Reasonable bridge state prediction method and system for cable-stayed bridge based on deep learning

Abstract

The present invention discloses a method and system for predicting the reasonable state of a cable-stayed bridge based on deep learning, including: S1: determining the construction target of the cable-stayed bridge, collecting the state data of the historical cable-stayed bridge, defining the state space of the reasonable state prediction model of the cable-stayed bridge and initializing the state space; S2: constructing a finite element model of the cable-stayed bridge, and calculating the force parameters and deformation parameters of the cable-stayed bridge; S3: generating the original morphological image of the cable-stayed bridge and preprocessing it, then extracting the morphological features of the cable-stayed bridge, enhancing the state space, and obtaining the enhanced state space; S4: defining the action space, value function and reward function of the reasonable state prediction model of the cable-stayed bridge; S5: performing reinforcement learning training on the reasonable state prediction model of the cable-stayed bridge, and taking the state space at the end of the training as the reasonable state of the cable-stayed bridge. The present invention solves the problem that the traditional method for calculating the state of a cable-stayed bridge relies on manual calculation and is complex in calculation.

Description

Technical Field

[0001] The present invention relates to cable-stayed bridge technology, and in particular to a method and system for predicting a reasonable completed bridge state of a cable-stayed bridge based on deep learning. Background Art

[0002] As an important bridge structure, the determination of the completed state of the cable-stayed bridge is crucial to ensure the safety, stability and durability of the bridge. The calculation method of the completed state of the traditional cable-stayed bridge usually adopts the analytical method or the finite element method. The analytical method is suitable for bridges with simple structures and clear boundary conditions. By establishing a mechanical model, the analytical expression of the stress state and deformation of the bridge in the completed state is derived. The finite element method is suitable for bridges with complex structures and variable boundary conditions. By discretizing the bridge into a finite number of units, a finite element model is established, and the stress state and deformation of the bridge are obtained through numerical calculation.

[0003] Although traditional methods play an important role in determining the completed state of cable-stayed bridges, they have the following disadvantages: Traditional methods rely on a large amount of manual calculations, which are not only time-consuming and laborious, but also prone to errors. Especially when dealing with complex structures and long-span bridges, the amount of calculation is huge, and it is difficult to ensure the accuracy and reliability of the calculation results; the completed state of the bridge obtained based on traditional methods may have deviations or deficiencies, and needs to be adjusted and optimized. However, due to the large amount of calculations and limited measured data, the adjustment and optimization process is often difficult and it is difficult to achieve the desired effect. Summary of the invention

[0004] In view of this, the present invention aims to provide a method and system for predicting the reasonable completion status of a cable-stayed bridge based on deep learning, so as to solve the problem that the traditional method for calculating the completion status of a cable-stayed bridge relies on manual calculation and the calculation is complex.

[0005] The reasonable bridge state prediction method of cable-stayed bridge based on deep learning includes:

[0006] S1: Determine the construction target of the cable-stayed bridge, collect the historical cable-stayed bridge completion status data, define the state space of the reasonable cable-stayed bridge completion status prediction model, calculate the state space initialization parameters, and initialize the state space, including:

[0007] According to the historical cable-stayed bridge status data, the preliminary main tower height is calculated;

[0008] Calculate the target main tower height based on the preliminary main tower height, the status characteristics of historical cable-stayed bridges, and the difference characteristics of historical cable-stayed bridges;

[0009] Extract the characteristics of the target cable-stayed bridge's completed state based on the target main tower height, target load, target main beam length and target material properties;

[0010] Similarity matching is performed based on the completed state characteristics of the target cable-stayed bridge, and the historical completed state data of the cable-stayed bridge with the highest similarity is selected to initialize the state space;

[0011] S2: According to the construction target of the cable-stayed bridge and the initialization parameters of the state space, a finite element model of the cable-stayed bridge is constructed, and the force parameters and deformation parameters of the cable-stayed bridge are calculated;

[0012] S3: According to the stress parameters and deformation parameters of the cable-stayed bridge, the original morphological image of the cable-stayed bridge is generated by simulation software, and the original morphological image of the cable-stayed bridge is preprocessed to obtain a preprocessed cable-stayed bridge image; the morphological features of the cable-stayed bridge are extracted from the preprocessed cable-stayed bridge image by using a convolutional neural network; the state space is enhanced according to the morphological features of the cable-stayed bridge to obtain an enhanced state space;

[0013] S4: Define the action space, value function and reward function of the reasonable bridge state prediction model for cable-stayed bridges;

[0014] S5: Design a segmented optimization training strategy, use the reward function to perform reinforcement learning training on the reasonable completion state prediction model of the cable-stayed bridge, and use the state space at the end of the training as the reasonable completion state of the cable-stayed bridge, including:

[0015] S51: Design a segmented optimization training strategy, and use the reward function to perform reinforcement training on the reasonable bridge state prediction model of the cable-stayed bridge. In the first stage of reinforcement learning training, the target main tower height is optimized, the target number of cables is optimized in the second stage, and the upper and lower anchor point coordinates of the target cables are optimized in the third stage.

[0016] S52: Initialize the Q table of the reasonable completed bridge state prediction model of the cable-stayed bridge according to the state space initialization parameters. First, optimize the target main tower height until the target main tower height converges. The optimization calculation method is:

[0017] ;

[0018] in, is the state-action pair in the Q table The value of To update the symbol, is the learning rate, is the reward for executing an action in the current state, s is the current state, a is the action to be executed, To obtain the maximum value, is the discount factor, is the state-action pair in the Q table The value of For the new state, For new status Next, execute the action;

[0019] S53: After the target main tower height converges, fix it as a constant and enter the second stage to optimize the target number of cables; after the target number of cables converges, fix both the target main tower height and the target number of cables as constants and enter the third stage to optimize the upper and lower anchor point coordinates of the target cables. The optimization method of the target number of cables and the optimization method of the upper and lower anchor point coordinates of the target cables are the same as step S52;

[0020] S54: After the optimization of the coordinates of the upper and lower anchor points of the target cables is completed, the state space of the reasonable completed state prediction model of the cable-stayed bridge is output as the reasonable completed state of the cable-stayed bridge.

