Asymmetric suspension bridge empty cable shape error prediction method and device and medium

Abstract

The invention relates to an asymmetric suspension bridge empty cable alignment error prediction method and device and a medium, and the method comprises the following steps: obtaining feature values of all feature attributes according to the feature attributes affecting the suspension bridge empty cable alignment error; performing standardization processing on the characteristic values of the characteristic attributes; establishing a suspension bridge empty cable state model, and obtaining a theoretical calculation result of an empty cable linear error through finite element calculation; calculating the influence purity of each characteristic attribute according to a theoretical calculation result, and selecting a plurality of optimal characteristic attributes based on the influence purity; the feature value of the selected feature attribute and the theoretical calculation result serve as the feature value and the target value of a CNN convolutional neural network respectively, and the network is trained; and comparing a theoretical calculation result with a training result, optimizing the CNN, selecting an optimal network with an error within an allowable range as a prediction model, and performing empty cable alignment error prediction. Compared with the prior art, the method has the advantages that the prediction efficiency and the prediction accuracy of the unloaded cable shape error of the asymmetric suspension bridge are improved, and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of main cable construction of suspension bridges, and in particular to a method, a device and a medium for predicting linear errors of empty cables of an asymmetric suspension bridge based on a CNN convolutional neural network. Background Art

[0002] With the vigorous development of the construction industry, the demand for bridges has gradually increased, and the requirements for the aesthetics, economy, and functional performance of bridges have become increasingly higher. Large-span bridges continue to appear in various places. The most advanced of these is the suspension bridge, which refers to a bridge with cables suspended and anchored on both sides through a cable tower. It is a bridge that serves as the main load-bearing component of the superstructure. The geometric shape of its cables is determined by the balance of forces and is generally close to a parabola. Many hangers are hung from the cables to suspend the bridge deck. Stiffening beams are often set between the bridge deck and the hangers to form a combined system with the cables to reduce the deflection deformation caused by the load. The suspension bridge is known as the most ornamental bridge because of its unique appearance. It has become the preferred bridge type for extra-large cross-border bridges because of its large spanning capacity, good seismic performance, and diverse forms.

[0003] The main load-bearing components of a suspension bridge are the main cable, anchor and bridge tower. As one of the main load-bearing components, the main cable bears huge tensile force and is a geometrically variable system, so the shape of the main cable can be changed.

[0004] Once the main cable is installed in place during the construction of a suspension bridge, the internal force deflection of the main cable depends entirely on the structural system, structural deadweight, construction load and temperature changes, and cannot be adjusted freely. Therefore, the analysis of the empty cable line shape of the main cable under the deadweight becomes the key to ensure the smooth and safe construction of the suspension bridge. During the construction process, it is difficult for the actual values ​​of relevant characteristics such as construction factors, environmental factors, and material factors to be consistent with the calculated theoretical values, resulting in errors between the actual empty cable line shape and the line shape after theoretical calculation. The line shape of the empty cable is mainly obtained by measuring the coordinates of each control point on site.

[0005] Due to the diversity of bridge functional requirements and geological conditions, designers will design different bridge structural forms to meet construction needs, so many asymmetric suspension bridges began to appear.

[0006] Table 1 Classification of asymmetric suspension bridges

[0007] Asymmetric type Feature description Anchor type asymmetric Gravity anchor and tunnel anchor are used together The main tower is asymmetrical in height The towers on both sides are of different heights Asymmetric span Multi-span suspension bridges with different spans Main cables are not of the same height Main cable shape is asymmetric Towerless suspension bridge Main cable anchored on the anchor Hybrid Asymmetric Mixed types

[0008] There is an asymmetric line shape in the empty cable state of an asymmetric suspension bridge, which leads to an increase in the number of monitoring control points compared to a symmetric suspension bridge. Ultimately, the actual empty cable line shape is more susceptible to measurement errors, so this part of the work needs to be optimized.

[0009] With the development of computer science and technology and artificial intelligence, various algorithm models have gradually replaced manual applications in all walks of life. Deep learning is a subset of machine learning in artificial intelligence, which is based on artificial neural networks. The learning process is deep. Since the structure of artificial neural networks consists of multiple inputs, outputs, and hidden layers, machines can learn through their own data processing thanks to this structure. The learning process is based on the following steps: 1. Input data into the algorithm. 2. Run the algorithm model with this data. 3. Test and deploy the model. 4. Use the model to perform automated work. Due to the use of artificial neural networks. Deep learning has excellent performance in many modes. For this reason, deep learning is rapidly changing many industries, including healthcare, finance, energy and other industries. Civil engineering, as a national basic project, is currently reflecting on and improving traditional business processes and moving towards a new era of intelligent construction. Deep learning is indispensable. However, deep learning models require a large amount of training data during training. At present, there is no complete data set for the prediction of the air cable line shape of suspension bridges, which makes it difficult for deep learning models to meet the prediction accuracy requirements.

