Method for estimating effective temperature extreme value of main beam under combined driving of structure monitoring and meteorology
By improving the moss growth optimization algorithm and LSTM-Attention model, the problems of insufficient hyperparameter optimization and insufficient feature selection in the effective temperature extreme estimation model of the bridge main beam are solved, and higher temperature prediction accuracy and model stability are achieved.
Patent Information
- Application Number
- CN202411729819.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-03-07
AI Technical Summary
In the prior art, the effective temperature extreme value estimation model of the main beam of the bridge inadequate technology has problems such as insufficient hyperparameter optimization and lack of effective feature selection, resulting in low estimation accuracy.
The improved moss growth optimization algorithm (IMGO) and LSTM-Attention model were used to optimize hyperparameters by introducing environmental mutation operations, and the attention mechanism was used to automatically identify important features for temperature prediction.
The accuracy and robustness of the effective temperature prediction of the main bridge beam is improved, the adaptability and stability of the model are enhanced, and the prediction error is reduced.
Smart Images
Figure CN119577916A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of civil engineering structure health monitoring, and proposes a main beam effective temperature extreme value estimation method driven jointly by structure monitoring and meteorology. Background Art
[0002] As an important part of the transportation system, the safety and durability of bridge structures have a significant impact on social and economic development. However, bridges are exposed to the natural environment all year round and are affected by a variety of environmental factors. Among them, temperature change is one of the key factors affecting bridge performance. Temperature fluctuations can cause expansion and contraction of materials, leading to internal stress and deformation. If these temperature-induced effects are not monitored and handled in a timely manner, they may cause fatigue and cracks in the materials, and even structural failure in severe cases. Therefore, establishing a method for estimating the effective temperature extremes of bridge main beams is crucial to ensure their safety and durability.
[0003] At present, some scholars have tried to apply deep learning to bridge temperature estimation. Liu Yang et al. established a BP-LSTM network based on the advantages and disadvantages of traditional BP networks and LSTM networks, and estimated the bridge node temperature with structural characteristics, time parameters, and meteorological temperature as network inputs; Tian Gu et al. discussed the influence of temperature stress during the hydration heat of large-volume concrete on bridge structures, indicating the necessity of temperature monitoring for large-volume concrete structures, and proposed to establish a BP neural network to estimate temperature changes; Wedel et al. established a fully connected neural network model, using environmental variables around the bridge (air temperature, solar radiation, air humidity) as inputs and cross-sectional temperature as output, and accurately estimated the cross-sectional temperature of the bridge.
[0004] It can be seen from the above literature that existing studies usually use traditional neural network models, such as LSTM networks and fully connected neural networks. Although these methods can estimate the bridge temperature to a certain extent, they have the following shortcomings: On the one hand, these methods do not reduce the impact of non-important features on the performance of the neural network model, lack of effective feature selection or attention mechanism, which may cause the model to focus on irrelevant or noise features, thereby affecting the estimation accuracy; on the other hand, in the process of model training, the adjustment and optimization of hyperparameters are often ignored, resulting in the performance of the model not being fully explored. Summary of the invention
[0005] The present invention aims to solve the problems of hyperparameter optimization and lack of effective feature selection in the bridge girder effective temperature extreme value estimation model in the prior art, and proposes a bridge girder effective temperature extreme value estimation method based on improved moss growth optimization algorithm (IMGO) and LSTM-Attention.