[0021] Furthermore, the S1 step includes:

[0022] S11: Determine the construction target of the cable-stayed bridge, including the target main beam length, target load and target material properties;

[0023] Collect historical cable-stayed bridge status data, including historical main beam length, historical load, historical material properties, historical main tower height, historical number of cables, and historical upper and lower anchor point coordinates of cables;

[0024] Define the state space of the reasonable completed state prediction model of the cable-stayed bridge, including the target main tower height, target number of cables, and target upper and lower anchor point coordinates of the cables;

[0025] According to the historical cable-stayed bridge state data, the state space initialization parameters are calculated, including the target main tower height, the target number of cables, and the target upper and lower anchor point coordinates of the cables:

[0026] The preliminary calculation of the main tower height is:

[0027] ;

[0028] in, is the preliminary main tower height, is the amount of historical cable-stayed bridge status data, is the similarity weight between the target material attribute and the i-th historical material attribute, i is the data index of the completed state of the historical cable-stayed bridge, is the height of the i-th historical main tower, is the target load, is the historical load of the ith article, is the target main beam length, is the length of the i-th historical main beam; is the target material property vector, is the attribute vector of the i-th historical material, To take the modulus length, is a natural constant, is the regulating factor, To take the absolute value;

[0029] S12: Calculate the target main tower height based on the preliminary main tower height, the status characteristics of historical cable-stayed bridges, and the difference characteristics of historical cable-stayed bridges:

[0030] ;

[0031] in, The characteristics of the completed state of the historical cable-stayed bridge. is a multi-layer perceptron, The differences in the status of historical cable-stayed bridges are shown in Figure 1. is the target main tower height;

[0032] S13: Extract the characteristics of the target cable-stayed bridge’s completed state based on the target main tower height, target load, target main beam length, and target material properties:

[0033] ;

[0034] in, The completed state characteristics of the target cable-stayed bridge;

[0035] S14: Perform similarity matching based on the target cable-stayed bridge completion state characteristics, and select the historical cable-stayed bridge completion state data with the highest similarity to obtain the target cable quantity and the upper and lower anchor point coordinates of the target cables:

[0036] ;

[0037] in, is the similarity between the completed state characteristics of the target cable-stayed bridge and the completed state characteristics of the i-th historical cable-stayed bridge, is the cosine similarity calculation, The data index of the historical cable-stayed bridge status with the highest similarity. To get the maximum value index i;

[0038] Calculate the historical cable-stayed bridge status data index with the highest similarity Then select The number of historical cables and the coordinates of the upper and lower anchor points of the historical cables in the completed state data of the historical cable-stayed bridges are used as the target number of cables and the coordinates of the upper and lower anchor points of the target cables;

[0039] S15: Initializing the state space of the reasonable completed state prediction model of the cable-stayed bridge according to the calculated state space initialization parameters.

[0040] It should be noted that the present invention uses a multi-layer perceptron and cosine similarity to evaluate the similarity between the target cable-stayed bridge and the historical cable-stayed bridge in terms of their completion status, and accordingly assigns weights to the historical data to calculate the preliminary main tower height; at the same time, the multi-layer perceptron is also used to encode the state characteristics of the historical cable-stayed bridge, and by comparing the differences between the completion status characteristics of the target cable-stayed bridge and the completion status characteristics of the historical cable-stayed bridge, the preliminary main tower height of the main tower height is further fine-tuned; finally, by selecting the historical cable-stayed bridge that is most similar to the completion status of the target cable-stayed bridge, the initial values ​​of the target cable quantity and the upper and lower anchor point coordinates are directly determined;

[0041] In reinforcement learning tasks, the initialization of the state space directly affects the exploration efficiency and convergence speed of the algorithm. Traditional state space initialization methods are often initialized to all zeros or random values. Although this method is simple and easy, it will lead to blind exploration of the algorithm in the learning process and slow down the convergence speed. The present invention initializes the state space by drawing on historical data, which can provide a more reasonable and optimal starting point for the optimization algorithm, making the optimization process more in line with engineering practice. The algorithm can converge to the optimal solution faster, thereby significantly improving the optimization efficiency and providing an accurate and efficient parameter initialization scheme for the reasonable prediction of the completed bridge state of the cable-stayed bridge.

[0042] Furthermore, the S2 step further includes:

[0043] S21: According to the target main beam length and the target material properties, a finite element model of the cable-stayed bridge is constructed in the finite element modeling software, and the target material properties are assigned to the main beam in the finite element model of the cable-stayed bridge;

[0044] S22: Add main towers and cables to the finite element model of the cable-stayed bridge according to the state space initialization parameters;

[0045] S23: Apply loads, including dead loads and dynamic loads, to the finite element model of the cable-stayed bridge according to the target load amount;

[0046] S24: Calculate the stress parameters and deformation parameters of the cable-stayed bridge by using finite element modeling software. The stress parameters of the cable-stayed bridge include the internal force of the main beam, the cable force and the internal force of the main tower. The deformation parameters of the cable-stayed bridge include the deflection of the main beam, the displacement of the main tower and the elongation of the cable.

[0047] Furthermore, the step of extracting the morphological features of the cable-stayed bridge using a convolutional neural network in S3 includes:

[0048] Firstly, a cable-stayed bridge morphological feature extraction model is constructed, the first morphological feature of the cable-stayed bridge is extracted according to the preprocessed cable-stayed bridge image, and the convolution fusion gate is calculated based on the first morphological feature of the cable-stayed bridge;

[0049] Combine the convolution fusion gate to fuse the Gaussian kernel and the edge detection kernel to obtain the fused convolution kernel;

[0050] The morphological features of the cable-stayed bridge are extracted by fusing the convolution kernel and channel splicing.

[0051] Furthermore, the S3 step also includes:

[0052] S31: In the simulation software, force parameters and deformation parameters of the cable-stayed bridge are input, the cable-stayed bridge is simulated, and an original shape image of the cable-stayed bridge is obtained;

[0053] S32: graying and downsampling the original image of the cable-stayed bridge to obtain a preprocessed image of the cable-stayed bridge, wherein the pixel size of the preprocessed image of the cable-stayed bridge is 640×640;

[0054] S33: Construct a cable-stayed bridge morphological feature extraction model, and extract the morphological features of the cable-stayed bridge according to the preprocessed cable-stayed bridge image. The calculation method is:

[0055] ;

[0056] in, It is the first morphological feature of the cable-stayed bridge. is a convolutional neural network, is the preprocessed cable-stayed bridge image, is the convolution fusion gate, is the convolution fusion weight, is the convolution fusion bias vector, To fuse the convolution kernel, is the Gaussian kernel, is the edge detection kernel, The second morphological feature of the cable-stayed bridge is is the sigmoid function, The first morphological feature of a cable-stayed bridge The number of feature maps contained in , j is the feature map index, is the convolution operation, is the jth characteristic graph of the first morphological feature of the cable-stayed bridge, is the morphological characteristic offset vector of the cable-stayed bridge, The morphological characteristics of cable-stayed bridges are: For a convolutional neural network using fused convolution kernels, For channel splicing;

[0057] S34: According to the morphological characteristics of the cable-stayed bridge, the state space of the reasonable bridge state prediction model of the cable-stayed bridge is enhanced to obtain an enhanced state space; the enhanced state space includes the target main tower height, the target number of cables, the coordinates of the upper and lower anchor points of the target cables, and the morphological characteristics of the cable-stayed bridge.

