CNN-BiLSTM-Attention building structure health monitoring missing data restoration method based on AO optimization

Abstract

According to the building structure health monitoring missing data restoration method based on the CNN-BiLSTM-Attention optimized by AO, spatial and temporal characteristics of structure health monitoring data are effectively captured, and high-precision structure health monitoring missing data restoration is achieved. The specific implementation process is as follows: A, pre-collecting and preprocessing structural health monitoring complete data, and constructing a training and verification data set; b, constructing a CNN-SE framework to extract spatial features of the monitoring data; c, constructing a BiLSTM-TA framework to extract time features of the monitoring data; d, constructing a CNN-BiLSTM-Attention model, and optimizing model hyper-parameters by adopting an AO algorithm; and E, inputting data acquired by a normal sensor, and repairing missing data of a fault sensor. According to the method disclosed by the invention, high-precision repair of structural health monitoring missing data can be realized, and support is provided for structural state diagnosis and prediction.

Description

Technical Field

[0001] The present invention relates to a method for repairing missing data in structural health monitoring of buildings. By combining CNN and SE attention mechanism, and combining BiLSTM and TA attention mechanism, and using the AO optimization algorithm to optimize the model hyperparameters, the missing data in structural health monitoring is repaired with high precision, belonging to the technical field of structural health monitoring. Background Art

[0002] Structural health monitoring technology can evaluate the performance state of building structures in real time, playing a crucial role in ensuring the safe operation of civil engineering structures. The effectiveness of the structural health monitoring system depends on reliable sensor data. However, due to the harsh environment where high-rise buildings are located (such as strong winds, earthquakes, and drastic temperature changes), sensors often have problems such as battery failures and communication interruptions, resulting in missing monitoring data. These missing data will affect the accuracy of structural state assessment. Therefore, developing effective methods for repairing missing data has important engineering application value.

[0003] Currently, data repair methods are mainly based on deep learning models, which repair missing monitoring data by modeling the correlation between complete and missing data. However, these methods usually only focus on the local features of the data, ignoring the complex internal correlations of the data in the time dimension and space dimension, and not considering the importance differences between different features. Therefore, it is necessary to develop a high-precision data repair method that can effectively capture the spatio-temporal correlations between different sensor data and consider the importance of different feature channels and key time steps. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention proposes a method for repairing missing data in structural health monitoring of buildings based on AO-optimized CNN-BiLSTM-Attention to improve the repair accuracy of missing data in structural health monitoring. The specific content includes:

[0005] The method for repairing missing data in structural health monitoring of buildings based on AO-optimized CNN-BiLSTM-Attention includes the following steps:

[0006] A. Pre-collect and preprocess the complete data of structural health monitoring, and construct a training and validation data set;

[0007] B. Construct a CNN-SE framework to extract the spatial features of the monitoring data;

[0008] C. Construct a BiLSTM-TA framework to extract the time features of the monitoring data;

[0009] D. Construct a CNN-BiLSTM-Attention model, and use the AO algorithm to optimize the model hyperparameters;

[0010] E. Input the data collected by normal sensors and repair the missing data of faulty sensors.

[0011] Further, step A specifically includes:

[0012] A1. Pre-collect complete structural health monitoring data;

[0013] A2. Use the maximum-minimum normalization method to process the pre-collected signal data, scale the data to the range of [-1, 1], so as to reduce the impact on the generalization ability of the network model;

[0014] A3. Divide the normalized data to construct training and validation data sets.