[0006] The technical solution adopted by the present invention to solve the technical problem is: A main beam effective temperature extreme value estimation method driven by structural monitoring and meteorology, the estimation method comprising the following contents: Step 1: Data preprocessing: (1) Collect hourly historical temperature data of the bridge structure and related environmental variable data, including air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD); calculate the effective temperature of the bridge based on historical temperature data , the formula is:
[0007] In the formula, is the cross-sectional area of the bridge girder; For the The area of the sub-region; For the The temperature of each sub-region; is the number of sub-regions; (2) Normalize the data to the air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) and the corresponding effective temperature of the bridge Composed of sample data, a large amount of sample data constitutes a data set; Step 2: Model building and training: (3) Build an effective temperature prediction model. The estimation model includes an input layer, an LSTM layer, an Attention layer, and an output layer. Hourly air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) as feature inputs, and the effective temperature of the bridge as output; the input data is processed by the LSTM layer to extract time series feature information from the historical data of related environmental variables, and then processed by the Attention layer to obtain the final Attention output; finally, it is processed by the Linear regression layer to obtain the output of the effective temperature prediction model, (4) Model training: 1) Introducing environmental mutation operation into the moss growth optimization algorithm, the improved moss growth optimization algorithm was obtained. The improved moss growth optimization algorithm (IMGO) was used to optimize the hyperparameters of the effective temperature prediction model, including the learning rate, the number of hidden layer neurons, and the batch size. The loss value of the effective temperature prediction model, namely the mean square error (MSE), was used as the fitness value of the improved moss growth optimization algorithm for iterative update to obtain the optimized hyperparameters. 2) Using the data set to train the effective temperature prediction model after parameter optimization, a trained effective temperature prediction model is obtained to estimate the effective temperature of the bridge structure; Step 3: Model prediction considering errors: (5) Model prediction: Input new relevant environmental variable data into the trained effective temperature prediction model to obtain the effective temperature estimation result; (6) Assume that the error sample is , , ..., , the probability density function of the kernel density estimation is expressed as:
[0008] In the formula, represents the probability density function of the error; is the number of error samples; is the bandwidth parameter; is the kernel function; Integrating the probability density function of the kernel density estimate gives the corresponding cumulative distribution function:
[0009] In the formula, represents the cumulative distribution function of the error; (7) Using the inverse cumulative distribution function of the error, the Monte Carlo sampling method is used to extract error samples equal to the number of temperature estimation results. Assume that The specific steps are as follows: 1) Generate A random number uniformly distributed in the interval [0,1] , ,…; 2) Through the inverse cumulative distribution function of the error, the uniformly distributed random number Mapping to error value On, for each , find the corresponding error sample ,satisfy:
[0010] In the formula, is the inverse cumulative distribution function of the error; 3) Repeat the above steps to generate Error samples , , ..., these error samples are added one by one to the temperature estimate to obtain the corrected estimate:
[0011] In the formula, It is The temperature estimate, is the corresponding sampling error sample, is the corrected temperature estimation result; Step 4: Extreme value calculation: (8) Based on the corrected temperature estimation results, calculate the annual extreme temperature of the bridge girder, that is, the maximum and minimum values in a year.
[0012] Furthermore, the environmental mutation operation of the improved moss growth optimization algorithm is: Set the environmental mutation probability and the number of mutation iteration intervals. Mutation condition: Every time the number of mutation iteration intervals is reached, the individual is mutated with the environmental mutation probability. Mutation method: Randomly reinitialize some individuals:
[0013] in, and are the lower and upper bounds of the search space, respectively. Indicates the generation of a size A random number vector, each element of which is a random number in the interval [0,1); are some individuals after random reinitialization; Update the optimal solution: If the individual fitness after mutation is better than the current optimal solution, then update the optimal solution.
[0014] Furthermore, the probability of environmental mutation is 20%, and the number of mutation iteration intervals is 20 times.
[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention first collects historical bridge temperature and environmental variable data, performs preprocessing, and divides the test set into a training set; secondly, builds and trains the LSTM-Attention model, optimizes the model's hyperparameters by introducing the moss growth optimization algorithm with an environmental mutation strategy, and calculates the error between the estimated value and the actual value in the test set; then, inputs new environmental variable data for estimation, and uses Monte Carlo sampling to randomly extract error samples with the same number as the estimated value from the probability distribution of the error, and adds them one by one to the corresponding estimated value; finally, based on the estimated value of the effective temperature, calculates the annual extreme temperature of the bridge main beam. By introducing the attention mechanism, the model can automatically identify and focus on the most important features for bridge temperature prediction, thereby improving the accuracy and robustness of the prediction.
