A method for predicting lateral / vertical smoothness of a high-speed maglev train carriage

Abstract

The present application relates to a kind of for high-speed maglev train carriage transverse / vertical smoothness prediction method, comprising the following steps: S1, extract transverse and vertical vibration information at different speeds with predetermined sampling frequency, and carry out transverse and vertical smoothness index calculation;S2, with different speed domain, the transverse and vertical smoothness index calculation result in step S1 is divided into training sample and test sample at set interval;S3, construct Elman neural network prediction model;S4, using learning index function according to the performance evaluation of the prediction model network constructed according to weight change;S5, according to the weight change situation, algorithm optimization is carried out in combination with setting algorithm, and the weight is adjusted;S6, input training sample carries out model training, and algorithm verification is carried out using test sample, and the transverse and vertical smoothness index are predicted, and the train smoothness grade is obtained by referring to GB5599-2019 standard.Compared with prior art, the present application has the advantages of improving prediction accuracy, fast operation speed and the like.

Description

Technical Field

[0001] This invention relates to the field of maglev trains, and in particular to a method for predicting the lateral / vertical stability of high-speed maglev train carriages. Background Technology

[0002] Maglev trains operate without any mechanical contact between the train and rails except for the power supply shoe contact, reducing mechanical wear and friction noise, resulting in significant environmental advantages and gradually becoming a new type of rail transportation. High-speed maglev trains can effectively overcome the speed limits imposed by wheel-rail contact, achieving higher operating speeds with lower energy consumption. Currently, the development of maglev trains exhibits multi-system characteristics, which can be divided into three main categories based on different levitation principles: conventional electromagnetic levitation (EMS), electric levitation (EDS), and hybrid levitation combining some features of EMS and EDS. Among these, the development of EMS maglev trains is the most mature, and they have already achieved engineering operation in Shanghai, Beijing, Changsha, and other cities. EDS maglev trains are relatively mature in China. In Japan, a low-temperature superconducting maglev test vehicle and a supporting test line (Yamanashi Line) have been built. In China, Southwest Jiaotong University has built the world's first high-temperature superconducting high-speed maglev test vehicle, but due to the short test line, it is still under in-depth research. Regarding the third type of hybrid levitation technology, current research lacks the conditions for engineering implementation and remains in the laboratory research stage. However, the high-speed maglev train that has achieved commercial operation and the highest speed is the Shanghai Maglev Train, which uses electromagnetic levitation (EMS) and has a commercial operating speed of 430 km / h.

[0003] Due to the relatively high technological maturity of EMS-type high-speed maglev trains, further speed increases based on existing conventional electromagnetic levitation technology are technically more feasible. However, as a type of ground-based rail transit, it is inevitably affected by complex ground conditions, especially the superposition of complex disturbances at higher speeds. This poses a greater challenge to the train's contactless levitation performance and necessitates further consideration of passenger comfort. Therefore, the issue of carriage stability must be considered when considering further speed increases. Currently, due to the lack of sufficiently long track lines to verify train stability at higher speeds, it is impossible to verify the lateral / vertical stability of the carriages during high-speed operation through actual running tests. Therefore, methods for predicting stability at high speeds are of significant engineering importance for verifying the feasibility of further speed increases. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for predicting the lateral / vertical stability of high-speed maglev train carriages.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for predicting the lateral / vertical stability of a high-speed maglev train car includes the following steps:

[0007] S1. Extract lateral and vertical vibration information at different speeds from the carriage vibration sensor at a predetermined sampling frequency, and calculate the lateral and vertical stability index.

[0008] S2. Divide the calculation results of the lateral and vertical stability indices in step S1 into training samples and test samples according to different speed domains at set intervals.

[0009] S3. Construct an Elman neural network prediction model. After normalizing the training samples, input them into the Elman neural network prediction model to reduce the adverse effects caused by singular sample data.

[0010] S4. Use the learning index function to evaluate the performance of the constructed Elman neural network prediction model based on the changes in weights;

[0011] S5. Optimize the algorithm based on the changes in weights and adjust the weights to avoid getting stuck in local optima or slow training speed during the weight update process.

[0012] S6. The algorithm is verified using test samples to predict the lateral and vertical stability indices, and the train stability level is obtained with reference to the GB5599-2019 standard.

[0013] Furthermore, in step S1, the expression for calculating stationarity is:

[0014]

[0015]

[0016] Among them, i=1, 2,...,n, A i For vibration acceleration, f i Let F(f) be the vibration frequency. i ) represents the frequency correction factor, W i For a frequency of f i The stability index component over time.

[0017] Furthermore, in step S2, the training samples are the stability calculation results at speeds of 0 km / h, 10 km / h, 20 km / h, ..., 300 km / h; the test samples are the stability calculation results at speeds of 310 km / h, 320 km / h, ..., 430 km / h.

[0018] Furthermore, in step S4, 0.5 times the squared error function is used as the learning index function to evaluate the performance of the current network model under the conditions of changes in the input layer-hidden layer weights, the connecting layer-hidden layer weights, and the hidden layer-output layer weights.

[0019] Furthermore, in step S5, the algorithm is set to Particle Swarm Optimization (PSO) and Gravity Search Algorithm (GSA).

[0020] Furthermore, the particle swarm optimization algorithm first optimizes the gravity search algorithm by considering the information interaction characteristics between massive particles, thus obtaining the optimized particle search movement speed:

[0021]

[0022] in, Let μ be the velocity of particle i in the d-th dimension considering the PSO-GSA algorithm, μ be the weighting function, gb(t) be the global optimal position, and κ be the velocity of particle i. PG1 and κ PG2 As a weighting factor, Let be the quotient of the net force acting on particle i in the d-th dimension and the inertial mass of particle i. Let i be the current position of particle i in the d-th dimension, and rand(0,1) be a random number in the interval (0,1). The optimization results are used as the weight optimization path for the Elman model for predicting the stability of maglev trains.

[0023] Furthermore, the particle's optimization movement speed, and its position, are represented as follows:

[0024]

[0025] The particle position at time t+1 is obtained and regarded as the ideal weight at time t+1.

[0026] Furthermore, in step S5, the calculated weights are substituted into the Elman model used for predicting the stability of high-speed maglev trains to calculate the hidden layer output and the output layer output, respectively:

[0027] x(k)=f(ω (1) x(k-1)+ω (2) u(k-1))

[0028] y(k)=g(ω (3) x(k))

[0029] Where x(k) is the hidden layer output, ω (1) ω (2) ω (3)These are the weight vectors of the receiving layer-hidden layer, the input layer-hidden layer, and the hidden layer-output layer, respectively. y(k) is the output of the output layer, which is the prediction result.

[0030] Furthermore, the lateral and vertical vibration information refers to the lateral / vertical vibration acceleration data of the train carriage. Acceleration sensors are respectively arranged at the ends, middle, and rear of the carriage to measure the lateral / vertical vibration of the carriage.

[0031] Furthermore, the lateral and vertical stability indices in step S1 are calculated using the relevant definitions of stability indices in the GB5599-2019 standard.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The Elman neural network prediction model constructed in this invention, compared with traditional methods for measuring train stability, has been optimized by the algorithm and can make a preliminary estimate of the stability of the carriage at higher operating speeds under conditions where there is no actual train test.

[0034] 2. This invention optimizes the GSA algorithm by using the PSO algorithm, which can improve the global optimization speed of the GSA algorithm and make it less likely to get trapped in local optima. This provides an algorithmic basis for the constructed Elman neural network prediction model and accelerates the prediction of the lateral / vertical stability index of the carriage.

[0035] 3. This invention optimizes the Elman neural network prediction model using the PSO-GSA algorithm. The optimization result serves as the weight optimization path for the Elman neural network model for predicting the stability of maglev trains. Compared with the previous technology that used gradient descent to adjust the weights, this invention can obtain more accurate weights, thereby improving prediction accuracy. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the process of the present invention.

[0037] Figure 2 This is a schematic diagram of the basic prediction structure of the Elman network involved in this invention.

[0038] Figure 3 This is a schematic diagram of the PSO-GSA algorithm optimization process involved in this invention.

[0039] Figure 4 This is a schematic diagram of the PSO-GSA-Elman algorithm involved in this invention.

[0040] Figure 5This is a schematic diagram of the weight correction process of the PSO-GSA-Elman algorithm involved in this invention. Detailed Implementation

[0041] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0042] like Figure 1 The diagram shown illustrates the process of this invention, a method for predicting the lateral / vertical stability of a high-speed maglev train carriage, comprising the following steps:

[0043] S1. Extract lateral and vertical vibration information at different speeds from the carriage vibration sensor at a predetermined sampling frequency, and calculate the lateral and vertical stability index.

[0044] S2. Divide the calculation results of the lateral and vertical stability indices in step S1 into training samples and test samples according to different speed domains at set intervals.

[0045] S3. Construct an Elman neural network prediction model. Normalize the training samples and then input them into the Elman neural network prediction model.

[0046] S4. Use the learning index function to evaluate the performance of the constructed Elman neural network prediction model based on the changes in weights;

[0047] S5. Optimize the algorithm based on the changes in weights and the set algorithm, and adjust the weights accordingly;

[0048] S6. The algorithm is verified using test samples to predict the lateral and vertical stability indices, and the train stability level is obtained with reference to the GB5599-2019 standard.

[0049] like Figure 2 As shown, to obtain raw data, sensors were installed at the front, middle, and rear of the carriage. Lateral and longitudinal vibration data were collected at predetermined frequencies under actual operating speeds ranging from 0 km / h to 430 km / h. The speed domain was divided into units of 10 km / h. After division, stability indices were calculated for the lateral and vertical vibration data in different speed domains based on the GB5599-2019 standard. The training samples (stability indices at speeds of 0 km / h, 10 km / h, 20 km / h, ..., 300 km / h) were normalized and used as... Figure 2The Elman neural network prediction model input vector is shown. In addition, the number of neurons, training functions f(·) and g(·), number of iterations, error tolerance, and initial weights are defined to obtain the basic structure of the prediction method involved in this paper.

[0050] like Figure 3 As shown, the PSO-GSA algorithm is used to optimize the weights of the Elman neural network prediction model. The expression for the optimization speed (particle movement speed) is:

[0051]

[0052] in, The net external acceleration shown is the quotient of the net force acting on particle i in the d-th dimension and the inertial mass of particle i. The net force acting on particle i in the d-th dimension... The calculation expression is:

[0053]

[0054] in, Let represent the interaction force between particle i and particle j in the d-th dimension.

[0055] Inertial mass can be calculated based on fitness. A higher fitness level results in a larger particle mass, making it more likely to attract other particles to its location. Therefore, it can be understood that particles with larger masses are closer to the optimal solution in the search space. The formula for calculating particle inertial mass is:

[0056] M GSA_imp_i (t)=ω m_adj_i (t)M GSA_i (t)

[0057] Where, ω m_adj_i (t) The inertial mass adjustment weight of particle i, M GSA_i (t) represents the standard mass of particle i.

[0058] The Elman network weights are adjusted based on the obtained optimal weights (optimal particle positions), and then... Figure 4 The output is updated according to the process shown. In addition, the performance of the network model is evaluated by using 0.5 times the squared error function as the learning index function to determine whether the system error meets the set conditions.

[0059] like Figure 5 As shown, the error defined within the PSO-GSA algorithm can be used to judge the obtained weights (particle positions). If it is the global optimal solution, the weights (particle positions) are adjusted. If it is not the optimal solution, the iteration continues until the stopping condition is met.

[0060] The working principle of this embodiment is as follows:

[0061] The Elman neural network is input with stationarity index data in the speed range of 0 km / h to 310 km / h from the training sample set. The data is calculated using the hidden layer f(·) function, then fed back to the hidden layer for further calculation. The result is then processed by the nonlinear function g(·) before being output by the output layer. The weights for f(·) and g(·) are optimized using the PSO-GSA algorithm, and network performance is evaluated using a factor of 0.5 (squared error). If the performance evaluation criteria are not met, the weights are continuously updated, and the network is continuously trained based on these updated weights until the performance requirements or other stopping conditions are met. After training, stationarity index data in the speed range of 310 km / h to 430 km / h from the test sample set are used for comparison. If the goodness of fit meets the requirements, the network performance is considered good and can be used to predict the lateral / vertical stationarity of a train carriage at higher speeds.

[0062] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for predicting lateral / vertical smoothness of a high-speed maglev train car, characterized in that, The method comprises the following steps: S1, extracting lateral and vertical vibration information at different speeds from the car vibration sensor at a predetermined sampling frequency, and calculating lateral and vertical smoothness indexes; S2, dividing the calculation results of the lateral and vertical smoothness indexes in step S1 into training samples and test samples according to different speed domains at a set interval; S3, constructing an Elman neural network prediction model, and inputting the normalized training samples into the Elman neural network prediction model; S4, using a learning index function to evaluate the performance of the constructed Elman neural network prediction model according to the weight change; S5, optimizing the algorithm according to the weight change and combining a set algorithm, and adjusting the weight; S6, verifying the algorithm using test samples, predicting the lateral and vertical smoothness indexes, and referring to the GB5599-2019 standard to obtain the train smoothness level; The set algorithm in step S5 is a particle swarm optimization algorithm and a gravitational search algorithm, the particle swarm optimization algorithm first optimizes the gravitational search algorithm, considers the information interaction characteristics between particles with mass, and obtains the optimized particle optimization moving speed: wherein, is the moving speed of particle i considering PSO-GSA algorithm in the dth dimensional space, μ is a weighting function, gb(t) is the global optimal position, κ PG1 and κ PG2 is a weighting factor, is the quotient of the resultant force experienced by particle i in the dth dimensional space and the inertial mass of particle i, is the current position of particle i in the dth dimensional space, rand(0,1) is a random number in the interval (0,1), and the optimization result is sequentially optimized as the weight optimization path of the maglev train stability prediction Elman model.

2. A method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, characterized in that, In step S1, the smoothness calculation expression is: where i = 1, 2,..., n, A i is the vibration acceleration, f i is the vibration frequency, F(f i ) is the frequency correction coefficient, W i is the stationary index component when the frequency is f i .

3. A method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, characterized in that, In step S2, the training samples are the smoothness calculation results at speeds of 0 km / h, 10 km / h, 20 km / h, …, and 300 km / h; and the test samples are the smoothness calculation results at speeds of 310 km / h, 320 km / h, …, and 430 km / h.

4. The method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, characterized in that, In step S4, the 0.5 times of the error square function is used as the learning index function to evaluate the performance of the current network model according to the changes of the input layer-implicit layer weight, the receiving layer-implicit layer weight, and the implicit layer-output layer weight.

5. The method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, wherein, The position of the particle optimization moving speed is represented as: The particle position at t+1 is obtained, which is regarded as the ideal weight at t+1.

6. The method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, wherein, In step S5, the obtained weight is brought into the Elman model for high-speed maglev train smoothness prediction, and the implicit layer output and the output layer output are calculated respectively: x(k) = f(ω (1) x(k-1) + ω (2) u(k-1)) y(k) = g(ω (3) x(k)) where x(k) is the output of the hidden layer, ω (1) , ω (2) , and ω (3) are the weight vectors of the receiving layer-hidden layer, input layer-hidden layer, and hidden layer-output layer, respectively, and y(k) is the output of the output layer, i.e., the prediction result.

7. The method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, wherein, The lateral and vertical vibration information is train car lateral / vertical vibration acceleration data, and acceleration sensors are arranged at the car end, the car middle, and the car tail to measure the car lateral / vertical vibration.

8. The method for predicting lateral / vertical smoothness of a high-speed maglev train car according to claim 1, wherein, In step S1, the lateral and vertical smoothness indexes are calculated according to the related definitions of the smoothness indexes in the GB5599-2019 standard.

Images