Temperature-response intelligent mapping method for large-span stiff skeleton arch bridge

Abstract

The invention discloses a temperature-response intelligent mapping method for a large-span stiff skeleton arch bridge, and belongs to the technical field of bridge structure health monitoring. Comprising the following steps: S100, acquiring data; s200, preprocessing the data to generate a data set, and segmenting the data set into a training set and a verification set; s300, based on the training set / verification set, constructing and training / verifying a Pry-SETCN model, and using MAE, RMSE and R2 to evaluate strain data obtained by mapping the Pry-SETCN model; and S400, carrying out real-time detection and abnormal response early warning on the temperature-induced strain of the bridge through the Pry-SETCN model. The method can be widely applied to a large-span bridge structure health monitoring system, effectively identifies the temperature response through the high-precision temperature-strain mapping model, provides a reliable basis for bridge safety assessment, and reduces the operation and maintenance cost.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge structure health monitoring, and more particularly to a temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge. Background Art

[0002] Bridge structures are highly sensitive to temperature, and temperature-induced responses often obscure key information such as vehicle load effects and structural damage, leading to misjudgments of safety status. This is particularly true for long-span rigid-frame concrete arch bridges, where their large spatial scale and high degree of static indeterminacy lead to non-uniform temperature distribution, significant time lags, and secondary temperature effects, severely hindering the accurate acquisition of temperature-induced responses.

[0003] Among existing methods, finite element models rely on precise material parameters and are computationally complex, while data-driven methods (such as LSTM and CNN) lack physical interpretability. Furthermore, the non-uniform distribution of temperature fields and significant time lag effects in long-span arch bridges further complicate the modeling of temperature-induced responses.

[0004] Therefore, how to provide a temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge to solve the technical problems existing in the above-mentioned prior art.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge includes:

[0008] S100: Acquire multi-channel temperature data of a specified cross section and strain data of corresponding measuring points in a bridge health monitoring system;

[0009] S200: Preprocess the multi-channel temperature data and the strain data of the corresponding measuring points to generate a data set, and divide it into a training set and a validation set;

[0010] S300: Build and train the Phy-SETCN model based on the training set, and verify the generalization ability of the Phy-SETCN model through the validation set, using MAE, RMSE, R 2 Evaluate the strain data mapped by the Phy-SETCN model;

[0011] S400: Real-time detection of bridge temperature-induced strain and abnormal response warning using the Phy-SETCN model.

[0012] Preferably, S100 includes:

[0013] The temperature increment and strain increment are calculated based on the initial bridge monitoring data. The input is multi-channel temperature increment data, and the output is the residual between the theoretical strain and the measured strain.

[0014] Preferably, the output is the residual between the theoretical strain and the measured strain, including:

[0015] Based on the calculation formula of additional internal force of arch bridge under the action of temperature, the theoretical strain of the scaled model arch under the action of temperature is derived, and the residual between the theoretical strain and the measured strain is used as the model output data.

[0016] Preferably, S200 includes:

[0017] Construct input and output data sets through sliding windows;

[0018] Perform minimum and maximum normalization processing on input and output data to eliminate dimensional differences;

[0019] Split the dataset into training and validation sets.

[0020] Preferably, the Phy-SETCN model integrates the SENet module and the TCN module, and introduces physical information construction residual as a driver, wherein the specific operation process is:

[0021] S310: Use the permute(·) function to adjust the input data;

[0022] S320: After passing through the adaptive average pooling layer, the spatial dimensions of the feature map are compressed and the shape is adjusted to match the input of the fully connected layer;

[0023] S330: Define two fully connected layers as the excitation operation, use the Sigmoid activation function, output the importance coefficient of each channel, and adjust the shape to match the shape of the original feature map;

[0024] S340: The output of the SENet module is dimensionally adjusted through the transpose(·) function to facilitate its input into the TCN module;

[0025] S350: The TCN module performs convolution operations on the input time series and uses residual connections to propagate feature information across layers to capture nonlinear data features.

[0026] S360: Use the transpose(·) function to reshape the output of the TCN module to forward it to the linear layer;

[0027] S370: Select the output of the last time step as the final prediction result.

