Method and device for predicting nonlinear earthquake destructive power of bridge structure

Abstract

The invention discloses a bridge structure nonlinear earthquake destructive power prediction method and device, and the method comprises the following steps: collecting and preprocessing the historical earthquake response acceleration of a bridge, and obtaining an original vibration acceleration sequence; performing decomposition processing on the original vibration acceleration sequence by adopting an ensemble empirical mode decomposition method to obtain sample data; performing training optimization on the Bi LSTM model based on the sample data by adopting a black-wing plinary algorithm to obtain an optimized Bi LSTM model; and decomposing acceleration data acquired in real time through an ensemble empirical mode decomposition method, inputting the decomposed acceleration data into the optimized Bi LSTM model, and outputting predicted seismic destructive power. According to the Bi LSTM model based on black-wing optimization, the nonlinearity and randomness of bridge seismic response acceleration data can be better processed, the model after black-wing optimization has better generalization ability when processing a complex bridge structure, and the seismic destructive power of the bridge structure can be quickly and accurately predicted by using a small amount of data; the method has important application potential in bridge anti-seismic performance evaluation.

Description

Technical Field

[0001] The present invention relates to the field of civil structure engineering and disaster prevention and mitigation technology, and in particular to a method and device for predicting the nonlinear earthquake destructive force of a bridge structure. Background Art

[0002] As an important part of modern transportation networks, the safety and reliability of bridges are directly related to the safety of people's lives and property. In order to ensure the safe operation of bridges, bridge structural health monitoring has become an indispensable link. Among them, vibration response prediction, as a key technology for evaluating the structural status of bridges, is of great significance for timely discovering potential structural problems and preventing earthquake damage.

[0003] In the field of earthquake destructive force prediction of bridge structures, there are currently two main methods: methods based on physical models and methods based on data-driven methods. Methods based on physical models mainly rely on structural mechanics models such as finite element analysis. This type of method predicts the response of a bridge by simulating its vibration behavior under different loads and environmental conditions. However, due to the complexity and diversity of bridge structures, as well as the uncertainty factors in actual working conditions (such as wind loads, temperature changes, material aging, etc.), traditional physical models are difficult to accurately capture these complex nonlinear and random characteristics, resulting in large prediction errors. In addition, the establishment and solution process of the physical model requires huge computational complexity, which makes it difficult to meet the timeliness requirements of real-time monitoring.

[0004] With the rapid development of big data and artificial intelligence technologies, data-driven methods (such as artificial neural networks, support vector machines, etc.) have gradually been applied to the prediction of seismic destructive force of bridge structures. This type of method collects historical data from bridge operation and trains models for prediction. Although data-driven models have certain advantages in processing complex data, due to the multidimensional nonlinear characteristics of bridge vibration, existing models are prone to fall into local optimality, and the prediction accuracy and generalization ability are insufficient. In addition, when processing large-scale data, the training efficiency of these methods is low, and the structure and parameter optimization process of the model lack an effective control mechanism, which further limits its effectiveness in practical applications. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is to provide an efficient and accurate method and device for predicting the nonlinear seismic destructive force of bridge structures.

[0006] In order to solve the above technical problems, a technical solution adopted by the present invention is to provide a method for predicting the nonlinear seismic destructive force of a bridge structure, comprising the following steps:

[0007] The historical earthquake response acceleration of the bridge is collected and preprocessed to obtain the original vibration acceleration sequence;

[0008] Decomposing the original vibration acceleration sequence by using the ensemble empirical mode decomposition method to obtain sample data;

[0009] The Black Kite algorithm is used to train and optimize the Bi LSTM model based on sample data to obtain the optimized BiLSTM model;

[0010] The acceleration data collected in real time is decomposed by the ensemble empirical mode decomposition method and then input into the optimized Bi LSTM model to output the predicted earthquake destructive force.

[0011] Furthermore, after the step of decomposing the acceleration data collected in real time by the ensemble empirical mode decomposition method and inputting the decomposed data into the optimized Bi LSTM model to output the predicted earthquake destructive force, the method further includes:

[0012] The predicted earthquake destructive force is compared with the set alarm threshold. If the predicted earthquake destructive force is greater than or equal to the alarm threshold, an alarm is triggered, otherwise, no alarm is triggered.