[0010] Finite element method is a numerical technique for finding approximate solutions to boundary value problems of partial differential equations. When solving, the entire region is decomposed, and each sub-region becomes a simple part. This simple part is the finite element. Since most engineering problems are difficult to obtain accurate solutions, and finite element method not only has high calculation accuracy and wide application range, it has become an effective engineering analysis method for civil engineering. Modeling suspension bridges through finite element software such as Midas and substituting the model into finite element method calculation is currently the most commonly used method for suspension bridge design. Since calculating the linear error of the empty cable of an asymmetric suspension bridge is a nonlinear problem, there are many control points on the main cable, the data dimension is high, and the magnitude is small, it is difficult for finite element method calculation to improve calculation efficiency while ensuring the accuracy of the results.

[0011] Therefore, there is an urgent need for a suspension bridge empty cable linear error prediction method that can integrate deep learning models and traditional numerical methods. Summary of the invention

[0012] The purpose of the present invention is to overcome the problems of low processing efficiency in the prior art, the inability to easily predict the linear error of the empty cable of an asymmetric suspension bridge, which may lead to inaccurate judgment of the linear shape of the empty cable and affect the smooth and safe construction of the superstructure of the suspension bridge, and to provide a method, device and medium for predicting the linear error of the empty cable of an asymmetric suspension bridge based on a CNN convolutional neural network.

[0013] The purpose of the present invention can be achieved by the following technical solutions:

[0014] A method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network comprises the following steps:

[0015] S1. According to the characteristic attributes that affect the linear error of the empty cable of the suspension bridge, the characteristic values ​​of all characteristic attributes are obtained;

[0016] S2, standardizing the characteristic values ​​of characteristic attributes;

[0017] S3. Establish a suspension bridge empty cable state model and obtain the theoretical calculation results of the empty cable linear error through finite element calculation;

[0018] S4. Calculate the influence purity of each characteristic attribute according to the theoretical calculation results, and select the optimal multiple characteristic attributes based on the influence purity;

[0019] S5, using the characteristic value of the selected characteristic attribute and the theoretical calculation result of the empty cable linear error as the characteristic value and target value of the CNN convolutional neural network, respectively, to train the network;

[0020] S6. Compare the theoretical calculation results with the training results, optimize the CNN convolutional neural network, select the optimal network with an error within the allowable range as the prediction model, and perform air cable linear error prediction.

[0021] The characteristic attributes include characteristics related to bridge construction technology, characteristics related to bridge material properties and characteristics related to the environment, wherein the characteristics related to the bridge construction technology include the actual installation error of the main cable saddle, the displacement of the main cable saddle consolidated with it caused by the tower deviation of the main tower, the main tower pre-lift elevation and the main tower displacement caused by the catwalk load; the characteristics related to the bridge material properties include the elastic modulus, bulk density, shear modulus and Poisson's ratio of the main tower and the main cable; the characteristics related to the environment include the parameters affecting the elevation of the main tower and the main cable due to temperature.

[0022] The process of standardizing the eigenvalues ​​of the characteristic attributes is: standardizing the eigenvalues ​​of the continuous random variable to a discrete random variable type, specifically, selecting a control point to truncate the continuous parameter, eliminating the eigenvalue data in the extreme intervals of the preset percentage range before and after the variable distribution, and discretizing the remaining intervals into multiple sub-intervals according to an equal-width probability distribution, and using the endpoint values ​​of each sub-interval as the eigenvalue of the characteristic attribute.

[0023] The impact purity is calculated by sensitivity analysis, which includes the following steps:

[0024] Change the characteristic value of a characteristic attribute within a preset range;

[0025] Taking the displacement of the linear control point of the empty cable of the suspension bridge as the control target, the change of the control target after the change of each characteristic value is calculated by finite element software;

[0026] The dimensionless sensitivity coefficient is calculated according to the change of the control target and the change of the characteristic value as the influence purity.