[0058] It is further explained that step S33 of the present invention introduces a convolution fusion gate and a fusion convolution kernel, and fuses the characteristics of the Gaussian kernel and the edge detection kernel with the feature map in the first morphological feature of the cable-stayed bridge, so that the feature contains both the global structural information of the image (Gaussian kernel) and the edge and detail information (edge ​​detection kernel), thereby enhancing the expressive power of the feature; then, the fused features are nonlinearly transformed by the sigmoid function, and combined with the initially extracted feature map, the diversity of the features is further integrated and enhanced; finally, the first morphological feature of the cable-stayed bridge, the second morphological feature of the cable-stayed bridge, and the preprocessed cable-stayed bridge image are channel-joined, and feature extraction is performed by a deep convolutional neural network, thereby realizing multi-scale and multi-level feature integration and utilization;

[0059] Traditional convolutional neural networks generally use fixed convolution kernels to extract image features, while the present invention integrates the characteristics of the Gaussian kernel and the edge detection kernel into the convolution kernel according to the characteristics of the cable-stayed bridge image. This adaptive convolution kernel design can more accurately capture the key features in the cable-stayed bridge image, so that the model has good generalization ability for different types of cable-stayed bridge images and different bridge completion states, and further captures subtle changes in the cable-stayed bridge completion state, thereby achieving accurate prediction of the bridge completion state.

[0060] Furthermore, the S4 step further includes:

[0061] S41: define the action space of the reasonable bridge state prediction model of the cable-stayed bridge, including: main tower height +1, main tower height -1, number of cables +1, number of cables -1, cable upper anchor point coordinate +1, cable upper anchor point coordinate -1, cable lower anchor point coordinate +1, cable lower anchor point coordinate -1;

[0062] S42: The value function of Q-learning is used as the value function of the reasonable bridge state prediction model of the cable-stayed bridge;

[0063] S43: Define the reward function of the reasonable bridge state prediction model of the cable-stayed bridge, and the calculation method is:

[0064] ;

[0065] in, is the reward for executing an action in the current state, s is the current state, a is the action to be executed, Reward for structural safety, Reward for structural stability, Rewards for exploration;

[0066] ;

[0067] in, is the safety bonus coefficient, To obtain the maximum value, For new status The stress index under is the stress index threshold; The new status The internal forces of the main beam, cable and main tower below;

[0068] ;

[0069] in, is the stability reward coefficient, For new status The displacement index under is the displacement index threshold; The new status The deflection of the main beam, the displacement of the main tower and the elongation of the cables;

[0070] ;

[0071] in, To explore the reward coefficient, State-action pairing in reasonable bridge state prediction model for cable-stayed bridges The number of visits, is the total number of visits.

[0072] Further, step S43 of the present invention defines a reward function of a reasonable bridge state prediction model for a cable-stayed bridge, which comprehensively considers three factors: structural safety, structural stability, and exploration.

[0073] In the structural safety reward, the reward function is proportional to the weighted sum of the internal forces of the main beam, the cable forces and the main tower. Therefore, the algorithm tends to select actions that can make the internal force distribution more uniform and reasonable, thereby optimizing the overall stress performance of the cable-stayed bridge. At the same time, the stress index and stress index threshold are introduced. The structural safety reward can ensure that the stress level of the cable-stayed bridge during the design process does not exceed the allowable range, thereby effectively ensuring the safety of the structure.

[0074] In the structural stability reward, the reward function is proportional to the square of the weighted sum of the main beam deflection, main tower offset and cable elongation. The algorithm tends to select actions that can reduce displacement and increase structural stiffness, thereby optimizing the displacement control performance of the cable-stayed bridge. The ratio of the displacement index to the displacement index threshold can limit the displacement level of the cable-stayed bridge during the design process and ensure the stability of the structure.

[0075] In the exploratory reward, by introducing the ratio of the number of visits to the state-action pair to the total number of visits and applying an exponential decay function, the reward function can encourage the algorithm to explore more unknown states during the training process and avoid falling into the local optimal solution; as the training progresses, the number of visits increases, the exploratory reward gradually decreases, and the algorithm gradually converges to a stable optimal solution or suboptimal solution.

[0076] It is further explained that the segmented optimization strategy of step S5 of the present invention decomposes the complex optimization problem into multiple smaller sub-problems, each of which focuses on only one or a few optimization objectives, so that the model only needs to consider the optimization objective of the current stage each time iterates, without having to process all possible parameter combinations at the same time, thereby greatly reducing the amount of calculation; in addition, through a gradual convergence approach, the optimization results of each stage can be used as the starting point of the next stage, avoiding a comprehensive search from scratch, and further saving computing resources; in the initial stage, the model only needs to focus on the optimization of the main tower height, which helps the model to understand and learn this specific task more quickly; as the optimization progresses, the model gradually introduces more parameters (such as the number of cables, the coordinates of the upper and lower anchor points of the cables), but each time is based on the previous stage, so the model can more easily adapt to these new optimization objectives; this method of gradually increasing complexity helps to avoid the model facing overly complex optimization problems in the initial stage, thereby reducing the difficulty of learning.

[0077] The present invention also discloses a reasonable bridge state prediction system for a cable-stayed bridge based on deep learning, comprising:

[0078] Cable-stayed bridge state space initialization module: determine the construction target of the cable-stayed bridge, collect the historical cable-stayed bridge state data, define the state space of the reasonable cable-stayed bridge state prediction model, calculate the state space initialization parameters, and perform state space initialization;

[0079] Cable-stayed bridge finite element model construction module: According to the cable-stayed bridge construction objectives and state space initialization parameters, the cable-stayed bridge finite element model is constructed, and the force parameters and deformation parameters of the cable-stayed bridge are calculated;

[0080] State space enhancement module: According to the stress parameters and deformation parameters of the cable-stayed bridge, the original morphological image of the cable-stayed bridge is generated by simulation software, and then the original morphological image of the cable-stayed bridge is preprocessed, and then the morphological features of the cable-stayed bridge are extracted by using a convolutional neural network, and the state space is enhanced to obtain the enhanced state space;

[0081] Reinforcement learning design module: defines the action space, value function and reward function of the reasonable bridge state prediction model of the cable-stayed bridge;

[0082] Reasonable bridge completion state prediction module: Design a segmented optimization training strategy, use the reward function to perform reinforcement learning training on the reasonable bridge completion state prediction model of the cable-stayed bridge, and use the state space at the end of the training as the reasonable bridge completion state of the cable-stayed bridge.

