Bridge health state prediction and early warning method

Abstract

The application provides a bridge health state prediction and early warning method, comprising the following steps: S1: data acquisition, obtaining bridge state data and external environment data; S2: data correction, correcting data exceeding the threshold; S3: establishing a bridge state prediction model, decomposing the obtained bridge state data and external environment data by using an LMD algorithm, and calculating the correlation degree of each subsequence of the bridge state data and each subsequence of the external data by using a GRA algorithm; S4: model training, using a BiLSTM model for prediction; S5: after the training is completed, inputting a test set into the BiLSTM model, comparing the final prediction result with the actual measured value, and using the comparison to evaluate the prediction performance of the model; the bridge health state prediction and early warning method provided by the application uses the time-frequency analysis LMD method to analyze and decompose the bridge state data, solves the shortcomings of the time-frequency analysis method aliasing and end effect, better maintains the integrity of the data, and solves the problems of gradient disappearance and gradient explosion when the artificial intelligence algorithm predicts the bridge state.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bridge state diagnosis, and particularly relates to a bridge health state prediction and early warning method. BACKGROUND

[0002] Bridges can be seen everywhere in China, and bridge collapse will cause serious threat and loss to the development of human society and the safety of people's life and property. In recent years, bridge collapse has occurred from time to time in China, and with the rapid increase in the number of vehicles in China, combined with the fact that overloading driving is common everywhere, this directly leads to the fact that the bearing capacity and health state of the bridges designed and constructed many years ago gradually decrease over time. Due to the complexity of the structure of the bridge, bridge early warning has always been the focus and difficulty of many studies, and there is a great need for an effective early warning method at present, so it is necessary to predict and early warn the health state of the bridge.

[0003] The core idea of local mean decomposition (LMD) is to adaptively decompose a complex non-stationary multi-component signal into several product function (PF) sums with physical meaning of instantaneous frequency, wherein each PF component is directly calculated from an envelope signal and a pure frequency modulation signal. The envelope signal is the instantaneous amplitude of the PF component, and the instantaneous frequency of the PF component can be directly calculated from the pure frequency modulation signal. The complete time-frequency distribution of the original signal is obtained by combining the instantaneous amplitude and the instantaneous frequency of all PF components. The LMD algorithm can decompose complex nonlinear bridge state data and external environmental data into several PFs with physical meaning. Grey relation analysis (GRA) is a method for calculating the similarity or difference between variables in terms of development trend, and can be used as a method for measuring the degree of correlation between factors. Bidirectional long-short term memory (BiLSTM) is a kind of time recurrent neural network, and the BiLSTM model is composed of forward and backward LSTM models. The forward LSTM can obtain past data information of the input sequence, and the backward LSTM can obtain future data information of the input sequence. The training process of the forward LSTM and the backward LSTM of the data can further improve the global integrity of feature extraction. SUMMARY

[0004] The purpose of the present application is to provide a bridge health state prediction and early warning method, which uses the time-frequency analysis LMD method to analyze and decompose the bridge state data, solves the shortcomings of time-frequency analysis method aliasing and end effect, better maintains the integrity of the data, solves the problems of gradient disappearance and gradient explosion when predicting the bridge state by using artificial intelligence algorithm, and solves the problem that external environmental influence factors cannot be effectively selected, thereby reducing the prediction accuracy.

[0005] To achieve the above technical purposes and effects, the present application is implemented by the following technical solutions:

[0006] The present application provides a bridge health state prediction and early warning method, comprising the following steps:

[0007] S1: data acquisition, using sensors to monitor the bridge state in real time, obtaining bridge state data, including bridge deflection, crack length, vehicle load on the bridge surface, etc. ; using monitoring equipment to monitor the factors affecting the bridge in real time, obtaining external environment data, including rainfall, temperature, humidity, wind size, etc. ;

[0008] S2: data correction, respectively setting threshold values to compare the obtained bridge state data and external environment data, and correcting the data exceeding the threshold value, the corrected data being the arithmetic mean of the first 5 data and the last 5 data of the data exceeding the threshold value;

[0009] S3: establishing a bridge state prediction model, using LMD algorithm to decompose the obtained bridge state data and external environment data, respectively, to obtain multiple different frequency PF components and u k of the bridge state data and external environment data; k , which constitutes a subsequence, and the actual cumulative data is equal to the sum of each subsequence;