[0013] Furthermore, the step of collecting the historical earthquake response acceleration of the bridge and preprocessing it to obtain the original vibration acceleration sequence includes the following sub-steps:

[0014] Using acceleration sensors to collect historical earthquake response acceleration data at key locations of the bridge, the historical earthquake response acceleration data being historical vibration acceleration data;

[0015] Filtering the historical vibration acceleration data to remove noise and remove or correct abnormal data;

[0016] The sliding window technology is used to extract the vibration acceleration sequence, which is recorded as the original vibration acceleration sequence. The original vibration acceleration sequence is presented in the form of a time series, and includes a series of vibration accelerations arranged in chronological order.

[0017] Furthermore, the step of decomposing the original vibration acceleration sequence by using the ensemble empirical mode decomposition method to obtain sample data specifically includes the following sub-steps:

[0018] The white noise sequence is superimposed on the original vibration acceleration sequence to obtain a noisy signal;

[0019] Repeat the above steps, superimpose different white noise sequences, and obtain several groups of noise-added signals;

[0020] Perform empirical mode decomposition on each group of noisy signals to obtain several groups of IMF components and residual sequences;

[0021] The average values ​​of several groups of IMF components and residual sequences obtained by decomposition are calculated respectively as sample data.

[0022] Furthermore, in the step of using the Black Kite algorithm to train and optimize the Bi LSTM model based on sample data to obtain an optimized Bi LSTM model, the following sub-steps are specifically included:

[0023] Setting the population size and the maximum number of iterations, initializing the search range of hyperparameters in the black-winged kite individuals in the population, wherein the hyperparameters include the learning rate, the number of hidden neurons, and the regularization parameter of the Bi LSTM model;

[0024] Randomly generate several black-winged kite individuals, each of which represents a set of candidate hyperparameters, and initialize the position and speed of each individual;

[0025] Calculate the fitness value of each individual at the current position, find the individual with the highest fitness value in the current population and mark it as the current global optimal solution;

[0026] Update the position and velocity of each individual according to the social behavior and hunting strategy of the black-winged kite;

[0027] Iteratively calculate the fitness value of each individual at the current position until the maximum number of iterations is reached or the fitness convergence condition is met;

[0028] At the end of the iteration, the hyperparameters of the final global optimal solution are substituted into the Bi LSTM model to obtain the optimized BiLSTM model.

[0029] Furthermore, in the step of calculating the fitness value of each individual at the current position, the following sub-steps are specifically included:

[0030] Build a Bi LSTM model based on the hyperparameters corresponding to each individual at the current position;

[0031] Dividing the sample data into a training set, a validation set and a test set according to a certain ratio;

[0032] Use the training set to train the Bi LSTM model to ensure that the model can capture the time dependency and feature relationship of the data;

[0033] After training is completed, the model performance is evaluated on the validation set and the accuracy is calculated;

[0034] Define the fitness value as accuracy.

[0035] Furthermore, in the step of updating the position and speed of each individual according to the social behavior and predation strategy of the black kite: if the hyperparameter corresponding to the updated position of the individual exceeds the search range, the value of the hyperparameter is set to the boundary value of the search range.

[0036] Furthermore, the Bi LSTM model includes an input layer, a hidden layer and a fully connected layer;

[0037] Input layer, used to receive input data;

[0038] The hidden layer includes a forward LSTM for capturing forward dependencies and a reverse LSTM for capturing backward dependencies. The outputs of the forward LSTM and the reverse LSTM are concatenated to form a new feature containing complete context information.

[0039] The fully connected layer, including one or more neurons, is used to convert the output of the hidden layer into earthquake destructive force.

[0040] In order to solve the above technical problems, another technical solution adopted by the present invention is to provide a bridge structure nonlinear earthquake destructive force prediction device, comprising:

[0041] The data acquisition module is used to collect the historical earthquake response acceleration of the bridge and perform preprocessing to obtain the original vibration acceleration sequence;

[0042] A decomposition module, used for decomposing the original vibration acceleration sequence by using a set empirical mode decomposition method to obtain sample data;

[0043] An optimization training module is used to train and optimize the Bi LSTM model based on sample data using the Black Kite algorithm to obtain an optimized Bi LSTM model;

[0044] The prediction module is used to decompose the acceleration data collected in real time by the ensemble empirical mode decomposition method and input it into the optimized BiLSTM model to output the predicted earthquake destructive force.