[0083] Compared with the prior art, the present invention has the following beneficial effects:

[0084] (1) The traditional calculation method of the completed state of a cable-stayed bridge is highly dependent on manual calculation. Not only is the calculation process complicated and cumbersome, but it is also difficult to fully consider various influencing factors, resulting in a large deviation between the calculation results and the actual situation. The present invention introduces reinforcement learning and deep learning technologies to construct an intelligent prediction model for the reasonable completed state of a cable-stayed bridge. The model can automatically learn and optimize the design parameters of the cable-stayed bridge, thereby quickly and accurately predicting the reasonable completed state of the bridge.

[0085] (2) To address the problems of slow convergence speed and blind algorithm exploration in traditional state space initialization methods of reinforcement learning, the present invention introduces a deep learning model in the state space initialization stage, combines multi-layer perceptron and cosine similarity, evaluates the similarity between the target cable-stayed bridge and the historical cable-stayed bridge in the completed state, and encodes the state characteristics of the historical cable-stayed bridge. Finally, the historical cable-stayed bridge that is most similar to the target cable-stayed bridge in the completed state is selected to directly determine the initial values ​​of the target number of cables and the upper and lower anchor point coordinates. This can provide a more reasonable starting point that is closer to the optimal solution for the optimization algorithm, making the optimization process more in line with engineering practice, significantly improving the optimization efficiency, and providing an accurate and efficient parameter initialization scheme for the reasonable completed state prediction of cable-stayed bridges.

[0086] (3) The present invention generates the original morphological image of the cable-stayed bridge and performs preprocessing, and then extracts the morphological features of the cable-stayed bridge, thereby effectively enhancing the state space. In this step, the present invention also designs a convolution fusion gate and a fusion convolution kernel, which fuses the characteristics of the Gaussian kernel and the edge detection kernel with the feature map in the first morphological feature of the cable-stayed bridge, so that the feature contains both the global structural information of the image and the edge and detail information, thereby enhancing the expressive power of the feature. This not only enriches the amount of information in the state space, but also enables the model to more accurately capture the morphological changes of the cable-stayed bridge, thereby further improving the accuracy of the prediction.

[0087] (4) The present invention proposes a reward function for the prediction model of the reasonable completion state of a cable-stayed bridge. This function comprehensively considers the three factors of structural safety, structural stability and exploratory nature, and optimizes the overall force performance and displacement control performance of the cable-stayed bridge. At the same time, since the reasonable completion state of a cable-stayed bridge involves many parameters and the model optimization is difficult, the present invention proposes a segmented optimization strategy to decompose the complex optimization problem into multiple smaller sub-problems, which greatly reduces the amount of calculation. In addition, through a step-by-step convergence approach, the optimization results of each stage can be used as the starting point of the next stage, avoiding a comprehensive search from scratch, further saving computing resources, and thus reducing the difficulty of learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 A schematic flow chart of a method for predicting a reasonable completed state of a cable-stayed bridge based on deep learning provided by the present invention;

[0089] Figure 2 A schematic diagram of the algorithm flow of the fused convolution kernel provided by the present invention. DETAILED DESCRIPTION

[0090] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0091] Embodiment 1: A reasonable bridge state prediction method for a cable-stayed bridge based on deep learning, such as Figure 1 As shown, the following steps are included:

[0092] S1: Determine the construction target of the cable-stayed bridge, collect the historical cable-stayed bridge completion status data, define the state space of the reasonable cable-stayed bridge completion status prediction model, calculate the state space initialization parameters, and initialize the state space;

[0093] S11: Determine the construction target of the cable-stayed bridge, including the target main beam length, target load and target material properties;

[0094] Collect historical cable-stayed bridge status data, including historical main beam length, historical load, historical material properties, historical main tower height, historical number of cables, and historical upper and lower anchor point coordinates of cables;

[0095] Define the state space of the reasonable completed state prediction model of the cable-stayed bridge, including the target main tower height, target number of cables, and target upper and lower anchor point coordinates of the cables;

[0096] According to the historical cable-stayed bridge state data, the state space initialization parameters are calculated, including the target main tower height, the target number of cables, and the specific values ​​of the upper and lower anchor point coordinates of the target cables. The calculation method is:

[0097] ;

[0098] in, is the preliminary main tower height, is the amount of historical cable-stayed bridge status data, is the similarity weight between the target material attribute and the i-th historical material attribute, i is the data index of the completed state of the historical cable-stayed bridge, is the height of the i-th historical main tower, is the target load, is the historical load of the ith article, is the target main beam length, is the length of the i-th historical main beam; is the target material property vector, is the attribute vector of the i-th historical material, To take the modulus length, is a natural constant, is the regulating factor, To take the absolute value;

[0099] S12: Calculate the target main tower height based on the preliminary main tower height, the status characteristics of historical cable-stayed bridges, and the difference characteristics of historical cable-stayed bridges:

[0100] ;

[0101] in, The characteristics of the completed state of the historical cable-stayed bridge. is a multi-layer perceptron, The differences in the status of historical cable-stayed bridges are shown in Figure 1. is the target main tower height;

[0102] S13: Extract the characteristics of the target cable-stayed bridge’s completed state based on the target main tower height, target load, target main beam length and target material properties:

[0103] ;

[0104] S14: Perform similarity matching based on the target cable-stayed bridge completion state characteristics, and select the historical cable-stayed bridge completion state data with the highest similarity to obtain the target cable quantity and the upper and lower anchor point coordinates of the target cables:

[0105] ;

[0106] in, is the similarity between the completed state characteristics of the target cable-stayed bridge and the completed state characteristics of the i-th historical cable-stayed bridge, is the cosine similarity calculation, The data index of the historical cable-stayed bridge status with the highest similarity. To get the maximum value index i;

[0107] Calculate the historical cable-stayed bridge status data index with the highest similarity Then select The number of historical cables and the coordinates of the upper and lower anchor points of the historical cables in the completed state data of the historical cable-stayed bridges are used as the target number of cables and the coordinates of the upper and lower anchor points of the target cables;

[0108] S15: Initializing the state space of the reasonable completed state prediction model of the cable-stayed bridge according to the calculated state space initialization parameters.