[0015] Further, step B specifically includes:

[0016] B1. Set up a Convolutional Neural Networks (CNN) network with two convolutional layers. After each convolutional layer, use the ReLU activation function to obtain the convolutional feature maps of the monitoring data, and transfer these feature maps to the dot product layer;

[0017] B2. After the first convolutional operation, the Squeeze-and-Excitation (SE) attention mechanism first performs global average pooling on the features of each channel to obtain the statistical information of the channel global features, which can be expressed as:

[0018]

[0019] In the formula: H and W respectively represent the width and height of the feature; u c is the input data; z c is the output value;

[0020] B3. The SE attention mechanism inputs z c into the fully connected layer. After being processed by the ReLU activation function, it then enters another fully connected layer and is activated by the sigmoid function to obtain the weight values of each channel, and transfer the weight values of each channel to the dot product layer, which can be expressed as:

[0021] S SE = σ(W2δ(W1z)) (2)

[0022] In the formula: W1 and W2 respectively represent the weights of the two fully connected layers; S SE is the output value; δ is the ReLU function; σ is the sigmoid function;

[0023] B4. The generated channel weight S SEPerform a dot product operation channel by channel on the convolutional feature map to achieve adaptive weighting of different channels, and flatten it into a vector through a flatten layer and pass it to the subsequent BiLSTM layer.

[0024] Further, step C specifically includes:

[0025] C1. Establish a bidirectional long short-term memory network (BiLSTM) to generate hidden states for each time step and pass them to the dot product layer;

[0026] C2. After the data passes through the first BiLSTM layer, the temporal attention mechanism (TA) uses a fully connected layer and the tanh function to perform a non-linear mapping on the input data, and converts the mapping result into an attention score, which can be expressed as:

[0027]

[0028] In the formula: e t is the attention score; W is the weight matrix; v is the weight vector; h t is the hidden state at the t-th time step; b is the bias vector;

[0029] C3. Normalize the attention score through the softmax function to obtain the attention weight for each time step and pass it to the dot product layer, which can be expressed as:

[0030]

[0031] In the formula: α t is the attention weight at the t-th time step;

[0032] C4. Perform a dot product operation on the hidden state at each time step and the attention weight α t to obtain the weighted time information, and then process it through the BiLSTM layer, fully connected layer and regression layer to output the final prediction result of the model.

[0033] Further, step D specifically includes:

[0034] D1. For hyperparameters such as the initial learning rate, L2 regularization coefficient, learning rate decay factor, and the number of hidden units in the BiLSTM layer and fully connected layer, the Aquila Optimization (AO) algorithm generates a certain number of random solutions, and each solution corresponds to a set of hyperparameter combinations to be optimized;

[0035] D2. Establish a CNN-BiLSTM-Attention model for each hyperparameter combination, perform model training, and calculate the fitness value (RMSE) based on the prediction results of the validation set, which can be expressed as:

[0036]

[0037] In the formula: k is the number of data points; y i is the true value of the data; is the data repair value;

[0038] D3. The AO optimization algorithm is used to minimize the fitness value, obtain the optimal hyperparameter combination, and establish the optimal CNN-BiLSTM-Attention model.

[0039] The advantages of the present invention are:

[0040] (1) The present invention combines CNN with the SE attention mechanism. The spatial features of the monitoring data are effectively extracted by CNN, and the weights of different spatial feature channels are obtained through the SE attention mechanism, effectively extracting the internal correlation in the data spatial dimension.

[0041] (2) The present invention combines BiLSTM with the TA attention mechanism. The time features of the sensor data are effectively extracted by BiLSTM, and the weights of the key time steps are obtained through the TA attention mechanism, effectively capturing the time features of the data.

[0042] (3) The present invention uses the AO optimization algorithm to optimize the hyperparameters of the neural network, effectively improving the accuracy of the model for missing data repair. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0044] Figure 2 is the layout diagram of the strain measurement points at the connection truss of the South Taihu CBD Building in the embodiment of the present invention;

[0045] Figure 3 is the schematic diagram of the CNN-BiLSTM-Attention framework of the present invention;

[0046] Figure 4 is the missing data repair result diagram of the present invention;

[0047] Figure 5 is the comparison diagram of the missing data repair results of different methods;

[0048] Figure 6 is the comparison diagram of the missing data repair indexes of different methods; DETAILED DESCRIPTION OF THE INVENTION

[0049] The present invention will be further described in detail below with reference to the drawings and an embodiment.