[0045] Furthermore, it also includes:

[0046] The alarm module is used to compare the earthquake destructive force predicted by the prediction module with the set alarm threshold. When the predicted earthquake destructive force is greater than or equal to the alarm threshold, the alarm is triggered; when the predicted earthquake destructive force is less than the alarm threshold, the alarm is not triggered.

[0047] The method and device for predicting the nonlinear earthquake destructive force of bridge structures of the present invention have at least the following beneficial effects: the present invention decomposes the vibration signal by using the ensemble empirical mode decomposition method, can extract multi-scale features, and combines the bidirectional time-dependent processing of Bi LSTM to make the prediction of the earthquake destructive force of the bridge structure more accurate; the black kite optimization algorithm can effectively improve the efficiency of the hyperparameter adjustment of the model, reduce the training time, and at the same time ensure the global optimal performance of the model, thereby improving the accuracy and robustness of the prediction; the hybrid structure of the ensemble empirical mode decomposition method and Bi LSTM can better handle the nonlinearity and randomness of bridge vibration data, and the model optimized by the black kite has better generalization ability when processing complex bridge structures; it can predict the earthquake destructive force of the bridge structure in real time, and give an alarm in time when it is detected that the earthquake destructive force reaches the alarm threshold, thereby realizing real-time monitoring and early warning, and improving the safety and stability of bridge operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 It is a flow chart of an implementation method of the nonlinear earthquake destructive force prediction method of the bridge structure of the present invention.

[0050] Figure 2 for Figure 1 Flow chart of step S1 in FIG.

[0051] Figure 3 for Figure 1 Flowchart of step S2 in FIG.

[0052] Figure 4 This is a structural diagram of the Bi LSTM model.

[0053] Figure 5 for Figure 1 Flow chart of step S3 in FIG.

[0054] Figure 6 for Figure 5 Flowchart of step S33 in FIG.

[0055] Figure 7 Schematic diagram of the implementation process of the Black Kite algorithm.

[0056] Figure 8 This is a statistical chart of the prediction accuracy of earthquake destructive force predicted by the optimized Bi LSTM model.

[0057] Fig. 9 It is a structural block diagram of one embodiment of the nonlinear earthquake destructive force prediction device for bridge structure of the present invention. DETAILED DESCRIPTION

[0058] The present invention will be further described below in conjunction with the accompanying drawings.

[0059] See also Figure 1 , is a flow chart of an implementation method of a bridge structure nonlinear earthquake destructive force prediction method of the present invention. This implementation method specifically includes the following steps:

[0060] S1. Obtain the original vibration acceleration sequence.

[0061] Specifically, the historical earthquake response acceleration of the bridge is collected and preprocessed to obtain the original vibration acceleration sequence. Figure 2 , this step S1 includes the following sub-steps:

[0062] S11. Collect historical earthquake response acceleration data.

[0063] In this embodiment, the historical earthquake response acceleration data is the historical vibration acceleration data. Specifically, according to the structural characteristics and monitoring requirements of the bridge, a number of acceleration sensors are set at key positions of the bridge, such as piers, bridge decks, cables, etc., and the acceleration sensors are used to continuously or periodically collect the historical vibration acceleration data of the key positions of the bridge.

[0064] S12. Data cleaning.

[0065] Specifically, the historical vibration acceleration data is filtered to remove noise and remove or correct abnormal data. Specifically, the historical vibration acceleration data is filtered to reduce total noise and interference and improve data accuracy; data anomalies are checked to remove abnormal values ​​caused by sensor failure, environmental interference, etc.; for data missing due to various reasons, interpolation or other methods are used to fill in the missing data to ensure data integrity.

[0066] S13, extracting vibration acceleration sequence.

[0067] Specifically, a sliding window technique is used to extract a vibration acceleration sequence, which is recorded as an original vibration acceleration sequence. The original vibration acceleration sequence is presented in the form of a time series, which includes a series of vibration accelerations arranged in chronological order. Specifically, a window of fixed length is set, and the window is slid on the original data, and the data in the window is extracted each time as a vibration acceleration sequence; the selection of the window size is determined according to the vibration characteristics and monitoring requirements of the bridge to ensure that the vibration characteristics of the bridge can be captured; according to the sliding window technique, multiple vibration acceleration sequences are extracted from the cleaned data, which are recorded as original vibration acceleration sequences.

[0068] S2. Decompose the original vibration acceleration sequence.