[0016] 2. The present invention adopts an improved moss growth optimization algorithm (IMGO) to optimize the model parameters and introduces environmental mutation operations, which effectively improves the accuracy of hyperparameter optimization, avoids local optimal problems, and enhances the adaptability and stability of the effective temperature prediction model.
[0017] 3. The present invention adopts the Monte Carlo sampling method to map random numbers to the target distribution through the inverse cumulative distribution function, which reduces the prediction error and improves the reliability and accuracy of the model prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 It is a flow chart of the method of the present invention.
[0019] Figure 2 It is a flowchart of the improved moss growth optimization algorithm. DETAILED DESCRIPTION
[0020] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and technical solutions.
[0021] The present invention provides a main beam effective temperature extreme value estimation method driven by structural monitoring and meteorology, comprising the following steps: Step 1: Data preprocessing: (1) Collect hourly historical temperature data of the bridge structure and related environmental variable data, including air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD); the effective temperature of the bridge is calculated based on the historical temperature data, and the formula is as follows:
[0022] In the formula, is the cross-sectional area of the bridge girder; For the The area of the sub-region; For the The temperature of each sub-region; is the number of sub-regions; (2) Normalize the collected data:
[0023] In the formula, is the original data value; and are the minimum and maximum values of the feature respectively; is the normalized data value.
[0024] The air temperature (T air), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) and the corresponding effective temperature of the bridge Composed of sample data, a large amount of sample data constitutes a data set.
[0025] Step 2: Model building and training: (3) Build an effective temperature prediction model. The estimation model includes an input layer, an LSTM layer, an Attention layer, and an output layer. Hourly air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) as feature inputs, and effective bridge temperature as output; the input data is processed by the LSTM layer to extract time series feature information from the historical data of relevant environmental variables, and then processed by the Attention layer to obtain the final Attention output; finally, it is processed by the Linear regression layer to obtain the output of the effective temperature prediction model; 1) Set the input layer to hour (e.g. 48 hours) of air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) as feature inputs, and the input layer reconstructs the data into a data format that can be read by the model.
[0026] 2) Build the LSTM layer of the network model to extract time series feature information from the historical data of environmental variables.
[0027] 21) Set the forget gate, use the sigmoid neural network layer and bitwise multiplication operation to remove irrelevant information.
[0028] 22) Set the memory gate, including the input gate, tanh neural network layer and bitwise multiplication operation, to extract and memorize valid information.
[0029] 23) Set the cell state, represented by Ct, to save the state information of the current LSTM and pass it to the LSTM at the next moment.
[0030] 24) Set the output gate. Through the combined action of the tanh function and the bitwise multiplication operation, the cell state and input signal are passed to the output gate. The output gate outputs the vector that is the integration of the current input value and the output value at the previous moment.
[0031] 3) Build the Attention layer and calculate the similarity between the Query and temperature features. The formula is as follows:
[0032] In the formula, represents the similarity score of the feature, is the Query vector, Indicates Key vectors; The dimension similarity score representing the key vector is processed by SoftMax to obtain the feature weight, and the formula is as follows:
[0033] In the formula, Indicates The weight value of each feature; the Value vectors of all features are weighted and summed according to the corresponding weights to obtain the final Attention output.
[0034] 4) Set the output layer, namely the Linear regression layer, and input the output of the Attention layer into the Linear layer to estimate the effective temperature of the bridge structure.
[0035] (4) Model training: 1) Introducing environmental mutation strategy into the moss growth optimization algorithm not only improves population diversity but also enhances the ability of global search. The improved moss growth optimization algorithm (IMGO) is used to optimize the hyperparameters of the effective temperature prediction model, including learning rate, number of hidden layer neurons and batch size; the loss value of the model, namely the mean square error (MSE), is used as the fitness value of the moss growth optimization algorithm for iterative update.
