A method for evaluating the safety of a highway bridge structure approaching a hydropower station

Abstract

The application provides a method for evaluating the safety of a highway bridge structure of a hydropower station. The method mainly comprises the following steps: collecting past bridge engineering project data; preprocessing the data; establishing a suitable neural network model using the processed data; training and optimizing the model using training set data; verifying and testing the model using validation set and test set data; using the trained model to evaluate the safety of the bridge structure; adjusting the model according to the actual bridge structure safety feedback; applying the evaluation results to practice to make decisions on bridge maintenance, reconstruction, demolition, etc.; recording and updating the new bridge structure safety evaluation results for the next evaluation; and establishing a continuous monitoring system for regular or as-needed evaluation. The method can effectively evaluate the safety of the bridge structure and improve the safety management level of the bridge engineering.

Description

Technical Field

[0001] This invention belongs to the field of construction safety, and more specifically relates to a method for assessing the structural safety of a bridge on a highway leading to a hydropower station. Background Technology

[0002] Assessing the safety of bridge structures is a complex task. Traditionally, this typically involves regular on-site inspections and maintenance by teams of engineers and experts. On-site personnel usually need to conduct detailed visual inspections of various parts of the bridge, looking for any potential structural damage such as cracks, peeling, corrosion, or displacement deformation. Such regular inspections and maintenance can consume significant human, material, and financial resources for large bridge projects.

[0003] However, this traditional assessment method has a problem: the extent of structural damage is often highly dependent on the engineer's experience and judgment, inevitably introducing subjectivity. Even experienced engineers may miss minor but critical damage, which could have a serious impact on the overall safety of the bridge.

[0004] Furthermore, most existing bridge inspection systems are rule-based, typically requiring a predefined set of rules and criteria to identify potential damage types and degrees. However, this approach often fails to adequately handle complex and variable real-world situations. For example, bridge safety can be influenced by numerous factors, including but not limited to: bridge design, materials, construction process, ongoing fatigue wear during use, and the climate and environmental conditions in which the bridge is located. These complex factors are often difficult to adequately consider and address in traditional rule-based systems.

[0005] Therefore, in some cases, the inability to comprehensively and accurately assess all possible influencing factors renders traditional assessment methods ineffective in providing accurate safety assessments. This undoubtedly presents significant difficulties and challenges for ensuring the safety of bridge structures, providing effective references for bridge maintenance and renovation, and preventing potential structural failures. Summary of the Invention

[0006] This invention employs a neural network model for the safety assessment of bridge structures on access roads to hydropower stations, enabling effective large-scale bridge safety assessments. Neural networks possess powerful self-learning and adaptive capabilities, allowing them to process and analyze massive amounts of data, uncover deep connections between data points, and comprehensively consider multiple factors affecting bridges, such as design, materials, construction, and environment. Furthermore, the neural network-based assessment method can utilize historical data for training and prediction, reducing reliance on manual intervention and improving the accuracy and efficiency of the assessment.

[0007] To achieve the above objectives, the present invention employs the following technical solution: the security assessment method includes: Step 1: Data Collection Collect data from past bridge engineering projects, such as bridge design, structure, environmental impact factors, construction process and status, maintenance process and status, and human factors; Step 2: Data Preprocessing The collected data is preprocessed, including cleaning, standardization, and normalization, to remove noise and outliers, making the data suitable for neural network models.

[0008] Step 3: Build a neural network model Utilize the collected data to build a suitable type of neural network model; Step 4: Training and Optimizing the Model The model is trained using the training set data. During the training process, the weights and biases of the model are continuously adjusted to make it fit the training data as closely as possible. At the same time, the model can be optimized by increasing the number of training iterations, changing the learning rate, adding regularization terms, etc. Step 5: Validate and test the model The performance of the model is validated and tested using validation and test set data, and the predictive ability of the model is evaluated. Step 6: Assess the safety of the bridge structure Furthermore, the data collection in step 1 is specifically as follows: (1) Good project record: Collect data from relevant bridge engineering projects, including design documents, construction reports, inspection reports, and maintenance records; (2) Weather and environmental data: Collect historical weather records and environmental information of the bridge location, such as temperature, humidity, wind speed, rainfall, geological structure, soil type, etc. (3) Expert interviews and questionnaires: Interviews and questionnaires were conducted with engineers, bridge designers, maintenance personnel and other relevant professionals to collect their views and experiences on bridge structural safety; (4) On-site investigation and testing: Combine on-site inspection and field testing to obtain the actual condition of the bridge, and collect data such as the bridge's structural type, materials and various testing indicators; (5) Public data and literature: Obtain relevant literature on bridge engineering through public channels such as the Internet and libraries.