[0109] It is further explained that the present invention uses a multi-layer perceptron and cosine similarity to evaluate the similarity between the target cable-stayed bridge and the historical cable-stayed bridge in terms of their completion status, and accordingly assigns weights to the historical data to calculate the preliminary main tower height; at the same time, the multi-layer perceptron is also used to encode the state characteristics of the historical cable-stayed bridge, and by comparing the differences between the completion status characteristics of the target cable-stayed bridge and the completion status characteristics of the historical cable-stayed bridge, the preliminary main tower height of the main tower height is further fine-tuned; finally, by selecting the historical cable-stayed bridge that is most similar to the completion status of the target cable-stayed bridge, the initial values ​​of the target cable quantity and the upper and lower anchor point coordinates are directly determined;

[0110] In reinforcement learning tasks, the initialization of the state space directly affects the exploration efficiency and convergence speed of the algorithm. Traditional state space initialization methods are often initialized to all zeros or random values. Although this method is simple and easy, it will lead to blind exploration of the algorithm in the learning process and slow down the convergence speed. The present invention initializes the state space by drawing on historical data, which can provide a more reasonable and optimal starting point for the optimization algorithm, making the optimization process more in line with engineering practice. The algorithm can converge to the optimal solution faster, thereby significantly improving the optimization efficiency and providing an accurate and efficient parameter initialization scheme for the reasonable prediction of the completed bridge state of the cable-stayed bridge.

[0111] For example:

[0112] Historical cable-stayed bridge status data (n=3):

[0113] Data 1: ;

[0114] Data 2: ;

[0115] Data 3: ;

[0116] Target cable-stayed bridge parameters:

[0117] ;

[0118] Calculate the preliminary main tower height , and then further adjust the target main tower height through the multi-layer perceptron model , after calculating the similarity, find the most similar historical data index , the number of target cables in this data is 10, and the coordinates of the upper and lower anchor points of the target cables are: . Then take the target main tower height , the target number of cables is 10, and the coordinates of the upper and lower anchor points of the target cables are: , as the state space initialization parameter.

[0119] S2: According to the construction target of the cable-stayed bridge and the initialization parameters of the state space, a finite element model of the cable-stayed bridge is constructed, and the force parameters and deformation parameters of the cable-stayed bridge are calculated;

[0120] S21: According to the target main beam length and the target material properties, a finite element model of the cable-stayed bridge is constructed in the finite element modeling software, and the target material properties are assigned to the main beam in the finite element model of the cable-stayed bridge;

[0121] S22: Add main towers and cables to the finite element model of the cable-stayed bridge according to the state space initialization parameters;

[0122] S23: Apply loads, including dead loads and dynamic loads, to the finite element model of the cable-stayed bridge according to the target load amount;

[0123] S24: Calculate the stress parameters and deformation parameters of the cable-stayed bridge by using finite element modeling software. The stress parameters of the cable-stayed bridge include the internal force of the main beam, the cable force and the internal force of the main tower. The deformation parameters of the cable-stayed bridge include the deflection of the main beam, the displacement of the main tower and the elongation of the cable.

[0124] S3: According to the stress parameters and deformation parameters of the cable-stayed bridge, the original morphological image of the cable-stayed bridge is generated by simulation software, and then the original morphological image of the cable-stayed bridge is preprocessed to obtain a preprocessed cable-stayed bridge image; then, a convolutional neural network is used to extract the morphological features of the cable-stayed bridge from the preprocessed cable-stayed bridge image, and the state space is enhanced to obtain an enhanced state space;

[0125] S31: In the simulation software, force parameters and deformation parameters of the cable-stayed bridge are input, the cable-stayed bridge is simulated, and an original shape image of the cable-stayed bridge is obtained;

[0126] S32: graying and downsampling the original image of the cable-stayed bridge to obtain a preprocessed image of the cable-stayed bridge, wherein the pixel size of the preprocessed image of the cable-stayed bridge is 640×640;

[0127] S33: Construct a cable-stayed bridge morphological feature extraction model, and extract the morphological features of the cable-stayed bridge according to the preprocessed cable-stayed bridge image, such as Figure 2 As shown, the calculation method is:

[0128] ;

[0129] in, It is the first morphological feature of the cable-stayed bridge. is a convolutional neural network, is the preprocessed cable-stayed bridge image, is the convolution fusion gate, is the convolution fusion weight, is the convolution fusion bias vector, To fuse the convolution kernel, is the Gaussian kernel, is the edge detection kernel, The second morphological feature of the cable-stayed bridge is is the sigmoid function, The first morphological feature of a cable-stayed bridge The number of feature maps contained in , j is the feature map index, is the convolution operation, is the jth characteristic graph of the first morphological feature of the cable-stayed bridge, is the morphological characteristic offset vector of the cable-stayed bridge, The morphological characteristics of cable-stayed bridges are: For a convolutional neural network using fused convolution kernels, For channel splicing;

[0130] S34: According to the morphological characteristics of the cable-stayed bridge, the state space of the reasonable bridge state prediction model of the cable-stayed bridge is enhanced to obtain an enhanced state space; the enhanced state space includes the target main tower height, the target number of cables, the coordinates of the upper and lower anchor points of the target cables, and the morphological characteristics of the cable-stayed bridge.

[0131] It is further explained that step S33 of the present invention introduces a convolution fusion gate and a fusion convolution kernel, and fuses the characteristics of the Gaussian kernel and the edge detection kernel with the feature map in the first morphological feature of the cable-stayed bridge, so that the feature contains both the global structural information of the image (Gaussian kernel) and the edge and detail information (edge ​​detection kernel), thereby enhancing the expressive power of the feature; then, the fused features are nonlinearly transformed by the sigmoid function, and combined with the initially extracted feature map, the diversity of the features is further integrated and enhanced; finally, the first morphological feature of the cable-stayed bridge, the second morphological feature of the cable-stayed bridge, and the preprocessed cable-stayed bridge image are channel-joined, and feature extraction is performed by a deep convolutional neural network, thereby realizing multi-scale and multi-level feature integration and utilization;

[0132] Traditional convolutional neural networks generally use fixed convolution kernels to extract image features, while the present invention integrates the characteristics of the Gaussian kernel and the edge detection kernel into the convolution kernel according to the characteristics of the cable-stayed bridge image. This adaptive convolution kernel design can more accurately capture the key features in the cable-stayed bridge image, so that the model has good generalization ability for different types of cable-stayed bridge images and different bridge completion states, and further captures subtle changes in the cable-stayed bridge completion state, thereby achieving accurate prediction of the bridge completion state.

