variable of the i-th training sample; is the corresponding BP neural network prediction value; M is the number of training samples; N is the number of output variables.

[0095] In step S6, the neural network optimized by the HIPSO algorithm is applied to predict the maximum temperature of the cable core of the multi-circuit cable in the test set, and the mean absolute error MAE, mean square error MSE, and absolute percentage error APE are selected as indicators for evaluating the accuracy of the neural network prediction results.

[0096] Example 2

[0097] This embodiment also relates to a method for predicting the steady-state temperature rise of a multi-circuit tunnel cable. Figure 1 As shown in the figure, the implementation process includes: establishing an 8-loop tunnel cable physical model, using the finite element model to calculate the maximum temperature of the cable core under different ambient temperatures and load currents, obtaining a sample data set and performing data preprocessing; integrating a single particle swarm algorithm with Sinusoidal chaotic mapping, dynamically adjusting inertia weights, T distribution perturbations, Gaussian-Cauchy mixed mutation and other strategies to form a HIPSO algorithm; constructing a BP neural network, optimizing the weights and thresholds of the neural network through the HIPSO algorithm, and obtaining the optimal BP neural network; and then predicting the steady-state temperature rise of the tunnel cable through the optimal BP neural network to obtain the maximum temperature of the cable core of the 8-loop tunnel cable.

[0098] Figure 2This is a schematic diagram of the physical model established by the present invention based on the actual 8-loop tunnel cable, the surface environment temperature T sur The load currents of the eight loops are taken as variable input variables, and the maximum cable core temperatures of the eight loops under different working conditions are calculated using the finite element model to obtain a sample data set.

[0099] like Figure 3 As shown in the figure, the ambient temperature and the load current of the 8 loops are used as input variables, and the maximum temperature of the cable core of the 8 loops is used as the output variable to construct a BP neural network. The threshold value θ of each node in the BP neural network is i and the weights W of the nodes between layers i The parameters to be optimized are determined by using the HIPSO algorithm proposed in the present invention.

[0100] The HIPSO algorithm is a new hybrid improved particle swarm optimization algorithm that integrates the strategies of Sinusoidal chaotic mapping, dynamic adjustment of inertia weight, T distribution perturbation, Gauss-Cauchy mixed mutation, etc. The HIPSO algorithm uses Sinusoidal chaotic mapping to initialize the particle population, dynamically adjusts the inertia weight to update the particle swarm velocity, uses T distribution perturbation to update the particle swarm position, and uses Gauss-Cauchy mixed mutation to perturb a particle near its historical optimal position.

[0101] like Figure 4(a) to Figure 4(d) As shown in the figure, due to the corresponding improvement strategies adopted in the process of initializing the particle swarm and updating the particle swarm position, the fitness value of the HIPSO-BPNN algorithm is always significantly lower than that of the PSO-BPNN and Levy-PSO-BPNN algorithms throughout the entire iteration process from the initial iteration, the early iteration to the end of the iteration. Compared with the IPSO-BPNN algorithm improved by combining several simple strategies, the fitness value of the HIPSO-BPNN algorithm at the beginning and end of the iteration is better than that of the IPSO-BPNN algorithm. At the same time, through a horizontal comparison with the three algorithms, it can be seen that in the early iteration, the HIPSO-BPNN algorithm tends to be stable the fastest among the four algorithms, and its number of iterations is the least under the same fitness value, indicating that the HIPSO-BPNN algorithm has obvious advantages in rapidity.

[0102] The coordinates of the global optimal particle position in each dimension at the end of the iteration are assigned to the weights and thresholds in the BP neural network to form the optimal BP neural network. The optimal BP neural network is used on the test set to predict the maximum temperature of the 8-loop cable core, and the prediction results of the HIPSO-BPNN algorithm are compared with the PSO-BPNN algorithm, the Levy-PSO-BPNN algorithm, and the IPSO-BPNN algorithm. The method of the present invention uses different evaluation indicators to comprehensively evaluate the accuracy of the prediction results. The smaller the value of the evaluation index, the smaller the deviation between the predicted value and the finite element calculation value, and the higher the prediction accuracy. The evaluation indicators include mean absolute error MAE, mean square error MSE, and absolute percentage error APE. The three indicator formulas are:

[0103]

[0104] The accuracy index comparison results of the HIPSO-BPNN algorithm proposed in the present invention and the PSO-BPNN, Levy-PSO-BPNN, and IPSO-BPNN algorithms are shown in Table 1:

[0105] Table 1

[0106]

[0107]

[0108] As can be seen from Table 1, the mean absolute error MAE, mean square error MSE, and absolute percentage error APE of the HIPSO-BPNN algorithm created by the present invention after optimization are 0.71, 0.85, and 0.21%, respectively. The three accuracy indicators are all at the minimum, indicating that the optimization effect of the HIPSO-BPNN algorithm is better than that of the other three algorithms.