[0009] Furthermore, the specific steps of data preprocessing in step 2 are as follows: Data standardization: This mainly involves scaling the data proportionally to make it fall within a defined small range. The main purpose of data standardization is to remove the influence of data units and eliminate the order-of-magnitude differences between data points. Z-Score standardization is used, and the formula is as follows: (1) Where X is a specific sample value, μ is the mean of all samples, and σ is the standard deviation of all samples; Data normalization: Transform all data to the range [0,1] or [-1,1]. The normalization method used is MinMax normalization, and the formula is as follows: (2) X is a specific sample value, and min and max are the minimum and maximum values ​​of all samples, respectively.

[0010] Furthermore, in step 3, the neural network model is established by introducing the ant colony optimization algorithm into the training process of SOM, which helps the network to better find the global optimal distribution of samples. Specifically, the ant colony algorithm guides the direction of weight modification. In each iteration, the ants search in the weight space based on pheromones and then update the weights. This not only maintains the unsupervised characteristics of the SOM network, but also enables the network to have better global search capabilities during training, thereby obtaining more accurate results. S301 Initialize the weight vector: For each neuron i (i=1,2,…N), assign it a weight vector with the same dimension as the input vector x. Each of its elements is generated by a random number between 0 and 1; S302 neurons win the competition: For a randomly selected input vector x(t), calculate the distance to each neuron i, where the distance function is: (3) if Neuron j wins the competition against all ; S303 updates the weights of the winning neuron and its neighboring neurons; (4) Let represent the weight vector of neuron v at time t. Here, t is the learning rate, and t is the time. It is a neighborhood function, representing the degree of influence of neuron i on neuron v; It is a constant in the range [0,1] used to balance the influence between the weight update of SOM and the weight update guided by ant colony algorithm; This represents the pheromone increment between neuron v and neuron i; The weight vector representing the highest pheromone concentration in history is the same as the weight vector corresponding to the best path in the ant colony algorithm. S304 updates the neighborhood width and learning rate; Steps 2 and 3 are repeated until the SOM network reaches stability. As for pheromone updates, they are determined by the ACO algorithm and are performed according to a certain pheromone evaporation rate and pheromone increment.

[0011] Furthermore, the specific steps of step 4: training and optimizing the model are as follows: S401 Initialize Weight Vector: For each neuron i (i=1,2,…N), assign it a weight vector with the same dimension as the input vector x. Each of its elements is generated by a random number between 0 and 1; S402 neurons win the competition: For a randomly selected input vector x(t), calculate the distance to each neuron i, where the distance function is: (5) if Neuron j wins the competition against all ; S403 updates the weights of the winning neuron and its neighboring neurons; (6) S404 updates neighborhood width and learning rate: After each iteration, the neighborhood width σ(t) and learning rate η(t) are updated according to a preset method, using the following exponential function: (7) (8) in and These are the initial learning rate and neighborhood width, and τ1 and τ2 are time constants; S405 determines the termination conditions: These conditions include the maximum number of iterations, the model convergence speed, etc. S406 Model Optimization: After the model training is completed, it needs to be optimized by adding L2 regularization; L2 regularization, also known as weight decay, is a method to control model complexity by adding a penalty to the weight size in the loss function, thereby preventing overfitting; it mainly targets weight parameters. Without regularization, the general form of the loss function of a neural network is: (9) When L2 regularization is introduced, the loss function becomes: (10) Here, λ is the regularization parameter, which controls the degree of influence of the penalty term on the loss function. It is the sum of squares of all weight parameters w; during training, when updating the weights using gradient descent, the addition of a squared penalty to the weight parameters w in the loss function causes the weights to decrease after each update, effectively preventing model overfitting. Finally, the optimal value of λ is usually determined through cross-validation. Specifically, the model is trained on the validation set with different values ​​of λ, and then the λ that best achieves the desired evaluation metrics (such as accuracy, F1 score, etc.) on the validation set is selected.

[0012] Furthermore, step 6, assessing the safety of the bridge structure, is conducted in the following specific manner: A trained neural network model is used to conduct a safety assessment of the bridge structure on the access road to the hydropower station. The existing conditions and data of the bridge are input, and the model makes predictions to obtain the safety level of the bridge structure.