[0133] S4: Define the action space, value function and reward function of the reasonable bridge state prediction model for cable-stayed bridges;

[0134] S41: define the action space of the reasonable bridge state prediction model of the cable-stayed bridge, including: main tower height +1, main tower height -1, number of cables +1, number of cables -1, cable upper anchor point coordinate +1, cable upper anchor point coordinate -1, cable lower anchor point coordinate +1, cable lower anchor point coordinate -1;

[0135] S42: The value function of Q-learning is used as the value function of the reasonable bridge state prediction model of the cable-stayed bridge;

[0136] S43: Define the reward function of the reasonable bridge state prediction model of the cable-stayed bridge, and the calculation method is:

[0137] ;

[0138] in, is the reward for executing an action in the current state, s is the current state, a is the action to be executed, Reward for structural safety, Reward for structural stability, Rewards for exploration;

[0139] ;

[0140] in, is the safety bonus coefficient, To obtain the maximum value, For new status The stress index under is the stress index threshold; The new status The internal forces of the main beam, cable and main tower below;

[0141] ;

[0142] in, is the stability reward coefficient, For new status The displacement index under is the displacement index threshold; The new status The deflection of the main beam, the displacement of the main tower and the elongation of the cables;

[0143] ;

[0144] in, To explore the reward coefficient, State-action pairing in reasonable bridge state prediction model for cable-stayed bridges The number of visits, is the total number of visits.

[0145] Further, step S43 of the present invention defines a reward function of a reasonable bridge state prediction model for a cable-stayed bridge, which comprehensively considers three factors: structural safety, structural stability, and exploration.

[0146] In the structural safety reward, the reward function is proportional to the weighted sum of the internal forces of the main beam, the cable forces and the main tower. Therefore, the algorithm tends to select actions that can make the internal force distribution more uniform and reasonable, thereby optimizing the overall stress performance of the cable-stayed bridge. At the same time, the stress index and stress index threshold are introduced. The structural safety reward can ensure that the stress level of the cable-stayed bridge during the design process does not exceed the allowable range, thereby effectively ensuring the safety of the structure.

[0147] In the structural stability reward, the reward function is proportional to the square of the weighted sum of the main beam deflection, main tower offset and cable elongation. The algorithm tends to select actions that can reduce displacement and increase structural stiffness, thereby optimizing the displacement control performance of the cable-stayed bridge. The ratio of the displacement index to the displacement index threshold can limit the displacement level of the cable-stayed bridge during the design process and ensure the stability of the structure.

[0148] In the exploratory reward, by introducing the ratio of the number of visits to the state-action pair to the total number of visits and applying an exponential decay function, the reward function can encourage the algorithm to explore more unknown states during the training process and avoid falling into the local optimal solution; as the training progresses, the number of visits increases, the exploratory reward gradually decreases, and the algorithm gradually converges to a stable optimal solution or suboptimal solution.

[0149] For example, parameter settings:

[0150] Safety bonus factor ;

[0151] Stability Reward Coefficient ;

[0152] Exploration Reward Coefficient ;

[0153] Stress Index Threshold ;

[0154] Displacement index threshold ;

[0155] Total visits ;

[0156] Current state-action pair access count ;

[0157] Current state(s) and action(a):

[0158] Main beam internal force ;

[0159] Cable force ;

[0160] Main tower internal force ;

[0161] Main beam deflection ;

[0162] Main tower offset ;

[0163] Cable elongation ;

[0164] but:

[0165] ;

[0166] Similarly, substituting the data into the formula yields:

[0167] ;

[0168] In this example, since the stress index does not exceed the threshold, the safety reward is 0; the stability reward is calculated based on the ratio of the weighted sum of the displacement index and the threshold, which is a negative value, indicating that there is a certain penalty on the stability of the current state; the exploration reward is calculated based on the ratio of the number of visits to the current state-action pair to the total number of visits, which is a positive value, encouraging the algorithm to continue exploring; the final total reward is the sum of these three parts of the reward, which is a positive value, indicating that this action is a relatively good choice in the current state.

[0169] S5: Design a segmented optimization training strategy, use the reward function to perform reinforcement learning training on the reasonable completion state prediction model of the cable-stayed bridge, and use the state space at the end of the training as the reasonable completion state of the cable-stayed bridge;

[0170] S51: Design a segmented optimization training strategy to optimize the target main tower height in the first stage of reinforcement learning training, optimize the target number of cables in the second stage, and optimize the upper and lower anchor point coordinates of the target cables in the third stage;

[0171] S52: Initialize the Q table of the reasonable completed bridge state prediction model of the cable-stayed bridge according to the state space initialization parameters. First, optimize the target main tower height until the target main tower height converges. The optimization calculation method is:

[0172] ;

[0173] in, is the state-action pair in the Q table The value of To update the symbol, is the learning rate, is the discount factor, is the state-action pair in the Q table the value of

[0174] S53: After the target main tower height converges, fix it as a constant and enter the second stage to optimize the target number of cables; after the target number of cables converges, fix both the target main tower height and the target number of cables as constants and enter the third stage to optimize the upper and lower anchor point coordinates of the target cables. The optimization method of the target number of cables and the optimization method of the upper and lower anchor point coordinates of the target cables are the same as step S52;

[0175] S54: After the optimization of the coordinates of the upper and lower anchor points of the target cables is completed, the state space of the reasonable completed state prediction model of the cable-stayed bridge is output as the reasonable completed state of the cable-stayed bridge.

[0176] It is further explained that the segmented optimization strategy of step S5 of the present invention decomposes the complex optimization problem into multiple smaller sub-problems, each of which focuses on only one or a few optimization objectives, so that the model only needs to consider the optimization objective of the current stage each time iterates, without having to process all possible parameter combinations at the same time, thereby greatly reducing the amount of calculation; in addition, through a gradual convergence approach, the optimization results of each stage can be used as the starting point of the next stage, avoiding a comprehensive search from scratch, and further saving computing resources; in the initial stage, the model only needs to focus on the optimization of the main tower height, which helps the model to understand and learn this specific task more quickly; as the optimization progresses, the model gradually introduces more parameters (such as the number of cables, the coordinates of the upper and lower anchor points of the cables), but each time is based on the previous stage, so the model can more easily adapt to these new optimization objectives; this method of gradually increasing complexity helps to avoid the model facing overly complex optimization problems in the initial stage, thereby reducing the difficulty of learning.