[0010] After obtaining each subsequence of the bridge state data and external environment data, the correlation degree of each subsequence of the bridge state data and each subsequence of the external data is calculated using the GRA algorithm, and if the correlation degree is greater than a preset value, it is determined that the correlation is greater, and the subsequence of the external data is related to the subsequence of the bridge state data;

[0011] S4: model training, selecting the input variable of the bridge state data subsequence with a correlation value greater than a preset value for prediction, using a BiLSTM model for prediction, and dividing the data into a training set and a test set at a fixed ratio, the training set being responsible for the parameters of the model, and the test set being responsible for verifying and evaluating the performance of the model;

[0012] S5: model verification and evaluation, after the training is completed, inputting the test set into the BiLSTM model to obtain the final prediction result, and comparing it with the actual measurement value to evaluate the prediction performance of the model; comparing the prediction value with the pre-set threshold value to obtain the early warning result.

[0013] As a further improvement of the present application, the calculation steps of the LMD algorithm in step S3 are as follows:

[0014]

[0015]

[0016] Where m i It is two consecutive extreme values ​​n i and n i+1 The i-th average value, α i This represents the local amplitude of each half-wave oscillation, i = 1, 2, ..., M-1 = 1 (M = the number of extreme values). The mean m of all consecutive extreme values ​​is... i A straight line must be formed, and the moving average method must be used for m. i and α i After smoothing, the local mean function m is obtained. 11 (t) and envelope estimation function α 11 (t). Then, m 11 The residual signal h is obtained by separating the original signal X(t) from the original signal X(t). 11 (t);

[0017] h 11 (t)=X(t)-m 11 (t);

[0018] Use h again 11 (t) divided by α 11 (t) is adjusted as follows:

[0019]

[0020] Where s 11 Let (t) be a purely frequency-modulated signal. Repeat this process q times until an envelope function satisfying α is obtained. 1(q+1) The envelope α1(t) of a purely frequency-modulated signal s1(t) = 1 is shown in the following equation:

[0021]

[0022] Where q represents the number of iterations, and the first PF1(t) is as follows:

[0023] PF1(t)=s 1q (t)×α1(t);

[0024] Subtracting PF1(t) from the original signal X(t) yields a new signal, and repeating the above steps yields PF2(t). These steps are repeated until the last signal becomes a constant or contains no further oscillations, thus obtaining the residual signal u. k Therefore, the original signal X(t) can be decomposed into PF components and u. k The sum of (t):

[0025]

[0026] PF p (t) represents the pth product function, u k (t) represents the residual component, and X(t) represents the original data.

[0027] As a further improvement of the present application, the GRA algorithm described in step S3 is calculated as follows:

[0028] S31: Set a sub-sequence X0 of the bridge state data as the main sequence, X i Construct a sequence matrix for the sub-sequences:

[0029] X = [X0, X1, X2, X3, …, X i ], i = 0, 1, …, m; k = 0, 1, …, n;

[0030]

[0031] Wherein m is the number of external environmental factor types, and n is the number of sub-sequences in the external environmental factor types.

[0032] S32: The correlation degree of each sub-sequence and each sub-sequence of the external data is calculated as follows:

[0033]

[0034]

[0035] Wherein ξ(x0(k)′, x i (k)′) is the correlation coefficient between x i and x j , ρ is the resolution coefficient, and F(x0, x i ) is the final correlation degree.

[0036] As a further improvement of the present application, the BiLSTM model described in step S4 is as follows:

[0037] f t = σ(w f · [h t-1 , x t ] + b f );

[0038] i t = σ(w i · [h t-1 , x t ] + b i );

[0039] o t = σ(w o · [ht1 , x t ]+b o );

[0040]

[0041] h t =0 t ·tanh(C t );

[0042]

[0043]

[0044] where: f t is the forget gate state of the forward propagation calculation at time t, i t is the input gate state at time t, o t is the output gate state at time t, h t is the output data at time t, C t is the cell state at time t, is the alternative state at time t, is the LSTM state of the reverse output, is the LSTM state of the forward output, H t is the output state of the BiLSTM, w f is the adjustment forget gate weight calculated by the backward propagation, w o is the output gate weight, w i is the input gate weight, w c is the alternative state weight, b f is the forget gate bias, b i is the input gate bias, b o is the output gate bias, b c is the alternative state bias.