[0027] The calculation formula of the sensitivity coefficient is as follows:

[0028]

[0029] Among them, S is the sensitivity coefficient; Δz is the change in displacement error of the air cable control point after changing a certain eigenvalue, that is, the change in the air cable linear error; z is the displacement error value of the air cable control point obtained by theoretical analysis of the original eigenvalue, that is, the air cable linear error; Δx is the change in the eigenvalue, and x is the original value of the characteristic parameter.

[0030] The step S4 is specifically as follows: calculating the sensitivity coefficient according to all the characteristic attributes, performing purity comparison according to the sensitivity coefficient, sorting the sensitivity coefficient values ​​from large to small, and selecting the characteristic values ​​with the top preset number or the top preset percentage as the training data sample set of the CNN convolutional neural network.

[0031] The input of the CNN convolutional neural network is the characteristic value of each characteristic attribute of each control point of the suspension bridge cable, and the structure is an m×n-order matrix, where m is the number of control points and n is the number of selected characteristic attribute types; the output of the CNN convolutional neural network is the linear error of the cable at each control point.

[0032] The training process of the CNN convolutional neural network specifically includes the following steps:

[0033] Initialize the grid structure of the CNN convolutional neural network, including the size and number of convolutional layers, pooling layers, and fully connected layers, as well as the initial weights;

[0034] Determine the hyperparameters of the CNN convolutional neural network, including the number of iterations, learning rate, batch size, step size, and activation function;

[0035] The input data is forward propagated through the convolutional layer, pooling layer, and fully connected layer to obtain the output value;

[0036] Calculate the error between the network output value and the target value;

[0037] When the error value is greater than the expected error, the error is back-propagated back to the network, the weights are improved and retrained; when the error is equal to or less than the expected error, the training ends.

[0038] A device for predicting the linear error of an empty cable of an asymmetric suspension bridge based on a CNN convolutional neural network comprises a memory, a processor, and a program stored in the memory. When the processor executes the program, the method described above is implemented.

[0039] A storage medium stores a program, which implements the method described above when executed.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) Based on the CNN convolutional neural network, the present invention intelligently and simply applies the existing method for predicting the linear error of the empty cable calculated by finite element theory, effectively reducing the difficulty of predicting the linear error of the empty cable and improving the engineering efficiency. At the same time, by comparing the difference between the model operation results and the theoretical calculation results, the method of optimizing the model can select and continuously improve the training model, and automatically improve the accuracy of the calculation of the linear shape of the empty cable. The present invention applies the deep learning method in the field of artificial intelligence to the field of bridge engineering, provides new ideas for the structural calculation method of bridge engineering, and can effectively promote the development of bridge engineering into the era of intelligent construction.

[0042] (2) The present invention determines the characteristic attributes that affect the linear error of the empty cable of the suspension bridge, and after calculating the characteristics, screens the characteristic value pairs by affecting the purity, thereby constructing a training data set for the neural network, so that the prediction results of the trained neural network are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the method of the present invention;

[0044] Figure 2 Schematic diagram of selecting control points for the empty cable of the suspension bridge;

[0045] Figure 3 Schematic diagram of the algorithm flow of CNN convolutional neural network. DETAILED DESCRIPTION

[0046] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0047] This embodiment provides a method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network. Figure 1 As shown, the following steps are included:

[0048] S1. According to the characteristic attributes that affect the linear error of the empty cable of the suspension bridge, the characteristic values ​​of all characteristic attributes are obtained.

[0049] The characteristic attributes include characteristics related to bridge construction technology, characteristics related to bridge material properties and characteristics related to the environment, wherein the characteristics related to the bridge construction technology include the actual installation error of the main cable saddle, the displacement of the main cable saddle consolidated with it caused by the tower deviation of the main tower, the main tower pre-lift elevation and the main tower displacement caused by the catwalk load; the characteristics related to the bridge material properties include the elastic modulus, bulk density, shear modulus and Poisson's ratio of the main tower and the main cable; the characteristics related to the environment include the parameters affecting the elevation of the main tower and the main cable due to temperature.

[0050] This embodiment is described in detail by taking a 4-main-cable long-span suspension bridge with different vertical angles in high and low spaces as an example. The elastic modulus, bulk density, Poisson's ratio and other properties of the main cable are obtained through material property experiments; the temperature properties are obtained through weather monitoring; and the deviation of the main cable saddle consolidating it caused by the tower deviation of the main tower, the pre-lift elevation of the main tower, the main tower offset caused by the catwalk load, and the installation error of the main cable saddle are obtained through construction inspection.