[0028] As can be seen from the above technical solution, compared with the prior art, the present invention provides a temperature-response intelligent mapping method for long-span rigid skeleton arch bridges, which has good generalization ability and is suitable for long-term analysis of real bridge monitoring data, and has the following beneficial effects:

[0029] (1) SENet is used to dynamically adjust the TCN channel weights, alleviating the imbalance of TCN focusing too much on early information and ignoring recent information. The impact of non-uniform temperature distribution is considered through multi-channel input, and the need for pre-processing the time lag effect of raw data is eliminated. Raw data can be directly input into the model, significantly improving the model efficiency and the ability to capture key time series features.

[0030] (2) Through the theoretical temperature-induced response calculation of the rigid skeleton arch bridge, the residuals of theoretical strain and measured strain were constructed during the neural network training process, and a physically guided temperature-strain mapping model was built driven by the residuals, which significantly improved the mapping accuracy and interpretability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0032] Figure 1 It is a structural schematic diagram of the present invention;

[0033] Figure 2 Calculation diagram of additional internal forces of the main arch caused by temperature changes;

[0034] Figure 3 The position of each test section and the distribution diagram of cross-sectional measurement points;

[0035] Figure 4(a) is the time course curve corresponding to 3-T3;

[0036] Figure 4(b) is the spectrum corresponding to 3-T3;

[0037] Figure 5(a) is the time course curve corresponding to 3-S2;

[0038] Figure 5(b) is the spectrum corresponding to 3-S2;

[0039] Figure 6 This is the correlation analysis diagram of 3-T3 and 3-S2 at section 3;

[0040] Figure 7(a) is a scatter plot of the full-time correlation of the temperatures 15-T3 and 3-S2 of the section 15 that is symmetrical about the dome at section 3;

[0041] Figure 7(b) is a scatter plot of the correlation between the temperatures 15-T3 and 3-S2 of section 15, which is symmetrical about the dome, on a certain day after downsampling;

[0042] Figure 8(a) is the temperature probability density distribution diagram;

[0043] Figure 8(b) is the strain probability density distribution diagram;

[0044] Figure 9(a) is a schematic diagram of the input temperature data for model training;

[0045] Figure 9(b) is a schematic diagram of the strain data output from model training;

[0046] Figure 10 This is a schematic diagram of the sliding window operation details;

[0047] Figure 11 Schematic diagram of the specific training process of each model;

[0048] Figure 12 Comparison diagram of the mapped strain and measured strain of the validation set;

[0049] Figure 13 Radial bar charts of evaluation indicators for each model;

[0050] Figure 14 Schematic diagram of the validation set results of Phy-SETCN. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Example 1:

[0053] The embodiment of the present invention discloses a temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge. The overall process of the method is as follows: Figure 1 The specific steps are as follows:

[0054] (1) Data Preparation: Using the measured value at the start of the period of interest as the baseline value, the strain increment and temperature increment data within that period are calculated, with the temperature increment as the input. Based on the calculation formula for the additional internal force of the arch bridge under the action of temperature, the theoretical strain under the action of temperature on the scaled model arch of this embodiment is derived, and the residual between the theoretical strain and the measured strain is used as the model output data.

[0055] (2) Dataset segmentation: Use sliding windows to appropriately establish the input-output correspondence. According to the model complexity and dataset size, the dataset is reasonably segmented to ensure effective model training, where the training set is X1 = {x1, x2, ..., x c-1 ,x c}, the validation set is X2={x c+1 ,x c+2 ,...,x c+m-1 ,x c+m}, where c and m are the lengths of the training set data and the validation set data respectively.

[0056] (3) Data normalization: To improve the speed and stability of model training and enhance the performance of the final model, this embodiment uses maximum and minimum normalization to eliminate the negative impact of data magnitude differences on subsequent model training and prediction.

[0057] (4) Model training: The training set is used for feature learning, the validation set is used to adjust the model's hyperparameters, and the trained model is used to map strains in unknown time periods to evaluate the model's generalization ability.

[0058] (5) Denormalization: To ensure that all model output values ​​are consistent with the range and dimension of the original data, the model output values ​​are denormalized.