[0036] 2) The training data set is the first 80% of the acquired monitoring data, and the test data set is the remaining 20%. The model training is completed by inputting the training data set into the effective temperature prediction model LSTM-Attention after parameter optimization; the test data set output is obtained by inputting the test data set into the trained model, and the error between the predicted value and the actual value in the test data set is calculated; the determination coefficient, root mean square error, and mean absolute error are used as the basis for evaluating the model performance; Step 3: Model prediction considering errors: (5) Model prediction: Input new environmental variable data into the trained effective temperature prediction model to obtain temperature estimation results; (6) Assume that the error sample is , , ..., , the probability density function of the kernel density estimation can be expressed as:
[0037] In the formula, represents the probability density function of the error; is the number of error samples; is the bandwidth parameter; It is a kernel function. Commonly used kernel functions include Gaussian kernel, uniform kernel and triangular kernel. By integrating the probability density function of the kernel density estimate, we can get the corresponding cumulative distribution function, as follows:
[0038] In the formula, represents the cumulative distribution function of the error; (7) Using the inverse cumulative distribution function of the error, the Monte Carlo sampling method is used to extract error samples equal to the number of temperature estimation results. Assume that The specific steps are as follows: 1) Generate A random number uniformly distributed in the interval [0,1] , ,…; 2) Through the inverse cumulative distribution function of the error, the uniformly distributed random number Mapping to error value For each , find the corresponding error sample ,satisfy:
[0039] In the formula, is the inverse cumulative distribution function of the error.
[0040] 3) Repeat the above steps to generate Error samples , , .... These error samples are added one by one to the temperature estimate to obtain the corrected estimate:
[0041] In the formula, It is The temperature estimate, is the corresponding sampling error sample, is the corrected temperature estimation result.
[0042] Step 4: Extreme value calculation: (8) Based on the corrected temperature estimation results, calculate the annual extreme temperature of the bridge girder, that is, the highest and lowest temperatures in a year.
[0043] The steps to improve the moss growth optimization algorithm IMGO are as follows: (1) Initialization Initialize IMGO parameters, including the maximum number of iterations , probability and interval of environmental mutation; Initialize the population matrix , where each individual is a dimension of The specific expression of the population is as follows:
[0044] in, Indicates The individual Particles, is a random number that does not exceed the maximum dimension of the individual; represents the number of moss individuals; (2) Determination of wind direction The growth direction of moss is determined by the wind. Here, the wind direction is determined by comparing the position of the individual with the optimal individual. Suppose the optimal individual is , will Dimension value is greater than or less than The individuals are divided into two groups, and the set with more individuals is selected:
[0045] in, , ;function It means to calculate the number of moss individuals in a certain set; Further obtain multiple partition sets:
[0046] in Indicates the number of times to be divided. The value of is ⌊ ⌋, and not less than 1; the symbol ⌊⋅⌋ represents a closed operation of rounding down; Representative Random numbers in the range , and satisfies the following formula: Ø The wind always blows from the area The wind blows towards the area where the best individuals are located. The precise calculation method of the wind direction is as follows:
[0047] in, Represents the calculated wind direction, with the same dimensions as the individual. Representing a collection The total number of individuals in . The calculation method is as follows:
[0048] in, Indicates in the area Internal, individual and The set of distances between (3) Spore dispersion search Spore dispersion search is used for global exploration to simulate the propagation of moss spores in the wind. The propagation is divided into two cases: steady wind and turbulent wind: Steady Wind:
[0049] in is a constant parameter, is a random number in the range (0, 1), is the wind strength, which decreases with the number of iterations. The formula is as follows:
[0050] in, Indicates the current evaluation number, Indicates the maximum number of iterations.