[0045] Advantages of the present application:

[0046] The bridge health state prediction and early warning method provided by the present application adopts the time frequency analysis LMD method to analyze and decompose the bridge state data, solves the aliasing and end effect shortcomings of the time frequency analysis method EMD or EEMD method, and better maintains the integrity of the data; solves the problems of gradient disappearance and gradient explosion when the artificial intelligence algorithm predicts the bridge state, and the problem that the external environmental influence factors cannot be effectively selected, thereby reducing the prediction accuracy; a plurality of BiLSTM models are used to predict each subsequence of the bridge state, improve the prediction accuracy, solve the problem of limited prediction accuracy of the LSTM model, and also solve the problem that the occurrence of abnormal data reduces the prediction accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 This is a flowchart of the bridge health status prediction and early warning method described in this invention;

[0048] Figure 2 This is a schematic diagram of the bridge deflection data detection results described in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the temperature data detection results described in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the strain data detection results described in an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram of the PF1 component after decomposing the bridge deflection data in an embodiment of the present invention;

[0052] Figure 6 This is a schematic diagram of the PF2 component after decomposing the bridge deflection data in an embodiment of the present invention;

[0053] Figure 7 This is a schematic diagram of the PF3 components after decomposing the bridge deflection data in an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of the PF4 components after decomposing the bridge deflection data in an embodiment of the present invention;

[0055] Figure 9 This is a schematic diagram of the PF5 component after decomposing the bridge deflection data in an embodiment of the present invention;

[0056] Figure 10 In this embodiment of the invention, u is the decomposed bridge deflection data. k Component diagram;

[0057] Figure 11 This is a schematic diagram of the prediction results of the PF1 component in the bridge deflection state in an embodiment of the present invention.

[0058] Figure 12 This is a schematic diagram of the prediction results of the PF2 component in the bridge deflection state in an embodiment of the present invention;

[0059] Figure 13 This is a schematic diagram of the prediction results of the PF3 component in the bridge deflection state in an embodiment of the present invention;

[0060] Figure 14 This is a schematic diagram of the prediction results of the PF4 component in the bridge deflection state in an embodiment of the present invention.

[0061] Figure 15 This is a schematic diagram of the prediction results of the PF5 component in the bridge deflection state in an embodiment of the present invention.

[0062] Figure 16 The prediction result schematic diagram of the u k component in the bridge deflection state in the embodiment of the present application is shown in the figure.

[0063] Figure 17 The final deflection value prediction result schematic diagram in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0064] In order to make the purpose of the present application, the technical scheme and the advantages more clear and obvious, the present application is further described in detail by specific embodiments in combination with the drawings. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0065] The embodiment of the present application provides a bridge health state prediction and early warning method, and the flow steps are as shown in the figure. Figure 1

[0066] The embodiment uses the bridge deflection data monitored by the bridge deflection sensor on the right side of a certain bridge. The data is collected every hour from October 29, 2022 to November 5, 2022, and there are a total of 168 data, as shown in the figure. The X-axis in the figure is the number of data, and the Y-axis is the data value. At the same time, the temperature data and the bridge strain data measured on the right side are used as external environment data. The data is collected every hour from October 29, 2022 to November 5, 2022, and there are 168 temperature data and strain data, as shown in the figures. Figure 2 Figure 3 Figure 4

[0067] The bridge deflection data, temperature data and strain data are decomposed into multiple components by using the LMD algorithm, as shown in the figure. Figures 5 to 10

[0068] Then, the GRA is used to calculate the correlation of each subsequence of the bridge deflection data and each subsequence of the external data (temperature data and strain data). Finally, the number of input variables selected for each of the 6 deflection subsequences is 2, 3, 3, 1, 5 and 2 respectively. Then, the BiLSTM model is used to predict the values of each subsequence. The forgetting gate weight w f , the output gate weight w o , the input gate weight w i , the alternative state weight w c , the forgetting gate bias b f , the input gate bias b i , the output gate bias b o , and the alternative state bias b c are set to the preset value, and the learning rate θ is set to the preset value.