[0109] like Figure 5 As shown in the figure, the changes in the mean square error (MSE) of the four algorithms under different sample numbers are studied. In the figure, the MSE value of the single PSO-BPNN algorithm is the largest, and the MSE value of the Levy flight algorithm has decreased, but it is still at a large value. The MSE values ​​of the IPSO-BPNN algorithm improved by combining several simple strategies and the hybrid improved HIPSO-BPNN algorithm are significantly lower than the above two algorithms. Compared with the two algorithms, the HIPSO-BPNN algorithm requires fewer finite element calculation samples under the same accuracy requirements, and can achieve higher accuracy with fewer data samples.

[0110] In order to study the stability of the algorithm of the present invention and the other three algorithms, independent repeated tests were performed, and the fitness function value after each test was recorded. The mean, standard deviation and variance of the fitness function of the four algorithms were calculated. The results are shown in Table 2:

[0111] Table 2

[0112] algorithm Mean Standard Deviation variance PSO-BPNN 0.0801 0.0098 9.57E-05 Levy-PSO-BPNN 0.0204 0.0090 8.05E-05 IPSO-BPNN 0.0046 0.0009 7.26E-07 HIPSO-BPNN 0.0041 0.0004 1.47E-07

[0113] It can be seen from Table 2 that, under several independent repeated tests, the mean value, standard deviation and variance of the fitness function value calculated by the HIPSO-BPNN algorithm are significantly lower than those of the PSO-BPNN and Levy-PSO-BPNN algorithms. The variance is reduced by 99.85% compared with the Levy-PSO-BPNN algorithm and by 99.82% compared with the PSO-BPNN algorithm. Although it is close to the mean value of the IPSO-BPNN algorithm, it still shows considerable advantages in terms of variance and standard deviation, indicating that the HIPSO-BPNN algorithm proposed in the present invention has the highest stability and stronger anti-interference ability, and can achieve higher prediction accuracy.

[0114] In order to study the effects of different modification strategies on BP neural network optimization in the HIPSO-BPNN algorithm proposed in the present invention, the effects of three modifications on the prediction accuracy of the BP neural network are compared, including hybrid modification (HIPSO), modification of particle swarm position update strategy (PSO+T distribution+Gaussian & Cauchy mixed mutation), and modification of particle swarm initialization strategy (PSO+Sinusoidal chaotic mapping). The accuracy index comparison of the corresponding calculation results is shown in Table 3:

[0115] Table 3

[0116] algorithm MAE MSE APE(%) HIPSO 0.71 0.85 0.21 PSO+T distribution+Gaussian & Cauchy mixed mutation 1.01 1.89 0.30 PSO+Sinusoidal Chaotic Mapping 3.32 17.46 0.97

[0117] As shown in Table 3, in terms of the three accuracy indicators of MAE, MSE and APE, compared with the PSO algorithm combined with multiple optimization strategies, the PSO+T distribution+Gauss & Cauchy mixed mutation and PSO+Sinusoidal chaotic mapping have poor results, indicating that the particle swarm algorithm without combining multiple strategies cannot optimize the BP neural network well, while the improved HIPSO algorithm can better handle the multi-objective optimization problem in the BP neural network. In addition, compared with the optimized initialization particle swarm update method (PSO+Sinusoidal chaotic mapping), the optimized particle swarm position update method (PSO+T distribution+Gauss & Cauchy mixed mutation) has a better optimization effect on the BP neural network, indicating that the position and speed of the particle swarm updated in the process have a greater impact on the algorithm performance, and are greater than the impact of the initialization particle swarm position on the algorithm performance.

[0118] The method of the present invention aims at the application scenario of steady-state temperature rise prediction of multi-circuit tunnel cables, and proposes a hybrid improved particle swarm algorithm (HIPSO) for optimizing BP neural network. The BP neural network optimized by HIPSO algorithm has higher accuracy, stability and rapidity in predicting the steady-state temperature rise of multi-circuit tunnel cables: in terms of accuracy, the three accuracy indicators of MAE, MSE and APE of HIPSO-BPNN prediction results are all the smallest, indicating that its prediction accuracy is the highest; in terms of rapidity, compared with other PSO optimized BPNN algorithms, the HIPSO-BPNN algorithm has a greatly reduced sample size requirement under the same accuracy requirement, and the prediction efficiency is greatly improved. At the same time, the number of iterations under the same fitness is the least, and it can quickly stabilize at the beginning of the iteration, and the algorithm has great advantages in rapidity; in terms of stability, the standard deviation and variance of the MSE value corresponding to the HIPSO-BPNN prediction results under several independent repeated tests are the smallest, indicating that the HIPSO-BPNN algorithm has higher accuracy and stability in predicting the steady-state temperature rise of multi-circuit tunnel cables.