[0051] Turbulent Wind:
[0052] in, is a random number in the range (0, 1), is the proportion of individuals in the selected group to the total individuals, and the formula is as follows:
[0053] The position of the new individual is updated as follows. If the random number , then it is carried out in steady wind; otherwise, it is carried out in turbulent wind:
[0054] in, Indicates a new individual; is a random number in the range (0, 1); is a constant parameter, set to 0.2; (4) Double breeding search Double reproduction search mimics sexual and vegetative reproduction of mosses for local exploitation. This search is performed when is a random number between (0, 1). , then sexual reproduction is carried out; otherwise, vegetative reproduction is carried out:
[0055] in, express The Particles; represents the current optimal individual; middle, express The jth particle in ; yes The Particles; is a random number in the range (0, 1); is a constant parameter, set to 0.5; evaluate Whether the particles in the The calculation of is as follows:
[0056]
[0057] in, , All are random numbers in the range (0, 1); (5) Cryptobiotic mechanism The cryptogenetic mechanism ensures that individuals will not lose good solutions due to local optimality by recording historical information. The individual positions and fitness values of each generation are recorded. When the maximum number of records is reached or the iteration ends, the historically optimal individual is selected for replacement to maintain the quality of the population.
[0058] (6) Environmental mutation operation Mutation condition: Every certain number of iterations (for example, 20 times), the individual is mutated with a certain probability (for example, 20%).
[0059] Mutation method: Randomly reinitialize some individuals:
[0060] in, and are the lower and upper bounds of the search space, respectively. Indicates the generation of a size A random number vector, each element of which is usually a random number in the interval [0,1).
[0061] Update the optimal solution: If the individual fitness after mutation is better than the current optimal solution, then update the optimal solution.
[0062] (7) Iteration stop condition Repeat the above steps, including wind direction determination, spore dispersion search, double reproduction search, cryptobiotic mechanism and environmental mutation operation, until the algorithm runs to the set maximum number of iterations and finally finds the global optimal solution or a better approximate solution.
[0063] Example The temperature monitoring data of a long-span cable-stayed bridge for one year and the ambient environmental variable data for 50 years (data measured per hour) were obtained to verify the effectiveness of this method. The training data set is the monitoring data of the first 9 months, and the test data set is the monitoring data of the last 3 months.
[0064] The specific implementation is as follows: (1) Calculate the effective temperature of the bridge based on the temperature monitoring data obtained, taking the air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) as feature inputs, and the effective temperature of the bridge structure at the current moment as output. The input and output data are normalized to build the effective temperature prediction model LSTM-Attention.
[0065] (2) Train the constructed effective temperature prediction model, and use IMGO to optimize the hyperparameters of LSTM-Attention, including learning rate, number of hidden layer neurons, and batch size; evaluate the model performance based on the determination coefficient, root mean square error, and mean absolute error. The closer the determination coefficient is to 1, the smaller the root mean square error and mean absolute error are, the better the model performance is. Calculate the error between the predicted value and the actual value in the test data set, and obtain the probability distribution of the error. The error follows a normal distribution with a mean of -0.0075 and a variance of 0.5618; (3) The environmental variables of the past 50 years are input into the trained effective temperature prediction model to obtain the temperature estimation results. Through Monte Carlo sampling, error samples with the same number as the temperature estimation results are randomly selected from the probability distribution of the errors, and these errors are added one by one to the corresponding temperature estimation results to obtain the final estimate of the effective temperature of the bridge. Finally, the extreme value of the effective temperature of the main beam of the long-span cable-stayed bridge can be calculated.
[0066] Through the present invention, the determination coefficient of the estimation result can reach 0.99, the root mean square error is 0.8, and the mean absolute error is 0.6, which verifies the effectiveness of the method.
[0067] Any matters not described in the present invention are applicable to the prior art.