[0069] ​​​​​The bridge deflection data and external data (temperature data and strain data) are divided into a training set and a data set in a ratio of 0.8:0.2, so that among the 168 groups of data, 134 groups are in the training set and 34 groups are in the test set. The training set data is used to train the model, and the model has forward propagation and back propagation during training; each training iteration calculates y t t t t t t f o i c f i o c t The value obtained by subtracting H t and y t is added to obtain the total error, and the total error is respectively derived from the forgetting gate weight w f , the output gate weight w o , the input gate weight w i , the alternative state weight w c , the forgetting gate bias b f , the input gate bias b i , the output gate bias b o , and the alternative state bias b c The derivative of the sum of the error terms with respect to each weight is derived from the forgetting gate weight w f , the output gate weight w o , the input gate weight w i , the alternative state weight w c , the forgetting gate bias b f , the input gate bias b i , the output gate bias b o , and the alternative state bias b c The derivative obtained is subtracted from the product of the learning rate to become the new forgetting gate weight w f , the output gate weight w o , the input gate weight w i , the alternative state weight w c , the forgetting gate bias b f , the input gate bias b i , the output gate bias b o , and the alternative state bias b c ; the loop is iterated until a set number of times or the total error is less than a set value, and the model training is completed;

[0070] ​​​​​​​​​​​​​​​​The model uses the test set data to predict, obtains a prediction result, only forward propagation is used in the prediction, and six sub-sequences of the bridge deflection state are respectively predicted, as shown in Figures 11 to 16

[0071] The predicted sub-sequences are added to obtain a final deflection value prediction, as shown in Figure 17 The final prediction result is compared with the actual measured value, and a difference between the two is used as an evaluation standard to evaluate the prediction performance of the model. The prediction value is compared with a pre-set threshold value to obtain a warning result.

[0072] Thus, several aspects of at least one embodiment of the present application are described. It can be understood that various changes, modifications and improvements can be easily made by those skilled in the art. Such changes, modifications and improvements are intended to be within the spirit and scope of the present application.​

Claims

1. A bridge health state prediction and early warning method, characterized in that, Comprise the following steps: S1: data acquisition, using sensors to monitor the bridge state in real time, obtain the bridge state data; using monitoring equipment to monitor the factors affecting the bridge in real time, obtain the external environment data; S2: data correction, respectively set threshold to compare the obtained bridge state data and external environment data, correct the data beyond the threshold, the corrected data is the arithmetic mean of the first 5 data and the last 5 data of the data beyond the threshold; S3: a bridge state prediction model is established, the LMD algorithm is used to respectively decompose the obtained bridge state data and external environment data, and multiple different frequency components of the bridge state data and the external environment data are obtained PF components and , PF components and That is, the subsequence is constituted, and the actual cumulative data is equal to the sum of each subsequence; After obtaining each subsequence of bridge state data and external environment data, the correlation degree of each subsequence of bridge state data and external data is calculated by using GRA algorithm, and the correlation degree greater than the preset value is determined as the greater correlation, then the subsequence of external data is related to the subsequence of bridge state data; S4: model training, the input variable of the bridge state data subsequence with the correlation value greater than the preset value is selected for prediction, and the BiLSTM model is used for prediction, the data is divided into training set and test set at a fixed ratio, the training set is responsible for the parameters of the model, and the test set is responsible for verifying and evaluating the performance of the model; S5: model verification and evaluation, after the training, the test set is input into the BiLSTM model, the final prediction result is compared with the actual measured value, and the prediction performance of the model is evaluated; the prediction value is compared with the threshold set in advance, and the early warning result is obtained; The LMD algorithm in step S3 is calculated according to the following formula: ； In the formula: is expressed as the pth product function, is expressed as the residual component, is expressed as the original data; The GRA algorithm in step S3 is calculated according to the following steps: S31: set one sub-sequence of bridge state data as main sequence, construct sequence matrix for sub-sequence: In the formula: m is the number of external environmental factor types, and n is the number of subsequences in the external environmental factor type; S32: the correlation degree of each subsequence and external data, the correlation degree is calculated according to the following formula: wherein: is the correlation coefficient between and is the resolution coefficient, is the final correlation degree.​ 2. The bridge health state prediction and early warning method of claim 1, wherein, The BiLSTM model in step S4 is as follows: where: is the forward pass computation is the forget gate state at time t, is the input gate state at time t, is the output gate state at time t, is the output data at time t, is the cell state at time t, is the candidate state at time t, is the LSTM state for the backward output, is the LSTM state for the forward output, is the output state for the BiLSTM, is the adjustment forget gate weight for the backward pass computation, is the output gate weight, is the input gate weight, is the candidate state weight, is the forget gate bias, is the input gate bias, is the output gate bias, is the candidate state bias.​​​​​

Images