[0050] Such as Figure 1, this embodiment relates to a method for repairing missing data in building structure health monitoring based on AO optimization of CNN-BiLSTM-Attention, which specifically includes the following steps:

[0051] A. Pre-collect and preprocess the complete data of structural health monitoring, and construct a training and validation data set, specifically including:

[0052] A1. A structural health monitoring system is installed in the South Taihu CBD Building. In the connecting truss part, 12 fiber Bragg grating-based strain sensors are deployed. The sampling frequency is 1 Hz, and continuous acquisition is carried out for 120 hours, obtaining a total of 432,000 pieces of data. The measuring point layout is as Figure 2 shown;

[0053] A2. Take the average value of the original data in groups of 60, obtain 7,200 pieces of processed data, and use the maximum-minimum normalization method to process the pre-collected signal data, and scale the data to the range of [-1, 1];

[0054] A3. Divide the normalized data. 5,760 pieces of data from 0 to 72 hours and 96 to 120 hours are used as the training set of the model, and 1,440 pieces of data from 72 to 96 hours are used as the validation set of the model.

[0055] B. Construct a CNN-SE framework to extract the spatial features of monitoring data, specifically including:

[0056] B1. Set up a CNN network containing two convolutional layers. After each convolutional layer, use the ReLU activation function to obtain the convolutional feature map of the monitoring data, and transfer it to the dot product layer;

[0057] B2. After the first and then through the BiLSTM layer and the fully connected layer convolutional operations, the SE attention mechanism first performs global average pooling on the features of each channel to obtain the statistical information z of the channel global features c :

[0058] B3. The SE attention mechanism inputs z c into the fully connected layer. After being processed by the ReLU activation function, it enters another fully connected layer and is activated by the sigmoid function to obtain the weight value S of each channel SE , and transfer it to the dot product layer;

[0059] B4. The generated channel weight S SE performs a dot product operation with the convolutional feature map channel by channel, realizes the adaptive weighting of different channels, and is flattened into a vector through the flatten layer and transferred to the subsequent BiLSTM layer.

[0060] C. Construct the BiLSTM-TA framework to extract the time features of monitoring data, specifically including:

[0061] C1. Establish a BiLSTM layer to generate hidden states for each time step and pass them to the dot product layer;

[0062] C2. After the data passes through the first BiLSTM layer, the TA attention mechanism uses a fully connected layer and the tanh function to perform a non-linear mapping on the input data, and converts the mapping result into an attention score e t ;

[0063] C3. Normalize the attention score through the softmax function to obtain the attention weight α for each time step t and pass it to the dot product layer;

[0064] C4. Perform a dot product operation on the hidden state of each time step and the attention weight α t to obtain the weighted time information, and then successively pass through the BiLSTM layer, the fully connected layer and the regression layer for processing to output the final prediction result of the model;

[0065] D. Construct the CNN-BiLSTM-Attention model as Figure 3 shown, and use the AO optimization algorithm to optimize the model hyperparameters, specifically including:

[0066] D1. For hyperparameters such as the initial learning rate, L2 regularization coefficient, learning rate decay factor, and the number of hidden units in the BiLSTM layer and the fully connected layer, the AO optimization algorithm generates a certain number of random solutions, and each solution corresponds to a set of hyperparameter combinations to be optimized;

[0067] D2. Establish a CNN-BiLSTM-Attention model for each hyperparameter combination and perform training, and calculate the fitness value according to the prediction performance of the validation set;

[0068] D3. Use the AO optimization algorithm to minimize the fitness value, obtain the optimal hyperparameter combination, and establish the optimal CNN-BiLSTM-Attention model.

[0069] E. Input the data collected by normal sensors to repair the missing data of faulty sensors. The repair effect of the CNN-BiLSTM-Attention model on the missing data is as Figure 4 shown, and the comparison with other methods is as Figure 5 shown, and the indicators of the repair effect are as Figure 6 shown.