[0013] Beneficial effects of this invention: This invention employs a neural network model for the safety assessment of bridge structures on access roads to hydropower stations, enabling effective large-scale bridge safety assessments. Neural networks possess powerful self-learning and adaptive capabilities, allowing them to process and analyze massive amounts of data, uncover deep connections between data points, and comprehensively consider multiple factors affecting bridges, such as design, materials, construction, and environment. Furthermore, the neural network-based assessment method can utilize historical data for training and prediction, reducing reliance on manual intervention and improving the accuracy and efficiency of the assessment. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the model training and optimization process of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0016] In the description of this invention, it should be noted that the terms "upper", "lower", "inner", "outer", "top / bottom", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0017] like Figure 1 As shown, the security assessment method includes: Step 1: Data Collection Collect data from past bridge engineering projects, such as bridge design, structure, environmental impact factors, construction process and status, maintenance process and status, and human factors; (1) Good project record: Collect data from relevant bridge engineering projects, including design documents, construction reports, inspection reports, and maintenance records; (2) Weather and environmental data: Collect historical weather records and environmental information of the bridge location, such as temperature, humidity, wind speed, rainfall, geological structure, soil type, etc. (3) Expert interviews and questionnaires: Interviews and questionnaires were conducted with engineers, bridge designers, maintenance personnel and other relevant professionals to collect their views and experiences on bridge structural safety; (4) On-site investigation and testing: Combine on-site inspection and field testing to obtain the actual condition of the bridge, and collect data such as the bridge's structural type, materials and various testing indicators; (5) Public data and literature: Obtain relevant literature on bridge engineering through public channels such as the Internet and libraries.

[0018] Step 2: Data Preprocessing The collected data is preprocessed, including cleaning, standardization, and normalization, to remove noise and outliers, making the data suitable for neural network models.

[0019] The specific steps of data preprocessing in step 2 are as follows: Data standardization: This mainly involves scaling the data proportionally to make it fall within a defined small range. The main purpose of data standardization is to remove the influence of data units and eliminate the order-of-magnitude differences between data points. Z-Score standardization is used, and the formula is as follows: (1) Where X is a specific sample value, μ is the mean of all samples, and σ is the standard deviation of all samples; Data normalization: Transform all data to the range [0,1] or [-1,1]. The normalization method used is MinMax normalization, and the formula is as follows: (2) X is a specific sample value, and min and max are the minimum and maximum values ​​of all samples, respectively.

[0020] Step 3: Build a neural network model Utilize the collected data to build a suitable type of neural network model; Step 3, establishing the neural network model, introduces the ant colony optimization algorithm into the training process of SOM to help the network find the global optimal distribution of samples. Specifically, the ant colony algorithm guides the direction of weight modification. In each iteration, the ants search in the weight space based on pheromones and then update the weights. This not only maintains the unsupervised nature of the SOM network but also gives the network better global search capabilities during training, thus obtaining more accurate results. S301 Initialize the weight vector: For each neuron i (i=1,2,…N), assign it a weight vector with the same dimension as the input vector x. Each of its elements is generated by a random number between 0 and 1; S302 neurons win the competition: For a randomly selected input vector x(t), calculate the distance to each neuron i, where the distance function is: (3) if Neuron j wins the competition against all ; S303 updates the weights of the winning neuron and its neighboring neurons; (4) Let represent the weight vector of neuron v at time t. The learning rate is t, where t is time or the current iteration number. It is a neighborhood function, representing the degree of influence of neuron i on neuron v; It is a constant in the range [0,1] used to balance the influence between the weight update of SOM and the weight update guided by ant colony algorithm; This represents the pheromone increment between neuron v and neuron i; The weight vector representing the highest pheromone concentration in history is the same as the weight vector corresponding to the best path in the ant colony algorithm. S304 updates the neighborhood width and learning rate; Steps 2 and 3 are repeated until the SOM network reaches stability. As for pheromone updates, they are determined by the ACO algorithm and are performed according to a certain pheromone evaporation rate and pheromone increment.