[0051] S2. Standardize the characteristic values ​​of the characteristic attributes.

[0052] The characteristic value calculated in step S1 is a continuous random variable, the variable conforms to the central limit law, and each characteristic parameter basically satisfies the normal distribution.

[0053] This step standardizes the eigenvalues ​​of the characteristic attributes, that is, standardizes and converts the eigenvalues ​​of continuous random variables into discrete random variable types. Specifically, the control point truncation is selected for the continuous parameters, the eigenvalue data in the extreme intervals of 5% before and after the variable distribution are eliminated, and the remaining intervals are discretized into multiple sub-intervals according to the equal-width probability distribution, and the endpoint values ​​of each sub-interval are used as the eigenvalues ​​of the characteristic attributes.

[0054] In this embodiment, all feature attributes are continuous random variables as described in step S2 and are standardized using the above method.

[0055] S3. Use Midas to establish the empty cable state model of the suspension bridge, and obtain the theoretical calculation results of the empty cable linear error through finite element calculation.

[0056] In this embodiment, the linear error of the empty cable is a discrete error value, which is obtained by subdividing the control points of the main cable. The control points selected in this embodiment are as follows: Figure 2 shown.

[0057] S4. Calculate the influence purity of each characteristic attribute according to the theoretical calculation results, and select the optimal multiple characteristic attributes based on the influence purity.

[0058] For the shape of the air cable, different characteristic factors have different influences. For the sake of simplicity, it is necessary to discard the characteristics with small influence and select and pay attention to the characteristics with large influence. The influence degree is referred to as purity. The calculation of purity is called sensitivity analysis, which is to find out several factors that are sensitive to the change of the air cable shape as the key control factors of the error analysis of the air cable shape of the asymmetric suspension bridge.

[0059] Sensitivity calculation is a method of sensitivity analysis. The specific method is to change a certain influencing parameter of the suspension bridge model to make its value change slightly, control other parameters unchanged, and then check the displacement change of the control point of the empty cable line shape. The ratio of the displacement change to the small change value of the influencing parameter is calculated as the sensitivity of the empty cable line shape to the influencing factor.

[0060] The sensitivity analysis to calculate the impact on purity includes the following steps:

[0061] 1. Change the characteristic value of the characteristic attribute within the preset range;

[0062] 2. Taking the displacement of the suspension bridge cable linear control point (i.e., the cable linear error) as the control target, the change in the control target after each characteristic value is changed is calculated using finite element software;

[0063] 3. Calculate the dimensionless sensitivity coefficient based on the change in the control target and the change in the characteristic value as the influence purity.

[0064] The dimensions of each influencing factor are different, so the sensitivity needs to be quantified using the processed dimensionless sensitivity coefficient. The calculation formula of the sensitivity coefficient is as follows:

[0065]

[0066] Among them, S is the sensitivity coefficient; Δz is the change in displacement error of the air cable control point after changing a certain eigenvalue, that is, the change in the air cable linear error; z is the displacement error value of the air cable control point obtained by theoretical analysis of the original eigenvalue, that is, the air cable linear error; Δx is the change in the eigenvalue, and x is the original value of the characteristic parameter.

[0067] After calculating the influence purity, the sensitivity coefficient is calculated according to all the feature attributes, and the purity is compared according to the sensitivity coefficient. After sorting the sensitivity coefficient values ​​from large to small, the feature values ​​with the top preset number or the top preset percentage are selected as the training data sample set of the CNN convolutional neural network.

[0068] S5. The eigenvalues ​​of the selected characteristic attributes and the theoretical calculation results of the empty cable linear error are used as the eigenvalues ​​and target values ​​of the CNN convolutional neural network respectively to train the network.

[0069] Convolutional neural network (CNN) is a deep learning model commonly used in image recognition, speech recognition and other fields. The core idea of ​​CNN is to extract data features through convolution operations, thereby achieving tasks such as data classification and prediction. Convolutional neural networks are widely used due to their accuracy and suitability for nonlinear prediction of complex data.

[0070] Convolution is a concept in the field of signal science, and its formula is

[0071]

[0072] Where f(x) and g(x) are two integrable functions.

[0073] A neural network is a simulation of the neuron mechanism of the human brain. Each neuron is a computing unit. The input data is multiplied by the weight, summed and biased, and the resulting data is then output through an activation function.