[0177] Embodiment 2: The present invention also discloses a reasonable bridge state prediction system for a cable-stayed bridge based on deep learning, comprising:

[0178] Cable-stayed bridge state space initialization module: determine the construction target of the cable-stayed bridge, collect the historical cable-stayed bridge state data, define the state space of the reasonable cable-stayed bridge state prediction model, calculate the state space initialization parameters, and perform state space initialization;

[0179] Cable-stayed bridge finite element model construction module: According to the cable-stayed bridge construction objectives and state space initialization parameters, the cable-stayed bridge finite element model is constructed, and the force parameters and deformation parameters of the cable-stayed bridge are calculated;

[0180] State space enhancement module: According to the stress parameters and deformation parameters of the cable-stayed bridge, the original morphological image of the cable-stayed bridge is generated by simulation software, and then the original morphological image of the cable-stayed bridge is preprocessed, and then the morphological features of the cable-stayed bridge are extracted by using a convolutional neural network, and the state space is enhanced to obtain the enhanced state space;

[0181] Reinforcement learning design module: defines the action space, value function and reward function of the reasonable bridge state prediction model of the cable-stayed bridge;

[0182] Reasonable bridge completion state prediction module: Design a segmented optimization training strategy, use the reward function to perform reinforcement learning training on the reasonable bridge completion state prediction model of the cable-stayed bridge, and use the state space at the end of the training as the reasonable bridge completion state of the cable-stayed bridge.

[0183] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0184] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0185] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A reasonable bridge state prediction method for cable-stayed bridges based on deep learning, characterized in that: The following steps are involved: S1: Determine the construction target of the cable-stayed bridge, collect the historical cable-stayed bridge completion status data, define the state space of the reasonable cable-stayed bridge completion status prediction model, calculate the state space initialization parameters, and initialize the state space, including: Calculate the preliminary main tower height based on the historical cable-stayed bridge status data; Calculate the target main tower height based on the preliminary main tower height, the status characteristics of historical cable-stayed bridges, and the difference characteristics of historical cable-stayed bridges; Extract the characteristics of the target cable-stayed bridge's completed state based on the target main tower height, target load, target main beam length and target material properties; Similarity matching is performed based on the completed state characteristics of the target cable-stayed bridge, and the historical completed state data of the cable-stayed bridge with the highest similarity is selected to initialize the state space; S2: According to the construction target of the cable-stayed bridge and the initialization parameters of the state space, a finite element model of the cable-stayed bridge is constructed, and the force parameters and deformation parameters of the cable-stayed bridge are calculated; S3: According to the stress parameters and deformation parameters of the cable-stayed bridge, the original morphological image of the cable-stayed bridge is generated by simulation software, and the original morphological image of the cable-stayed bridge is preprocessed to obtain a preprocessed cable-stayed bridge image; the morphological features of the cable-stayed bridge are extracted from the preprocessed cable-stayed bridge image by using a convolutional neural network; the state space is enhanced according to the morphological features of the cable-stayed bridge to obtain an enhanced state space; S4: Define the action space, value function and reward function of the reasonable bridge state prediction model for cable-stayed bridges; The reward function of the reasonable bridge state prediction model of the cable-stayed bridge is defined and calculated as follows: ； in, is the reward for executing an action in the current state, s is the current state, a is the action to be executed, Reward for structural safety, Reward for structural stability, Rewards for exploration; ； in, is the safety bonus coefficient, To obtain the maximum value, For new status The stress index under is the stress index threshold; The new status The internal forces of the main beam, cable and main tower below; ； in, is the stability reward coefficient, For new status The displacement index under is the displacement index threshold; The new status The deflection of the main beam, the displacement of the main tower and the elongation of the cables; ； in, To explore the reward coefficient, State-action pairing in reasonable bridge state prediction model for cable-stayed bridges The number of visits, is the total number of visits; S5: Design a segmented optimization training strategy, use the reward function to perform reinforcement learning training on the reasonable completion state prediction model of the cable-stayed bridge, and use the state space at the end of the training as the reasonable completion state of the cable-stayed bridge, including: S51: Design a segmented optimization training strategy, and use the reward function to perform reinforcement training on the reasonable bridge state prediction model of the cable-stayed bridge. In the first stage of reinforcement learning training, the target main tower height is optimized, the target number of cables is optimized in the second stage, and the upper and lower anchor point coordinates of the target cables are optimized in the third stage. S52: Initialize the Q table of the reasonable completed bridge state prediction model of the cable-stayed bridge according to the state space initialization parameters. First, optimize the target main tower height until the target main tower height converges. The optimization calculation method is: ； in, is the state-action pair in the Q table The value of To update the symbol, is the learning rate, is the reward for executing an action in the current state, s is the current state, a is the action to be executed, To obtain the maximum value, is the discount factor, is the state-action pair in the Q table The value of For the new state, For new status Next, execute the action; S53: After the target main tower height converges, fix it as a constant and enter the second stage to optimize the target number of cables; after the target number of cables converges, fix both the target main tower height and the target number of cables as constants and enter the third stage to optimize the upper and lower anchor point coordinates of the target cables. The optimization method of the target number of cables and the upper and lower anchor point coordinates of the target cables is the same as step S52; S54: After the optimization of the coordinates of the upper and lower anchor points of the target cables is completed, the state space of the reasonable completed state prediction model of the cable-stayed bridge is output as the reasonable completed state of the cable-stayed bridge.

2. The method for predicting reasonable completed state of a cable-stayed bridge based on deep learning according to claim 1 is characterized in that: The S1 step includes: S11: Determine the construction target of the cable-stayed bridge, including the target main beam length, target load and target material properties; Collect historical cable-stayed bridge status data, including historical main beam length, historical load, historical material properties, historical main tower height, historical number of cables, and historical upper and lower anchor point coordinates of cables; Define the state space of the reasonable completed state prediction model of the cable-stayed bridge, including the target main tower height, target number of cables, and target upper and lower anchor point coordinates of the cables; According to the historical cable-stayed bridge state data, the state space initialization parameters are calculated, including the target main tower height, the target number of cables, and the target upper and lower anchor point coordinates of the cables: The preliminary calculation of the main tower height is: ； in, is the preliminary main tower height, is the amount of historical cable-stayed bridge status data, is the similarity weight between the target material attribute and the i-th historical material attribute, i is the data index of the completed state of the historical cable-stayed bridge, is the height of the i-th historical main tower, is the target load, is the historical load of the ith article, is the target main beam length, is the length of the i-th historical main beam; is the target material property vector, is the attribute vector of the i-th historical material, To take the modulus length, is a natural constant, is the regulating factor, To take the absolute value; S12: Calculate the target main tower height based on the preliminary main tower height, the status characteristics of historical cable-stayed bridges, and the difference characteristics of historical cable-stayed bridges: ； in, The characteristics of the completed state of the historical cable-stayed bridge. is a multi-layer perceptron, The differences in the status of historical cable-stayed bridges are shown in Figure 1. is the target main tower height; S13: Extract the characteristics of the target cable-stayed bridge’s completed state based on the target main tower height, target load, target main beam length, and target material properties: ； in, The completed state characteristics of the target cable-stayed bridge; S14: Perform similarity matching based on the target cable-stayed bridge completion state characteristics, and select the historical cable-stayed bridge completion state data with the highest similarity to obtain the target cable quantity and the upper and lower anchor point coordinates of the target cables: ； in, is the similarity between the completed state characteristics of the target cable-stayed bridge and the completed state characteristics of the i-th historical cable-stayed bridge, is the cosine similarity calculation, The data index of the historical cable-stayed bridge status with the highest similarity. To get the maximum value index i; Calculate the historical cable-stayed bridge status data index with the highest similarity Then select The number of historical cables and the coordinates of the upper and lower anchor points of the historical cables in the completed state data of the historical cable-stayed bridges are used as the target number of cables and the coordinates of the upper and lower anchor points of the target cables; S15: Initializing the state space of the reasonable completed state prediction model of the cable-stayed bridge according to the calculated state space initialization parameters.