[0021] Step 4: Training and Optimizing the Model The model is trained using the training set data. During the training process, the weights and biases of the model are continuously adjusted to make it fit the training data as closely as possible. At the same time, the model can be optimized by increasing the number of training iterations, changing the learning rate, adding regularization terms, etc. like Figure 2 As shown, step 4: training and optimizing the model involves the following specific steps: S401 Initialize Weight Vector: For each neuron i (i=1,2,…N), assign it a weight vector with the same dimension as the input vector x. Each of its elements is generated by a random number between 0 and 1; S402 neurons win the competition: For a randomly selected input vector x(t), calculate the distance to each neuron i, where the distance function is: (5) if Neuron j wins the competition against all ; S403 updates the weights of the winning neuron and its neighboring neurons; (6) S404 updates neighborhood width and learning rate: After each iteration, the neighborhood width σ(t) and learning rate η(t) are updated according to a preset method, using the following exponential function: (7) (8) in and These are the initial learning rate and neighborhood width, and τ1 and τ2 are time constants; S405 determines the termination conditions: These conditions include the maximum number of iterations, the model convergence speed, etc. S406 Model Optimization: After the model training is completed, it needs to be optimized by adding L2 regularization; L2 regularization, also known as weight decay, is a method to control model complexity by adding a penalty to the weight size in the loss function, thereby preventing overfitting; it mainly targets weight parameters. Without regularization, the general form of the loss function of a neural network is: (9) When L2 regularization is introduced, the loss function becomes: (10) Here, λ is the regularization parameter, which controls the degree of influence of the penalty term on the loss function. It is the sum of squares of all weight parameters w; during training, when updating the weights using gradient descent, the addition of a squared penalty to the weight parameters w in the loss function causes the weights to decrease after each update, effectively preventing model overfitting. Finally, the optimal value of λ is usually determined through cross-validation. Specifically, the model is trained on the validation set with different values ​​of λ, and then the λ that best achieves the desired evaluation metrics (such as accuracy, F1 score, etc.) on the validation set is selected.

[0022] Step 5: Validate and test the model The performance of the model is validated and tested using validation and test set data, and the predictive ability of the model is evaluated. S501: Splitting the Dataset: First, we divide the collected dataset into three parts: a training set, a validation set, and a test set. The training set is used to train the neural network model, the validation set is used to verify the model's performance and adjust the model's parameters during training to prevent overfitting, and the test set is used to test the model's performance after training is complete.

[0023] S502: Training the Model: Train the neural network model using the training set data. During training, we will optimize the parameters of the neural network model to better fit the training data.

[0024] S503: Validate the model: After each round of training (or every few rounds of training), we will use validation set data to validate the model's performance, evaluate the model's predictive ability on unseen data, and adjust the model's parameters as needed to prevent overfitting.

[0025] S504: Testing the Model: After training, we will use test set data, which the model did not use during training, to evaluate the model's performance on real-world problems. Through testing, we can obtain a set of numerical values ​​regarding the model's performance, such as accuracy, recall, and F1 score.

[0026] S505: Evaluation Results: Compare the predicted results on the test set with the actual results, and calculate and record the model's evaluation metrics. These include prediction accuracy, error rate, precision, recall, F1 score, and confusion matrix, etc., to comprehensively evaluate the model's performance. If the model's performance meets the expected target, then the model is considered successfully trained and can be used for prediction tasks on real data. If the model's performance does not meet expectations, or if overfitting or underfitting is found in the learning curve on the validation set, then it is necessary to return to step 4, adjust the parameters of the neural network model, such as the number of layers, number of nodes, and learning rate, and then retrain.

[0027] Step 6: Assess the safety of the bridge structure Step 6: The specific method for assessing the safety of the bridge structure is as follows: A trained neural network model is used to conduct a safety assessment of the bridge structure on the access road to the hydropower station. The existing conditions and data of the bridge are input, and the model makes predictions to obtain the safety level of the bridge structure.