[0119] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting steady-state temperature rise of a multi-circuit tunnel cable, characterized in that: The method comprises the following steps: Step S1: obtaining a sample data set of a multi-loop tunnel cable; Step S2: performing data preprocessing on the sample data set obtained in step S1; Step S3: constructing a BP neural network model; Step S4: constructing a hybrid improved particle swarm optimization HIPSO algorithm, wherein the HIPSO algorithm is formed by fusing Sinusoidal chaotic mapping, dynamically adjusting inertia weight, T distribution perturbation and Gauss-Cauchy mixed mutation; Step S5: Use the HIPSO algorithm constructed in step S4 to optimize the BP neural network model, obtain the optimal BP neural network model, and establish a HIPSO-BPNN cable core temperature prediction model; Step S6: using the HIPSO-BPNN cable core temperature prediction model of step S5 to predict the steady-state temperature rise of the multi-loop tunnel cable, and obtain a predicted value of the maximum temperature of the cable core.

2. A method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 1, characterized in that: The HIPSO algorithm uses Sinusoidal chaotic mapping to initialize particle population, uses dynamic adjustment of inertia weight to update particle population velocity, uses T distribution perturbation to update particle population position, and uses Gaussian-Cauchy mixed mutation to perturb a particle near its historical optimal position.

3. A method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 2, characterized in that: The method of initializing the particle population by using the Sinusoidal chaotic map includes: Set the number of particles in the particle group to J, and the range of particle position change to [a i ,b i ], the particle dimension is selected as the total number L of inter-layer weights and node thresholds in the BP neural network, and the maximum number of iterations for iterative updating of particle positions is set, and the Sinusoidal chaotic mapping is used to initialize the particle population: X j =(X1,X2,...,X L ),j=1,2…J X i+1 =Sinusoidal·X i ·sin(πX i ) Where, X j (j=1,2,…,J) is the position of the jth particle; X i (i=1,2,…,L) is a random number between [0,1], representing the coordinates of the particle in each dimension; L is the dimension of the particle; Sinusoidal is the chaos coefficient.

4. The method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 2, characterized in that: The method of updating the particle swarm speed by dynamically adjusting the inertia weight is as follows: The formula for updating the particle swarm velocity is: In the formula, is the velocity of the jth particle at the tth iteration; is the individual historical optimal position of the particle, is the global optimal particle position at the tth iteration; ω t is the dynamic adjustment inertia weight of each particle velocity at the tth iteration, ω max is the maximum inertia weight, ω min is the minimum inertia weight, σ is the inertia adjustment factor, betarnd is a random number that obeys the Beta distribution, p and q are the parameters of the Beta distribution; r1 and r2 are random numbers between [0,1]; c1 and c2 are the individual learning factor improved by the linear decreasing strategy and the social learning factor improved by the linear increasing strategy, respectively, and their expressions are: Where N is the maximum number of iterations, and t represents the tth iteration.

5. The method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 2, characterized in that: The method of updating the particle swarm position by using T-distribution disturbance includes: If the particle position stagnates, then: otherwise: Where a is the speed weight factor; θ is the walking step length scaling factor; trnd t is a random number generated at the tth iteration that follows T distribution, is the position of the jth particle at the tth iteration, is the updated position of the jth particle at the t+1th iteration.

6. A method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 2, characterized in that: The perturbation of a particle near its historical optimal position by using Gauss-Cauchy mixed variation includes: The mutation method is as follows: Where ω1 and ω2 are the inertia weights of Gaussian distribution and Cauchy distribution respectively; is the individual optimal historical position of the particle; gauss~N(0,1) is an L-dimensional random array that obeys Gaussian distribution; cauchy is an L-dimensional random array that obeys Cauchy distribution, and the distribution function is as follows: cauchy=tan[(ξ-0.5)π],ξ~N(0,1) If the particle position after mutation Corresponding fitness value Less than the historical optimal position of the particle before mutation The corresponding fitness value Then Update the optimal position and optimal fitness value, otherwise keep them unchanged.

7. A method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 1, characterized in that: The BP neural network model uses the ambient temperature and the load currents of multiple loops as input variables, and uses the maximum cable core temperature of the multiple loops as output variables.

8. The method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 1, characterized in that: The optimal BP neural network can minimize the fitness function value obtained on the training set.

9. The method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 1, characterized in that: The method of obtaining a sample data set of a multi-loop tunnel cable includes: establishing a physical model of the multi-loop tunnel cable, calculating the maximum temperature of the cable core under different working conditions using a finite element model, and obtaining a sample data set.

10. A method for predicting steady-state temperature rise of a multi-circuit tunnel cable according to claim 1, characterized in that: The method further comprises comparing the predicted value with the result calculated by the finite element model to evaluate the accuracy of the predicted result, wherein the evaluation indicators comprise mean absolute error MAE, mean square error MSE and absolute percentage error APE.

Images