[0074] The activation function is a function that runs on the neurons of the neural network and is responsible for mapping the input of the neuron to the output. Commonly used activation functions include Sigmoid function, Tanh function, and ReLU function.

[0075] like Figure 3 As shown in the figure, the basic structure of a convolutional neural network consists of a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer extracts the features of the data by performing a convolution operation on the input data. The pooling layer is used to reduce the dimension of the feature map output by the convolutional layer, thereby reducing model parameters and the amount of calculation. The fully connected layer is used to map the feature vector output by the pooling layer to the output category, thereby achieving classification or recognition of the input data.

[0076] The evaluation indicators of the convolutional neural network used for the prediction model are root mean square error RMSE and mean absolute error MAE. The convolutional neural network is optimized according to the results of the evaluation indicators. The specific optimization scheme is to optimize the convolutional neural network structure or optimize the hyperparameters.

[0077] In this embodiment, the input of the CNN convolutional neural network is the characteristic value of each characteristic attribute of each control point of the suspension bridge cable, and the structure is an m×n-order matrix, where m is the number of control points and n is the number of selected characteristic attribute types; the output of the CNN convolutional neural network is the linear error of the cable at each control point.

[0078] The training process of CNN convolutional neural network specifically includes the following steps:

[0079] S501, initializing the grid structure of the CNN convolutional neural network, including the size and number of convolutional layers, pooling layers, and fully connected layers, as well as initial weights;

[0080] S502, determine the hyperparameters of the CNN convolutional neural network, including the number of iterations, learning rate, batch size, step size, and activation function;

[0081] S503, the input data is forward propagated through the convolution layer, the pooling layer, and the fully connected layer to obtain an output value;

[0082] S504, calculating the error between the output value of the network and the target value;

[0083] S505. When the error value is greater than the expected error, the error is back-propagated back to the network, the weights are improved and retrained; when the error is equal to or less than the expected error, the training is terminated.

[0084] S6. Compare the theoretical calculation results with the training results, optimize the CNN convolutional neural network, select the optimal network with an error within the allowable range as the prediction model, and perform air cable linear error prediction.

[0085] Among them, the training results are the output results of the CNN convolutional neural network test set, and the theoretical calculation results are the finite element calculation results of the Midas software.

[0086] The optimal network is obtained by calculating the evaluation index of the CNN convolutional neural network. In this embodiment, the evaluation index includes the root mean square error RMSE and the mean absolute error MAE. The specific calculation formula is as follows:

[0087]

[0088] The mean absolute error is the ratio of the sum of the absolute values ​​of the deviations between the observed value and the true value to the number of observations. The specific calculation formula is as follows:

[0089]

[0090] in, is the predicted value of the convolutional neural network model at each control point, y i is the theoretical calculated value of each control point, and m is the number of selected control points.

[0091] In this step, the optimization of the CNN convolutional neural network includes optimizing the convolutional neural network structure or optimizing hyperparameters. The specific optimization idea is to use optimization algorithms and manual experience to adjust parameters for optimization.

[0092] The above is an introduction to a method embodiment. The following is a further explanation of the solution of the present invention through an apparatus embodiment.

[0093] This embodiment provides an asymmetric suspension bridge empty cable linear error prediction device based on a CNN convolutional neural network, including a memory, a processor, and a program stored in the memory. When the processor executes the program, the method described above is implemented.

[0094] In a preferred embodiment, the device comprises:

[0095] Eigenvalue acquisition module: obtains the eigenvalues ​​of all eigenvalues ​​according to the eigenvalues ​​that affect the linear error of the empty cable of the suspension bridge;

[0096] Standardization processing module: standardizes the characteristic values ​​of characteristic attributes;

[0097] Theoretical value calculation module: establish the empty cable state model of the suspension bridge, and obtain the theoretical calculation results of the empty cable linear error through finite element calculation;

[0098] Screening module: Calculate the influence purity of each characteristic attribute according to the theoretical calculation results, and select the optimal multiple characteristic attributes based on the influence purity;

[0099] Convolutional neural network training module: The eigenvalues ​​of the selected feature attributes and the theoretical calculation results of the empty cable linear error are used as the eigenvalues ​​and target values ​​of the CNN convolutional neural network to train the network;

[0100] Network optimization and prediction module: Compare the theoretical calculation results with the training results, optimize the CNN convolutional neural network, select the optimal network with an error within the allowable range as the prediction model, and predict the empty cable linear error.