3. The method for predicting reasonable completed state of a cable-stayed bridge based on deep learning according to claim 2 is characterized in that: The S2 step includes: S21: According to the target main beam length and the target material properties, a finite element model of the cable-stayed bridge is constructed in the finite element modeling software, and the target material properties are assigned to the main beam in the finite element model of the cable-stayed bridge; S22: Add main towers and cables to the finite element model of the cable-stayed bridge according to the state space initialization parameters; S23: Apply loads, including dead loads and dynamic loads, to the finite element model of the cable-stayed bridge according to the target load amount; S24: Calculate the stress parameters and deformation parameters of the cable-stayed bridge by using finite element modeling software. The stress parameters of the cable-stayed bridge include the internal force of the main beam, the cable force and the internal force of the main tower. The deformation parameters of the cable-stayed bridge include the deflection of the main beam, the displacement of the main tower and the elongation of the cable.

4. The method for predicting reasonable completed state of a cable-stayed bridge based on deep learning according to claim 3 is characterized in that: The step of extracting the morphological features of the cable-stayed bridge by using a convolutional neural network in S3 includes: Firstly, a cable-stayed bridge morphological feature extraction model is constructed, the first morphological feature of the cable-stayed bridge is extracted according to the preprocessed cable-stayed bridge image, and the convolution fusion gate is calculated based on the first morphological feature of the cable-stayed bridge; Combine the convolution fusion gate to fuse the Gaussian kernel and the edge detection kernel to obtain the fused convolution kernel; The morphological features of the cable-stayed bridge are extracted by fusing the convolution kernel and channel splicing.

5. The method for predicting reasonable completed state of a cable-stayed bridge based on deep learning according to claim 4 is characterized in that: The S3 step further includes: S31: In the simulation software, force parameters and deformation parameters of the cable-stayed bridge are input, the cable-stayed bridge is simulated, and an original shape image of the cable-stayed bridge is obtained; S32: graying and downsampling the original image of the cable-stayed bridge to obtain a preprocessed image of the cable-stayed bridge, wherein the pixel size of the preprocessed image of the cable-stayed bridge is 640×640; S33: Construct a cable-stayed bridge morphological feature extraction model, and extract the morphological features of the cable-stayed bridge according to the preprocessed cable-stayed bridge image. The calculation method is: ； in, It is the first morphological feature of the cable-stayed bridge. is a convolutional neural network, is the preprocessed cable-stayed bridge image, is the convolution fusion gate, is the convolution fusion weight, is the convolution fusion bias vector, To fuse the convolution kernel, is the Gaussian kernel, is the edge detection kernel, The second morphological feature of the cable-stayed bridge is is the sigmoid function, The first morphological feature of a cable-stayed bridge The number of feature maps contained in , j is the feature map index, is the convolution operation, is the jth characteristic graph of the first morphological feature of the cable-stayed bridge, is the morphological characteristic offset vector of the cable-stayed bridge, The morphological characteristics of cable-stayed bridges are: For a convolutional neural network using fused convolution kernels, For channel splicing; S34: According to the morphological characteristics of the cable-stayed bridge, the state space of the reasonable bridge state prediction model of the cable-stayed bridge is enhanced to obtain an enhanced state space; the enhanced state space includes the target main tower height, the target number of cables, the coordinates of the upper and lower anchor points of the target cables, and the morphological characteristics of the cable-stayed bridge.

6. The method for predicting reasonable completed state of a cable-stayed bridge based on deep learning according to claim 5 is characterized in that: The S4 step includes: S41: define the action space of the reasonable bridge state prediction model of the cable-stayed bridge, including: main tower height +1, main tower height -1, number of cables +1, number of cables -1, cable upper anchor point coordinate +1, cable upper anchor point coordinate -1, cable lower anchor point coordinate +1, cable lower anchor point coordinate -1; S42: The value function of Q-learning is used as the value function of the reasonable bridge state prediction model of the cable-stayed bridge.

7. A reasonable bridge state prediction system for cable-stayed bridges based on deep learning, characterized in that: include: Cable-stayed bridge state space initialization module: determine the construction target of the cable-stayed bridge, collect the historical cable-stayed bridge state data, define the state space of the reasonable cable-stayed bridge state prediction model, calculate the state space initialization parameters, and perform state space initialization; Cable-stayed bridge finite element model construction module: According to the cable-stayed bridge construction objectives and state space initialization parameters, the cable-stayed bridge finite element model is constructed, and the force parameters and deformation parameters of the cable-stayed bridge are calculated; State space enhancement module: According to the stress parameters and deformation parameters of the cable-stayed bridge, the original morphological image of the cable-stayed bridge is generated by simulation software, and then the original morphological image of the cable-stayed bridge is preprocessed, and then the morphological features of the cable-stayed bridge are extracted by using a convolutional neural network, and the state space is enhanced to obtain the enhanced state space; Reinforcement learning design module: defines the action space, value function and reward function of the reasonable bridge state prediction model of the cable-stayed bridge; Reasonable bridge state prediction module: Design a segmented optimization training strategy, use the reward function to perform reinforcement learning training on the reasonable bridge state prediction model of the cable-stayed bridge, and use the state space at the end of the training as the reasonable bridge state of the cable-stayed bridge; To realize the reasonable bridge status prediction method of a cable-stayed bridge based on deep learning as described in any one of claims 1-6.

Images