[0069] Specifically, the ensemble empirical mode decomposition (EEMD) method is used to decompose the original vibration acceleration sequence to obtain sample data. Ensemble empirical mode decomposition is an improved signal processing method proposed on the basis of empirical mode decomposition. Ensemble empirical mode decomposition is to add white noise to the original signal for processing. The uniform distribution of the spectrum of white noise is used to eliminate the mode mixing problem caused by the uneven interval of the signal extreme points in the empirical mode decomposition, and at the same time, the sensitivity to the signal envelope fitting error is reduced.

[0070] See also Figure 3 , this step S2 includes the following sub-steps:

[0071] S21. Superimpose white noise sequence.

[0072] Specifically, add a white noise sequence o to the original vibration acceleration sequence s(t) j (t), and get the noise signal x j (t), the noise signal x j The calculation formula of (t) is as follows:

[0073] x j (t) = s(t) + o j (t)(i)

[0074] Wherein, j represents the jth addition of the white noise sequence.

[0075] S22. Obtain several groups of noise-added signals.

[0076] Specifically, the above steps are repeated, and the superimposed white noise sequence is different each time, and several groups of noise-added signals are obtained, which are x 1 (t),x 2 (t),x 3 (t)......x j (t). In this embodiment, N groups of noise-added signals are obtained, that is, j=1, 2, 3...N.

[0077] S23. Obtain several groups of IMF components and residual sequences.

[0078] Specifically, each group of noisy signals is subjected to empirical mode decomposition to obtain several groups of IMF components and residual sequences. j (t) Decompose to obtain a set of decomposed data, which includes i IMF components g ij (t) and a residual sequence m j(t); Similarly, by decomposing N groups of noisy signals, N groups of decomposed data can be obtained.

[0079] S24. Calculate the average value.

[0080] Specifically, the average values ​​of several groups of IMF components and residual sequences obtained by decomposition are calculated as sample data. ij The calculation formula for the average value of (t) is as follows:

[0081]

[0082] Among them, g i (t) represents N groups of IMF components g ij The average value of (t).

[0083] N groups of residual sequences m j The calculation formula for the average value of (t) is as follows:

[0084]

[0085] Where m(t) represents N groups m j The average value of (t).

[0086] S3. Train and optimize the BiLSTM model.

[0087] The Black Kite algorithm is used to train and optimize the BiLSTM (bidirectional long short-term memory neural network) model based on sample data to obtain the optimized BiLSTM model. For details, please refer to Figure 4 The BiLSTM model includes an input layer, a hidden layer and a fully connected layer, wherein: the input layer is used to receive input data; the hidden layer includes a forward LSTM for capturing forward dependencies and a reverse LSTM for capturing backward dependencies, and the outputs of the forward LSTM and the reverse LSTM are concatenated to form a new feature containing complete context information; the fully connected layer includes one or more neurons for converting the output of the hidden layer into earthquake destructive force. The forward LSTM and backward LSTM in the BiLSTM model can capture bidirectional time dependencies and historical trends in the sequence, thereby enhancing the model's ability to process bridge monitoring data.

[0088] See also Figure 5 , this step S3 includes the following sub-steps:

[0089] S31. Set relevant parameters.

[0090] Set the population size and the maximum number of iterations, and initialize the search range of the hyperparameters in the black kite individuals in the population. Specifically, set the population size to N, the maximum number of iterations to T, and the hyperparameters include the learning rate η, the regularization parameter λ, and the number of hidden neurons h of the BiLSTM model. In this embodiment, the search range of the learning rate η is [0.0001, 0.1], the search range of the regularization parameter λ is [0.0001, 0.01], and the search range of the number of hidden neurons h is [32, 512].

[0091] S32. Randomly generate a number of black-winged kite individuals.

[0092] Specifically, M black kite individuals are randomly generated, each of which represents a set of candidate hyperparameters, that is, each black kite individual can be represented as x = (η, λ, h), and the position of each individual is initialized as x i , initialize the velocity of each individual to v i .

[0093] S33, calculating the fitness value.

[0094] Specifically, the fitness value of each individual at the current position is calculated, and the individual with the highest fitness value is found in the current population and marked as the current global optimal solution.

[0095] See also Figure 6 , this step S33 includes the following sub-steps:

[0096] S331. Build a Bi LSTM model.

[0097] Specifically, for the data of each individual at the current position, a Bi LSTM model is constructed according to its corresponding hyperparameters, and based on the selected hyperparameters, a deep learning framework is used to construct the Bi LSTM model.