[0101] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0102] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several 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, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0103] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network, characterized in that: The following steps are involved: S1. According to the characteristic attributes that affect the linear error of the empty cable of the suspension bridge, the characteristic values ​​of all characteristic attributes are obtained; S2, standardizing the characteristic values ​​of characteristic attributes; S3. Establish a suspension bridge empty cable state model and obtain the theoretical calculation results of the empty cable linear error through finite element calculation; S4. Calculate the influence purity of each characteristic attribute according to the theoretical calculation results, and select the optimal multiple characteristic attributes based on the influence purity; S5, using the characteristic value of the selected characteristic attribute and the theoretical calculation result of the empty cable linear error as the characteristic value and target value of the CNN convolutional neural network, respectively, to train the network; S6. Compare the theoretical calculation results with the training results, optimize the CNN convolutional neural network, select the optimal network with an error within the allowable range as the prediction model, and predict the empty cable linear error.

2. According to the method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 1, it is characterized in that: The characteristic attributes include characteristics related to bridge construction technology, characteristics related to bridge material properties and characteristics related to the environment, wherein the characteristics related to the bridge construction technology include the actual installation error of the main cable saddle, the displacement of the main cable saddle consolidated with it caused by the tower deviation of the main tower, the main tower pre-lift elevation and the main tower displacement caused by the catwalk load; the characteristics related to the bridge material properties include the elastic modulus, bulk density, shear modulus and Poisson's ratio of the main tower and the main cable; the characteristics related to the environment include the parameters affecting the elevation of the main tower and the main cable due to temperature.

3. The method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 1 is characterized in that: The process of standardizing the eigenvalues ​​of the characteristic attributes is: standardizing the eigenvalues ​​of the continuous random variable to a discrete random variable type, specifically, selecting a control point to truncate the continuous parameter, eliminating the eigenvalue data in the extreme intervals of the preset percentage range before and after the variable distribution, and discretizing the remaining intervals into multiple sub-intervals according to an equal-width probability distribution, and using the endpoint values ​​of each sub-interval as the eigenvalue of the characteristic attribute.

4. The method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 1 is characterized in that: The impact purity is calculated by sensitivity analysis, which includes the following steps: Change the characteristic value of a characteristic attribute within a preset range; Taking the displacement of the linear control point of the empty cable of the suspension bridge as the control target, the change of the control target after the change of each characteristic value is calculated by finite element software; The dimensionless sensitivity coefficient is calculated according to the change of the control target and the change of the characteristic value as the influence purity.

5. The method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 4 is characterized in that: The calculation formula of the sensitivity coefficient is as follows: Among them, S is the sensitivity coefficient; Δz is the change in displacement error of the air cable control point after changing a certain eigenvalue, that is, the change in the air cable linear error; z is the displacement error value of the air cable control point obtained by theoretical analysis of the original eigenvalue, that is, the air cable linear error; Δx is the change in the eigenvalue, and x is the original value of the characteristic parameter.

6. The method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 4 is characterized in that: The step S4 is specifically as follows: calculating the sensitivity coefficient according to all the characteristic attributes, performing purity comparison according to the sensitivity coefficient, sorting the sensitivity coefficient values ​​from large to small, and selecting the characteristic values ​​with the top preset number or the top preset percentage as the training data sample set of the CNN convolutional neural network.

7. The method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 1 is characterized in that: The input of the CNN convolutional neural network is the characteristic value of each characteristic attribute of each control point of the suspension bridge cable, and the structure is an m×n-order matrix, where m is the number of control points and n is the number of selected characteristic attribute types; the output of the CNN convolutional neural network is the linear error of the cable at each control point.

8. The method for predicting the linear error of an asymmetric suspension bridge empty cable based on a CNN convolutional neural network according to claim 1 is characterized in that: The training process of the CNN convolutional neural network specifically includes the following steps: Initialize the grid structure of the CNN convolutional neural network, including the size and number of convolutional layers, pooling layers, and fully connected layers, as well as the initial weights; Determine the hyperparameters of the CNN convolutional neural network, including the number of iterations, learning rate, batch size, step size, and activation function; The input data is forward propagated through the convolutional layer, pooling layer, and fully connected layer to obtain the output value; Calculate the error between the network output value and the target value; When the error value is greater than the expected error, the error is back-propagated back to the network, the weights are improved and retrained; when the error is equal to or less than the expected error, the training ends.

9. A device for predicting the linear error of an empty cable of an asymmetric suspension bridge based on a CNN convolutional neural network, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images