Claims
1. A method for estimating the effective temperature extreme value of a main beam driven by structural monitoring and meteorology, characterized in that: The estimation method includes the following: Step 1: Data preprocessing: (1) Collect hourly historical temperature data of the bridge structure and related environmental variable data, including air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD); calculate the effective temperature of the bridge based on historical temperature data , the formula is: In the formula, is the cross-sectional area of the bridge girder; For the The area of the sub-region; For the The temperature of each sub-region; is the number of sub-regions; (2) Normalize the data to the air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) and the corresponding effective temperature of the bridge Composed of sample data, a large amount of sample data constitutes a data set; Step 2: Model building and training: (3) Build an effective temperature prediction model. The estimation model includes an input layer, an LSTM layer, an Attention layer, and an output layer. Hourly air temperature (T air ), solar radiation intensity (SRI), wind speed (WS) and wind direction (WD) as feature inputs, and the effective temperature of the bridge as output; the input data is processed by the LSTM layer to extract time series feature information from the historical data of related environmental variables, and then processed by the Attention layer to obtain the final Attention output; finally, it is processed by the Linear regression layer to obtain the output of the effective temperature prediction model, (4) Model training: 1) Introducing environmental mutation operation into the moss growth optimization algorithm, the improved moss growth optimization algorithm was obtained. The improved moss growth optimization algorithm (IMGO) was used to optimize the hyperparameters of the effective temperature prediction model, including the learning rate, the number of hidden layer neurons, and the batch size. The loss value of the effective temperature prediction model, namely the mean square error (MSE), was used as the fitness value of the improved moss growth optimization algorithm for iterative update to obtain the optimized hyperparameters. 2) Using the data set to train the effective temperature prediction model after parameter optimization, a trained effective temperature prediction model is obtained to estimate the effective temperature of the bridge structure; Step 3: Model prediction considering errors: (5) Model prediction: Input new relevant environmental variable data into the trained effective temperature prediction model to obtain the effective temperature estimation result; (6) Assume that the error sample is , , ..., , the probability density function of the kernel density estimation is expressed as: In the formula, represents the probability density function of the error; is the number of error samples; is the bandwidth parameter; is the kernel function; Integrating the probability density function of the kernel density estimate gives the corresponding cumulative distribution function: In the formula, represents the cumulative distribution function of the error; (7) Using the inverse cumulative distribution function of the error, the Monte Carlo sampling method is used to extract error samples equal to the number of temperature estimation results. Assume that The specific steps are as follows: 1) Generate A random number uniformly distributed in the interval [0,1] , ,…; 2) Through the inverse cumulative distribution function of the error, the uniformly distributed random number Mapping to error value On, for each , find the corresponding error sample ,satisfy: In the formula, is the inverse cumulative distribution function of the error; 3) Repeat the above steps to generate Error samples , , ..., these error samples are added one by one to the temperature estimate to obtain the corrected estimate: In the formula, It is The temperature estimate, is the corresponding sampling error sample, is the corrected temperature estimation result; Step 4: Extreme value calculation: (8) Based on the corrected temperature estimation results, calculate the annual extreme temperature of the bridge girder, that is, the maximum and minimum values in a year.
2. The estimation method according to claim 1, characterized in that The environmental mutation operation of the improved moss growth optimization algorithm is: Set the environmental mutation probability and the number of mutation iteration intervals. Mutation condition: Every time the number of mutation iteration intervals is reached, the individual is mutated with the environmental mutation probability. Mutation method: Randomly reinitialize some individuals: in, and are the lower and upper bounds of the search space, respectively. Indicates the generation of a size A random number vector, each element of which is a random number in the interval [0,1); are some individuals after random reinitialization; Update the optimal solution: If the individual fitness after mutation is better than the current optimal solution, then update the optimal solution.
3. The estimation method according to claim 1, characterized in that The probability of environmental mutation is 20%, and the number of mutation iteration intervals is 20 times.