[0098] S332. Divide sample data.

[0099] Specifically, the sample data is divided into a training set, a validation set, and a test set according to a certain ratio. In this embodiment, the sample data is divided into a training set, a validation set, and a test set according to a ratio of 6:2:2. Among them, the training set is used to train the Bi LSTM model, the validation set is used to adjust the hyperparameters of the model, and the test set is used to evaluate the final generalization ability of the model to ensure that the Bi LSTM model still has a good prediction effect on data that has not appeared before.

[0100] S333. Train the Bi LSTM model.

[0101] Use the training set to train the Bi LSTM model to ensure that the model can capture the time dependency and feature relationship of the data. Specifically, input the training set into the Bi LSTM model and train the model through forward propagation and back propagation algorithms. During the training process, the model will learn the time dependency and feature relationship of the data, and use optimization algorithms (such as Adam, SGD, etc.) to update the model weights to minimize the loss function.

[0102] S334. Calculate the accuracy.

[0103] Specifically, after training is completed, the validation set is used to evaluate the model performance, which helps to understand the performance of the model on unseen data and is used to adjust the model's hyperparameters. The validation set is input into the Bi LSTM model to calculate the model's performance indicators, including the accuracy rate, which is obtained by the following formula:

[0104]

[0105] Among them, TP represents the number of positive classes correctly predicted by the model, TN represents the number of negative classes correctly predicted by the model, FP represents the number of positive classes incorrectly predicted by the model, and FN represents the number of negative classes incorrectly predicted by the model. It can be seen that the greater the accuracy, the better the performance of the model.

[0106] S335. Define the fitness value as accuracy.

[0107] Specifically, the fitness value is defined as the accuracy rate. The maximum accuracy rate can be achieved by maximizing the fitness value. The higher the fitness value, the higher the accuracy rate, indicating that the hyperparameter combination of the individual is better. The fitness values ​​of the individuals are compared, and the individual with the highest fitness value is marked as the current global optimal solution x best .

[0108] S34. Update the position and speed of each individual.

[0109] For details, please refer to Figure 7 , the position and speed of each individual are updated according to the social behavior and predation strategy of the black kite. The speed update formula is as follows:

[0110] v i (t+1)=wv i (t)+c 1 r 1 (x best -x i (t))+c 2 r 2 (x j (t)-x i (t))+c 2 r 2(x j (t)-x i (t))(V)

[0111] Among them, w represents the inertia weight, c 1 and c 2 represents the acceleration constant, r 1 and r 2 Represents a random number, x j (t) represents the position of the other selected individuals, which can be adjacent individuals or randomly selected individuals. i (t) represents the current position of the updated individual.

[0112] The updated position of the updated individual is equal to the current position plus the updated speed. The position update formula is as follows:

[0113] x i (t+1)=x i (t)+v i (t+1)(Saturday)

[0114] It should be noted that in this embodiment, if the hyperparameters corresponding to the individual updated position exceed the preset search range, the values ​​of these hyperparameters are set to the boundary values ​​of the search range, thereby ensuring that all hyperparameters are within a reasonable range and avoiding model failure due to unreasonable hyperparameters.

[0115] S35, iteratively calculate the fitness value.

[0116] Specifically, steps S33-S34 are repeated to iteratively calculate the fitness value of each individual at the current position until the maximum number of iterations T is reached or the convergence condition of the fitness value is reached, that is, the change in the fitness value is less than the set threshold.

[0117] S36. Obtain the optimized Bi LSTM model.

[0118] Specifically, at the end of the iteration, the hyperparameters of the final global optimal solution are substituted into the Bi LSTM model to obtain the optimized Bi LSTM model, which is the model with the best performance found by the optimization algorithm under a given search space and fitness function.

[0119] S4. Predict the destructive power of earthquakes.

[0120] Specifically, the vibration acceleration data in the current environment is collected in real time, and the data is decomposed by EEMD to extract effective earthquake vibration characteristics. The processed data is then input into the optimized Bi LSTM model to output the corresponding earthquake destructive force prediction value. Figure 8, shows the comparison of the optimized Bi LSTM model in terms of earthquake destructive power prediction accuracy, and divides earthquake destructive power into four levels: no earthquake, small earthquake, moderate earthquake and large earthquake. The model can effectively identify these four levels of earthquake destructive power, and the predicted level can accurately reflect the intensity level of the earthquake.