[0059] (6) Data reconstruction: Add the denormalized output, i.e., the residual of the mapping, to the theoretical value of the corresponding time sequence to obtain the mapped strain increment. 2 Evaluate the strain data obtained from model mapping.

[0060] In this simplified embodiment, the temperature effect calculation is performed in two steps under temperature load: first, a micro-segment of the rigid skeleton concrete is cut to calculate the self-stress and axial free deformation, and then the axial deformation of the micro-segment is substituted into the basic arch structure to calculate the additional internal force. The calculation diagram of the additional internal force of the main arch caused by temperature change is shown in the figure below. Figure 2 shown.

[0061] The deformation coordination condition of steel pipe and concrete under uniform temperature change is:

[0062]

[0063] Where: ε z is the axial free strain of the micro segment; σ s , σ c1 and σ c2 are the axial self-stress of steel, concrete inside the tube and outer concrete, E s 、E c1 and E c2 are the elastic moduli of steel, concrete inside the tube and concrete outside the tube, αs , α c1 and α c2 are the linear expansion coefficients of steel, concrete inside the tube, and concrete outside the tube, ΔT s , ΔT c1 and ΔT c2 are the temperature changes of steel, concrete inside the tube and external concrete respectively.

[0064] The internal force equilibrium condition of the composite section is:

[0065] σ s A s +σ c1 A c1 +σ c2 A c2 =0 (2)

[0066] Where: A s 、A c1 and A c2 are the cross-sectional areas of steel, concrete inside the tube and concrete outside the tube, respectively.

[0067] Combining equations (1) and (2) yields:

[0068]

[0069] ε z Substitution Figure 2 The basic structure shown is used to calculate the additional internal forces of the arch. The redundant restraint force of the basic system is:

[0070]

[0071] Δl t =ε z l (8)

[0072] Where H t is the horizontal force generated at the elastic center by temperature change; δ′ 42 is the horizontal displacement of the elastic center by unit horizontal force; Δl t is the horizontal displacement of the arch axis caused by temperature change; EA and EI are the axial stiffness and bending stiffness of the composite section respectively; l is the calculated span; It is the horizontal angle at any position of the arch axis.

[0073] Therefore, the additional internal force of the composite section is:

[0074] M t =-H t y=-H t (y s -y1) (9)

[0075]

[0076] Where M t and N t is the additional bending moment and axial force of any section of the arch caused by temperature change, y s is the elastic center position, and y1 is the distance from any section to the dome.

[0077] The secondary internal forces of the steel tube, core concrete and outer concrete sections are:

[0078]

[0079] In a specific embodiment, model training specifically includes:

[0080] The temperature-strain mapping model used in this example integrates SENet and TCN, and introduces physical information construction residuals as a model driver. This combination aims to fully leverage the advantages of each model to improve mapping accuracy. In this composite model, the tasks of the different models are divided as follows:

[0081] (1) SENet explicitly models the dependencies between feature channels and recalibrates channel weights to emphasize key features and suppress useless features, thereby enhancing the representational capabilities of convolutional neural networks and solving the problem that early information in the TCN receptive field ignores recent information.

[0082] (2) The extended causal convolution and residual connection designs in TCN are used to help the model effectively capture the linear and nonlinear characteristics of the signal data sequence.

[0083] (3) The theoretical response of the structure under temperature is derived from the principles of mechanics, and interpretable deep learning with embedded physical mechanisms is achieved through residual-driven modeling between the theoretical response and the measured response.

[0084] The specific operation process of the combined model (Phy-SETCN model) is as follows:

[0085] Step 1: Use the permute( ) function to adjust the input data to match the input size of the subsequent model;

[0086] Step 2: After passing through the adaptive average pooling layer, the spatial dimension of the feature map is compressed to 1x1 and the shape is adjusted to match the input of the fully connected layer;

[0087] Step 3: Define two fully connected layers as the excitation operation, use the Sigmoid activation function, output the importance coefficient of each channel, and adjust the shape to match the shape of the original feature map;

[0088] Step 4: The transpose(·) function is used to resize the output of SENet to facilitate its input into TCN.

[0089] Step 5: TCN performs convolution operations on the input time series and uses residual connections to propagate feature information across layers to capture nonlinear data features;

[0090] