Method for predicting bridge cable force through machine learning

Abstract

The invention provides a method for predicting bridge cable force through machine learning, and the method comprises the following steps: building a BP neural network model, determining an input value and an output value of a BP neural network according to a method flow of testing bridge cable force through a frequency method, and determining functions between each hidden layer and the input and output layers of the BP neural network; optimizing the BP neural network by using a particle swarm intelligent algorithm, and optimizing an initial weight and an offset value of the neural network; training the model by using the training group data, adjusting the hyper-parameters of the model in the training process, and setting the learning rate, training iteration times, target and gradient of the BP neural network, the number of neurons and the number of hidden layers; and testing by applying the data of the test group, and importing the characteristic parameters of the inhaul cable as neural network input values into an input layer. On the basis of an artificial intelligence algorithm, a proper algorithm is determined by processing cable force vibration data collected on site.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge cable force calculation, and in particular, to a method for predicting bridge cable force using machine learning. Background Art

[0002] In the existing specifications, the method of identifying cable force through the cable force calculation formula based on the frequency method has the defects of uncertain value of cable calculation length and inaccurate definition of cable boundary conditions. The present invention abandons the disadvantage that it is difficult to accurately determine the cable length in the traditional cable force calculation formula, uses measured data to train the neural network, and combines another intelligent algorithm for parameter optimization, so as to obtain cable force prediction data efficiently and accurately. Summary of the invention

[0003] In view of the technical problems mentioned in the above background technology, a method for predicting bridge cable force using machine learning is provided.

[0004] The technical means adopted by the present invention are as follows:

[0005] A method for predicting bridge cable force using machine learning, comprising the following steps:

[0006] Step 1: Establish a BP neural network model, determine the input and output values ​​of the BP neural network according to the method flow of testing the bridge cable force by the frequency method, and determine the functions between each hidden layer and the input and output layers of the BP neural network;

[0007] Step 2: Optimize the BP neural network by using the particle swarm intelligence algorithm to optimize the initial weights and bias values ​​of the neural network;

[0008] Step 3: Divide the measured bridge cable force data into a training group and a test group, and use the training group data to train the model. During the training process, adjust the model's hyperparameters, set the BP neural network's learning rate, training iterations, target and gradient, as well as the number of neurons and hidden layers;

[0009] Step 4: Use the actual measured bridge test data to conduct the test, in which the characteristic parameters of the cables are imported into the input layer as the neural network input values.

[0010] Furthermore, in step 1, after the BP neural network inputs the data to be predicted, it is necessary to first perform normalization processing on the data to be predicted.

[0011] Furthermore, the normalization process refers to selecting a nonlinear activation function TanH function between the input layer and the hidden layer and between the hidden layers to limit the data range to [-1, 1].

[0012] Furthermore, in step 2, optimizing the BP neural network includes the following steps:

[0013] Step 21: Initialize the position and velocity of each particle; set the individual optimal position P of each particle i and the group's best position P g ;

[0014] Step 22: Update and optimize particles;

[0015] Step 23: The BP neural network calculates the output through forward propagation and calculates the error E;

[0016] Step 24: Determine whether the error E is in compliance;

[0017] If the error E is greater than or equal to the target error, the weights are updated by the gradient descent algorithm to update the connection weights from the hidden layer to the output layer; if the error E is less than the set target error, or reaches the set number of iterations, execute step 25;

[0018] Step 25: Perform data prediction, and output the obtained data to a pre-set Excel table through denormalization.

[0019] Furthermore, in step 21, the position of each particle i is expressed as:

[0020] X i =(x i1 ,x i2 ,...,x id );

[0021] Where d represents the dimension of the problem; x id represents the position coordinates of particle i in the dth dimension, that is, particle position X i A component in a vector;

[0022] The velocity of each particle is expressed as:

[0023] V i =(v i1 ,v i2 ,...,v id ).

[0024] Among them, v i1 Represents the position coordinates of particle i in the dth dimension, that is, the particle velocity vector V i A component of .

[0025] Furthermore, in step 22, the particle updating process is as follows:

[0026] V i (t+1)=w·c1·rand1·Pi (t)-X i (t)+c2·rand2·P g (t)-X i (t);

[0027] X i (t+1)=X i (t)+V i (t+1);

[0028] Where w represents the inertia weight, c1 and c2 represent the acceleration coefficient, rand1 and rand2 represent random numbers; P i (t) represents the optimal position of the individual particle at time t, P g (t) represents the global optimal position of the particle at time t.

[0029] Furthermore, the calculation formula of the error E is:

[0030]

[0031] Among them, d_k represents the expected output and y_k represents the output of the kth neuron.

[0032] Furthermore, in step 24, the weights are updated by the gradient descent algorithm, and the connection weights from the hidden layer to the output layer are:

[0033]

[0034] Among them, w_jk represents the connection weight from the hidden layer to the output layer, g_j represents the output of the jth neuron in the hidden layer, and f represents the activation function;

[0035] The connection weight from the input layer to the hidden layer is:

[0036]

[0037] Among them, vi_ij represents the connection weight between the input layer and the hidden layer, and the j input samples are set to x_j;

[0038] The weight update is:

[0039]

[0040] Among them, η represents the learning rate.

[0041] Furthermore, the set number of iterations is 2000 times.

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] The present invention is based on an artificial intelligence algorithm. By processing the cable force vibration data collected on site, a suitable algorithm is determined, and the algorithm parameters are optimized. The data of the cable force tested by the vibration frequency method are analyzed and processed, so that the calculated average absolute percentage error of the cable force is not higher than 5%. In addition, the calculation accuracy of the final algorithm obtained is higher than the formula calculation method recommended in Appendix B of the "Highway Bridge Load Test Code" (JTG / T J21-01-2015). BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0046] Figure 2 This is a flowchart of the neural network workflow of the present invention.

[0047] Figure 3 This is a schematic diagram of the absolute error percentage of the predicted cable force according to the present invention. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0049] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0050] like Figure 1 , Figure 2 As shown, the present invention provides a method for predicting bridge cable force using machine learning, comprising the following steps:

[0051] Step 1: Establish a BP neural network model, determine the input value and output value of the BP neural network according to the method flow of testing the bridge cable force by the frequency method, and determine the functions between each hidden layer and the input and output layers of the BP neural network; the BP neural network, when inputting the data to be predicted, needs to first perform normalization processing on the data to be predicted. The normalization processing refers to selecting the nonlinear activation function TanH function between the input layer and the hidden layer and between the hidden layers, and limiting the data range to [-1,1].

[0052] Step 2: By using the particle swarm intelligence algorithm, the BP neural network is optimized, and the initial weights and bias values ​​of the neural network are optimized; the optimization of the BP neural network includes the following steps:

[0053] Step 21: Initialize the position and velocity of each particle; set the individual optimal position P of each particle i and the group's best position P g ; In step 21, the position of each particle i is expressed as:

[0054] X i =(x i1 ,x i2 ,...,x id );

[0055] Where d represents the dimension of the problem; x id represents the position coordinate of particle i in the dth dimension, which is the particle position X i A component of a vector.

[0056] The velocity of each particle is expressed as:

[0057] V i =(v i1 ,v i2 ,...,v id ).

[0058] Among them, v i1 Represents the position coordinates of particle i in the dth dimension, which is the particle velocity vector V i A component of .

[0059] Step 22: Update and optimize the particles; the particle update process is:

[0060] V i (t+1)=w·c1·rand1·P i (t)-Xi (t)+c2·rand2·P g (t)-X i (t);

[0061] X i (t+1)=X i (t)+V i (t+1);

[0062] Among them, w represents the inertia weight, c1 and c2 represent the acceleration coefficient, and rand1 and rand2 are random numbers. i (t) represents the optimal position of the individual particle at time t, P g (t) represents the global optimal position of the particle at time t.

[0063] Step 23: The BP neural network calculates the output through forward propagation and calculates the error E; the calculation formula of the error E is:

[0064]

[0065] Among them, d_k represents the expected output and y_k represents the output of the kth neuron.

[0066] Step 24: Determine whether the error E is in compliance;

[0067] If the error E is greater than or equal to the target error, the weight is updated by the gradient descent algorithm, and the connection weight from the hidden layer to the output layer is:

[0068]

[0069] Among them, w_jk is the connection weight from the hidden layer to the output layer, g_j is the output of the jth neuron in the hidden layer, and f is the activation function.

[0070] The connection weight from the input layer to the hidden layer is:

[0071]

[0072] Among them, vi_ij is the connection weight between the input layer and the hidden layer, and the j input samples are set to x_j.

[0073] The weight update is:

[0074]

[0075] Wherein, η is the learning rate, and in the present invention, the learning rate is equal to 0.01.

[0076] If the error E is less than the set target error, or reaches the set number of iterations (the number of iterations is 2000), execute step 25;

[0077] Step 25: Perform data prediction, and output the obtained data to a pre-set Excel table through denormalization.

[0078] Step 3: Divide the measured bridge cable force data into a training group and a test group, and use the training group data to train the model. During the training process, adjust the model's hyperparameters, set the BP neural network's learning rate, training iterations, target and gradient, as well as the number of neurons and hidden layers;

[0079] Step 4: Use the actual measured bridge test data to conduct the test, in which the characteristic parameters of the cables are imported into the input layer as the neural network input values.

[0080] Example 1

[0081] like Figure 3 As shown in the figure, the average absolute error percentage between the cable force of the suspender predicted by the present invention and the cable force actually measured by the hydraulic jack. It can be seen that the error of the formula calculation method recommended in Appendix B of the "Highway Bridge Load Test Code" (JTG / TJ21-01-2015) is 9.7%,

[0082] The absolute error of the BP neural network prediction is 7.9%, and the absolute error percentage of the PSO-BP neural network after particle swarm intelligence algorithm optimization is 2.8%, showing good accuracy and engineering practical value.

[0083] Example 2

[0084] The Luoyang East Ring Bridge was applied according to the preset scheme to test the practicability of the present invention. The East Ring Road Bridge is located in Luoyang City, Henan Province. The main bridge is 320 meters long and a total of 4x25=100 hangers are arranged on the whole bridge. The frequency and actual cable force values ​​of 200 groups of hangers in the 50% and fully tensioned states of the hangers were measured by hydraulic jacks and wireless cable tension testers.

[0085] Step 1: Establish a neural network model and optimize it using the particle swarm intelligence algorithm. According to string vibration theory, the explicit relationship between the cable force and its natural frequency is:

[0086]

[0087] Wherein, T is the cable force, m represents the cable linear density, L represents the cable length, f represents the cable natural frequency, and n represents the frequency order. Therefore, in the present invention, the linear density, cable length and natural frequency of the cable are used as the input values ​​of the neural network. The TanH function is selected between the input layer and the hidden layer and between the hidden layers of the neural network established by the present invention; the Purelin function is selected between the hidden layer and the output layer. In the particle swarm algorithm, the optimal solution is found by simulating the search process of particles in the solution space.

[0088] Step 2: Randomly select 149 sets of hanger data, of which 100 sets are used as training sets and 39 sets are used as test sets. The measured cable data of the bridge are input to train the two neural network models and adjust the model hyperparameters. The measured cable length, linear density, frequency and cable force are respectively brought into the input and output of the neural network. The learning rate of the neural network is set to 0.01, the training iteration is set to 2000 times, and the target error is set to 1×10 -6 , the gradient is 1×10 -7 , determine the model of the neural network is 3-4-6-1.

[0089] Step 3: Input the cable parameters to be predicted into the input end of the trained neural network model to predict the cable force data. And output the scatter plot of the predicted data through machine language.

[0090] Step 4: Analyze the predicted data and compare them with the actual cable data obtained by the jack. The average absolute error percentage between the cable force predicted by the BP neural network and the actual cable force value is 7.93%, and the average absolute error percentage predicted by the PSO-BP neural network is 2.78%. The accuracy of the prediction method of the invention is verified to be better than the formula calculation method recommended in Appendix B of the "Highway Bridge Load Test Code" (JTG / T J21-01-2015).

[0091] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0092] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0093] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0094] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0095] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0096] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting bridge cable forces using machine learning, characterized in that: The following steps are involved: Step 1: Establish a BP neural network model, determine the input and output values ​​of the BP neural network according to the method flow of testing the bridge cable force by the frequency method, and determine the functions between each hidden layer and the input and output layers of the BP neural network; Step 2: Optimize the BP neural network by using the particle swarm intelligence algorithm to optimize the initial weights and bias values ​​of the neural network; Step 3: Divide the measured bridge cable force data into a training group and a test group, and use the training group data to train the model. During the training process, adjust the model's hyperparameters, set the BP neural network's learning rate, training iterations, target and gradient, as well as the number of neurons and hidden layers; Step 4: Use the actual measured bridge test data to conduct the test, in which the characteristic parameters of the cables are imported into the input layer as the neural network input values.

2. The method for predicting bridge cable force using machine learning according to claim 1, characterized in that: In step 1, the BP neural network needs to normalize the data to be predicted after inputting the data to be predicted.

3. The method for predicting bridge cable force using machine learning according to claim 1, characterized in that: The normalization process refers to selecting a nonlinear activation function TanH function between the input layer and the hidden layer and between the hidden layers to limit the data range to [-1, 1].

4. The method for predicting bridge cable force using machine learning according to claim 1, characterized in that: In step 2, optimizing the BP neural network includes the following steps: Step 21: Initialize the position and velocity of each particle; set the individual optimal position P of each particle i and the group's best position P g ; Step 22: Update and optimize particles; Step 23: The BP neural network calculates the output through forward propagation and calculates the error E; Step 24: Determine whether the error E is in compliance; If the error E is greater than or equal to the target error, the weights are updated by the gradient descent algorithm to update the connection weights from the hidden layer to the output layer; if the error E is less than the set target error, or reaches the set number of iterations, execute step 25; Step 25: Perform data prediction, and output the obtained data to a pre-set Excel table through denormalization.

5. The method for predicting bridge cable force using machine learning according to claim 4, characterized in that: In step 21, the position of each particle i is expressed as: X i =(x i1 ,x i2 ,...,x id ); Where d represents the dimension of the problem; x id represents the position coordinates of particle i in the dth dimension, that is, particle position X i A component in a vector; The velocity of each particle is expressed as: V i =(v i1 ,v i2 ,...,v id )。 Among them, v i1 Represents the position coordinates of particle i in the dth dimension, that is, the particle velocity vector V i A component of .

6. The method for predicting bridge cable force using machine learning according to claim 4, characterized in that: In step 22, the particle updating process is as follows: V i (t+1)=w·c1·rand1·P i (t)-X i (t)+c2·rand2·P g (t)-X i (t); X i (t+1)=X i (t)+V i (t+1); Where w represents the inertia weight, c1 and c2 represent the acceleration coefficient, rand1 and rand2 represent random numbers; P i (t) represents the optimal position of the individual particle at time t, P g (t) represents the global optimal position of the particle at time t.

7. The method for predicting bridge cable force using machine learning according to claim 4, characterized in that: The calculation formula of the error E is: Among them, d_k represents the expected output and y_k represents the output of the kth neuron.

8. The method for predicting bridge cable force using machine learning according to claim 4, characterized in that: In step 24, the weights are updated by the gradient descent algorithm, and the connection weights from the hidden layer to the output layer are: Among them, w_jk represents the connection weight from the hidden layer to the output layer, g_j represents the output of the jth neuron in the hidden layer, and f represents the activation function; The connection weight from the input layer to the hidden layer is: Among them, vi_ij represents the connection weight between the input layer and the hidden layer, and the j input samples are set to x_j; The weight update is: Among them, η represents the learning rate.

9. The method for predicting bridge cable force using machine learning according to claim 4, characterized in that: The set number of iterations is 2000.

Images