[0028] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0029] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for structural safety assessment of bridges on access roads to hydropower stations, characterized in that: The security assessment methods include: Step 1: Data Collection Collect data on past bridge engineering projects, including bridge design, structure, environmental impact factors, construction process and status, maintenance process and status, and human factors; Step 2: Data Preprocessing The collected data is preprocessed, including cleaning, standardization and normalization, to remove noise and outliers so that the data can be used in neural network models. Step 3: Build a neural network model Using the collected data, a neural network model is established. Step 3, establishing the neural network model, involves introducing the ant colony optimization algorithm into the training process of the SOM to help the network better find the globally optimal distribution of samples. Specifically, the ant colony algorithm guides the direction of weight modification. In each iteration, the ants search in the weight space based on pheromones and then update the weights. This not only maintains the unsupervised nature of the SOM network but also gives the network better global search capabilities during training, resulting in more accurate results. S301 Initialize the weight vector: For each neuron i, i=1,2,…N, assign it a weight vector with the same dimension as the input vector x. Each of its elements is generated by a random number between 0 and 1; S302 neurons win the competition: For a randomly selected input vector x(t), calculate the distance to each neuron i, where the distance function is: if Neuron j wins the competition against all ; S303 updates the weights of the winning neuron and its neighboring neurons; Let represent the weight vector of neuron v at time t. Here, t is the learning rate, and t is the time. It is a neighborhood function, representing the degree of influence of neuron i on neuron v; It is a constant in the range [0,1] used to balance the influence between the weight update of SOM and the weight update guided by ant colony algorithm; This represents the pheromone increment between neuron v and neuron i; The weight vector representing the highest pheromone concentration in history is the same as the weight vector corresponding to the best path in the ant colony algorithm. S304 updates the neighborhood width and learning rate; Repeat steps 2 and 3 until the SOM network reaches stability; as for pheromone updates, the ACO algorithm specifies that they should be performed according to a certain pheromone evaporation rate and pheromone increment. Step 4: Training and Optimizing the Model The model is trained using the training set data. During training, the model's weights and biases are continuously adjusted to make it fit the training data as closely as possible. Simultaneously, the model is optimized by increasing the number of training iterations, changing the learning rate, and adding regularization terms. Step 4: The specific steps for training and optimizing the model are as follows: S401 Initialize Weight Vector: For each neuron i, i=1,2,…N, assign it a weight vector with the same dimension as the input vector x. Each of its elements is generated by a random number between 0 and 1; S402 neurons win the competition: For a randomly selected input vector x(t), calculate the distance to each neuron i, where the distance function is: if Neuron j wins the competition against all ; S403 updates the weights of the winning neuron and its neighboring neurons; S404 updates neighborhood width and learning rate: After each iteration, the neighborhood width σ(t) and learning rate η(t) are updated according to a preset method, using the following exponential function: in and These are the initial learning rate and neighborhood width. and It is a time constant; S405 determines the termination conditions: These conditions include the maximum number of iterations and the model convergence rate; S406 Model Optimization: After the model training is completed, it needs to be optimized by adding L2 regularization; L2 regularization, also known as weight decay, is a method to control model complexity by adding a penalty to the weight size in the loss function, thereby preventing overfitting; it mainly targets weight parameters. Without regularization, the general form of the loss function of a neural network is: When L2 regularization is introduced, the loss function becomes: Where λ is the regularization parameter, which controls the degree of influence of the penalty term on the loss function; It is the sum of squares of all weight parameters w; during training, when updating the weights using gradient descent, the addition of a squared penalty to the weight parameters w in the loss function causes the weights to decrease after each update, effectively preventing model overfitting. The optimal value of λ is determined by cross-validation. Specifically, the model is trained on the validation set with different values ​​of λ, and then the λ that makes the evaluation metric on the validation set optimal is selected. Step 5: Validate and test the model The performance of the model is validated and tested using validation and test set data, and the predictive ability of the model is evaluated. Step 6: Assess the structural safety of the bridge.

2. The method for structural safety assessment of a hydropower station access road bridge according to claim 1, characterized in that: The specific data collection in step 1 is as follows: (1) Good project record: Collect data from relevant bridge engineering projects, including design documents, construction reports, inspection reports, and maintenance records; (2) Weather and environmental data: Collect historical weather records and environmental information of the bridge location, including temperature, humidity, wind speed, rainfall, geological structure, and soil type information; (3) Expert interviews and questionnaires: Interviews and questionnaires were conducted with engineers, bridge designers, maintenance personnel and other relevant professionals to collect their views and experiences on bridge structural safety; (4) On-site investigation and testing: Combine on-site inspection and field testing to obtain the actual condition of the bridge, and collect data on the bridge's structural type, materials and various testing indicators; (5) Public data and literature: Obtain relevant literature on bridge engineering through the Internet and library public channels.

3. The method for structural safety assessment of a hydropower station access road bridge according to claim 1, characterized in that: The specific steps of data preprocessing in step 2 are as follows: Data standardization: This mainly involves scaling the data proportionally to make it fall within a defined small range. The main purpose of data standardization is to remove the influence of data units and eliminate the order-of-magnitude differences between data points. Z-Score standardization is used, and the formula is as follows: Where X is a specific sample value, μ is the mean of all samples, and σ is the standard deviation of all samples; Data normalization: Transform all data to the range [0,1] or [-1,1]. The normalization method used is MinMax normalization, and the formula is as follows: X is a specific sample value, and min and max are the minimum and maximum values ​​of all samples, respectively.

4. The method for structural safety assessment of a hydropower station access road bridge according to claim 1, characterized in that: Step 6: The specific method for assessing the safety of the bridge structure is as follows: A trained neural network model is used to conduct a safety assessment of the bridge structure on the access road to the hydropower station. The existing conditions and data of the bridge are input, and the model makes predictions to obtain the safety level of the bridge structure.

Images