[0121] S5. Determine whether to trigger an alarm based on the prediction result.

[0122] Specifically, the predicted earthquake destructive force is compared with the set alarm threshold. If the predicted earthquake destructive force is greater than or equal to the alarm threshold, an alarm is triggered, otherwise, no alarm is triggered.

[0123] See also Fig. 9 , is a structural block diagram of an embodiment of a bridge structure nonlinear earthquake destructive force prediction device of the present invention. The bridge structure nonlinear earthquake destructive force prediction device of this embodiment is used to implement the bridge earthquake destructive force prediction method described in the above embodiment. Specifically, the bridge earthquake destructive force prediction device of this embodiment includes a data acquisition module 100, a decomposition module 200, an optimization training module 300 and a prediction module 400. Among them:

[0124] The data acquisition module 100 is used to collect the historical vibration acceleration of the bridge and perform preprocessing to obtain an original vibration acceleration sequence.

[0125] The decomposition module 200 is used to decompose the original vibration acceleration sequence obtained by the data acquisition module 100 using the ensemble empirical mode decomposition method to obtain sample data.

[0126] The optimization training module 300 is used to train and optimize the BiLSTM model based on the sample data obtained by the decomposition module 200 using the black kite algorithm to obtain an optimized Bi LSTM model. Specifically, the Bi LSTM model includes an input layer, a hidden layer, and a fully connected layer, wherein: the input layer is used to receive input data; the hidden layer includes a forward LSTM for capturing forward dependencies and a reverse LSTM for capturing backward dependencies, and the outputs of the forward LSTM and the reverse LSTM are concatenated to form a new feature containing complete context information; the fully connected layer includes one or more neurons for converting the output of the hidden layer into earthquake destructive force.

[0127] The prediction module 400 is used to input the vibration acceleration data collected in real time into the Bi LSTM model optimized by the optimization training module after EEMD decomposition, and output the predicted earthquake destructive force.

[0128] In order to better evaluate the health status of the bridge and to sound an alarm when the vibration of the bridge is abnormal, as a preferred embodiment, an alarm module 500 is also provided for comparing the vibration acceleration data predicted by the prediction module 400 with the set alarm threshold. When the predicted vibration acceleration data is greater than or equal to the alarm threshold, an alarm is triggered; otherwise, no alarm is triggered.

[0129] The present invention decomposes vibration signals by using the ensemble empirical mode decomposition method to extract multi-scale features, and combines the bidirectional time-dependent processing of Bi LSTM to make the prediction of earthquake destructive force of bridge structures more accurate; the black kite optimization algorithm can effectively improve the efficiency of hyperparameter adjustment of the model, reduce training time, and at the same time ensure the global optimal performance of the model, thereby improving the accuracy and robustness of prediction; the hybrid structure of the ensemble empirical mode decomposition method and Bi LSTM can better handle the nonlinearity and randomness of bridge vibration data, and the model optimized by the black kite has better generalization ability when processing complex bridge structures; it can predict the earthquake destructive force of bridge structures in real time, and give an alarm in time when it is detected that the earthquake destructive force reaches the alarm threshold, thereby realizing real-time monitoring and early warning, and improving the safety and stability of bridge operation.

[0130] The above content only expresses the preferred embodiments of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A method for predicting the nonlinear earthquake destructive force of a bridge structure, characterized in that: The following steps are involved: The historical earthquake response acceleration of the bridge is collected and preprocessed to obtain the original vibration acceleration sequence; Decomposing the original vibration acceleration sequence by using the ensemble empirical mode decomposition method to obtain sample data; The Black Kite algorithm is used to train and optimize the BiLSTM model based on sample data to obtain the optimized BiLSTM model; The acceleration data collected in real time is decomposed by the ensemble empirical mode decomposition method and input into the optimized BiLSTM model to output the predicted earthquake destructive force.

2. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 1, characterized in that: After the step of decomposing the acceleration data collected in real time by the ensemble empirical mode decomposition method and inputting the decomposed data into the optimized BiLSTM model to output the predicted earthquake destructive force, the method further includes: The predicted earthquake destructive force is compared with the set alarm threshold. If the predicted earthquake destructive force is greater than or equal to the alarm threshold, an alarm is triggered, otherwise, no alarm is triggered.

3. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 1, characterized in that: The step of collecting the historical earthquake response acceleration of the bridge and preprocessing it to obtain the original vibration acceleration sequence includes the following sub-steps: Using acceleration sensors to collect historical earthquake response acceleration data at key locations of the bridge, the historical earthquake response acceleration data being historical vibration acceleration data; Filtering the historical vibration acceleration data to remove noise and remove or correct abnormal data; The sliding window technology is used to extract the vibration acceleration sequence, which is recorded as the original vibration acceleration sequence. The original vibration acceleration sequence is presented in the form of a time series, and includes a series of vibration accelerations arranged in chronological order.

4. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 3, characterized in that: The step of using the ensemble empirical mode decomposition method to decompose the original vibration acceleration sequence to obtain sample data specifically includes the following sub-steps: The white noise sequence is superimposed on the original vibration acceleration sequence to obtain a noisy signal; Repeat the above steps, superimpose different white noise sequences, and obtain several groups of noise-added signals; Perform empirical mode decomposition on each group of noisy signals to obtain several groups of IMF components and residual sequences; The average values ​​of several groups of IMF components and residual sequences obtained by decomposition are calculated respectively as sample data.

5. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 1, characterized in that: In the step of using the Black Kite algorithm to train and optimize the BiLSTM model based on sample data to obtain an optimized BiLSTM model, the following sub-steps are specifically included: The population size and the maximum number of iterations are set, and the search range of hyperparameters in the black-winged kite individuals in the population is initialized, wherein the hyperparameters include the learning rate, the number of hidden neurons, and the regularization parameter of the BiLSTM model; Randomly generate several black-winged kite individuals, each of which represents a set of candidate hyperparameters, and initialize the position and speed of each individual; Calculate the fitness value of each individual at the current position, find the individual with the highest fitness value in the current population and mark it as the current global optimal solution; Update the position and velocity of each individual according to the social behavior and hunting strategy of the black-winged kite; Iteratively calculate the fitness value of each individual at the current position until the maximum number of iterations is reached or the fitness convergence condition is met; At the end of the iteration, the hyperparameters of the final global optimal solution are substituted into the BiLSTM model to obtain the optimized BiLSTM model.

6. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 5, characterized in that: In the step of calculating the fitness value of each individual at the current position, the following sub-steps are specifically included: Build a BiLSTM model based on the hyperparameters corresponding to each individual at the current position; Dividing the sample data into a training set, a validation set and a test set according to a certain ratio; Use the training set to train the BiLSTM model to ensure that the model can capture the time dependency and feature relationship of the data; After training is completed, the model performance is evaluated on the validation set and the accuracy is calculated; Define the fitness value as accuracy.

7. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 5, characterized in that: In the step of updating the position and speed of each individual according to the social behavior and predation strategy of the black kite: if the hyperparameter corresponding to the updated position of the individual exceeds the search range, the value of the hyperparameter is set to the boundary value of the search range.

8. The method for predicting the nonlinear earthquake destructive force of a bridge structure according to claim 1, characterized in that: The BiLSTM model includes an input layer, a hidden layer and a fully connected layer; Input layer, used to receive input data; The hidden layer includes a forward LSTM for capturing forward dependencies and a reverse LSTM for capturing backward dependencies. The outputs of the forward LSTM and the reverse LSTM are concatenated to form a new feature containing complete context information. The fully connected layer, including one or more neurons, is used to convert the output of the hidden layer into earthquake destructive force.

9. A bridge structure nonlinear earthquake destructive force prediction device, characterized in that: include: The data acquisition module is used to collect the historical earthquake response acceleration of the bridge and perform preprocessing to obtain the original vibration acceleration sequence; A decomposition module, used for decomposing the original vibration acceleration sequence by using a set empirical mode decomposition method to obtain sample data; The optimization training module is used to train and optimize the BiLSTM model based on sample data using the Black Kite algorithm to obtain an optimized BiLSTM model; The prediction module is used to decompose the acceleration data collected in real time by the ensemble empirical mode decomposition method and input it into the optimized BiLSTM model to output the predicted earthquake destructive force.

10. The bridge structure nonlinear earthquake destructive force prediction device according to claim 9, characterized in that: Also includes: The alarm module is used to compare the earthquake destructive force predicted by the prediction module with the set alarm threshold. When the predicted earthquake destructive force is greater than or equal to the alarm threshold, the alarm is triggered; when the predicted earthquake destructive force is less than the alarm threshold, the alarm is not triggered.

Images