[0070] The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for repairing missing data in building structure health monitoring based on AO-optimized CNN-BiLSTM-Attention includes the following steps: A. Pre-collect and preprocess the complete data of structure health monitoring, and construct a training set and a validation data set; B. Construct a CNN-SE framework to extract the spatial features of the monitoring data; C. Construct a BiLSTM-TA framework to extract the temporal features of the monitoring data; D. Construct a CNN-BiLSTM-Attention model, and use the Tianying optimization algorithm (AO) to optimize the model hyperparameters; E. Input the data collected by normal sensors to repair the missing data of faulty sensors.

2. The method for repairing missing data in building structure health monitoring based on AO-optimized CNN-BiLSTM-Attention according to claim 1, characterized in that, Step A specifically includes: A1. Pre-collect the complete data of building structure health monitoring in advance; A2. Use the maximum-minimum normalization method to process the pre-collected signal data, and scale the data to the range of [-1, 1] to reduce the impact on the generalization ability of the network model; A3. Divide the normalized data to construct a training set and a validation data set.

3. The method for repairing missing data in building structure health monitoring based on AO-optimized CNN-BiLSTM-Attention according to claim 1, wherein Step B specifically includes: B1. Set up a CNN network with two convolutional layers, and use the ReLU activation function after each convolutional layer to obtain the convolutional feature maps of the monitoring data, and transfer these feature maps to the dot product layer; B2. After the first convolutional operation, the SE attention mechanism first performs global average pooling on the features of each channel to obtain the statistical information of the channel global features, which can be expressed as: where: H and W respectively represent the width and height of the feature; u c is the input data; z c is the output value; The B3.SE attention mechanism takes z c and inputs it into the fully connected layer. After being processed by the ReLU activation function, it then enters another fully connected layer and is activated by the sigmoid function to obtain the weight values S for each channel SE . The weight values for each channel are passed to the dot product layer, which can be expressed as: S SE = σ(W2δ(W1z)) (2) Where: W1 and W2 respectively represent the weights of two fully connected layers; S SE is the output value; δ is the ReLU function; σ is the sigmoid function; B4. Generated channel weight S SE Perform a dot product operation with the convolutional feature map channel by channel to achieve adaptive weighting of different channels, and flatten it into a vector through a flatten layer and pass it to the subsequent BiLSTM layer.

4. The method for repairing missing data in building structure health monitoring based on AO-optimized CNN-BiLSTM-Attention according to claim 1, characterized in that, Step C specifically includes: C1. Establish a BiLSTM layer to generate hidden states for each time step and transfer them to the dot product layer; C2. After the data passes through the first BiLSTM layer, the TA attention mechanism uses a fully connected layer and the tanh function to perform a non-linear mapping on the input data, and converts the mapping result into an attention score, which can be expressed as: where: e t is the attention score; W is the weight matrix; v is the weight vector; h t is the hidden state at the t-th time step; b is the bias vector; C3. Normalize the attention scores through the softmax function to obtain the attention weights for each time step and transfer them to the dot product layer, which can be expressed as: where: α t is the attention weight at the t-th time step; C4. The hidden state at each time step and the attention weight α t Perform a dot product operation to obtain weighted time information, which is then processed through a BiLSTM layer, a fully connected layer, and a regression layer to output the final prediction result of the model.

5. The method for repairing missing data in building structure health monitoring based on AO-optimized CNN-BiLSTM-Attention according to claim 1, wherein, Step D specifically includes: D1. For hyperparameters such as the initial learning rate, L2 regularization coefficient, learning rate decay factor, and the number of hidden units in the BiLSTM layer and the fully connected layer, the AO optimization algorithm generates a certain number of random solutions, and each solution corresponds to a set of hyperparameter combinations to be optimized; D2. Establish a CNN-BiLSTM-Attention model for each hyperparameter combination and train it, and calculate the fitness value (RMSE) according to the prediction performance of the validation set, which can be expressed as: Where: k is the number of data points; y i is the true value of the data; is the data repair value; D3. Use the AO optimization algorithm to minimize the fitness value, obtain the optimal hyperparameter combination, and establish the optimal CNN-BiLSTM-Attention model.

Images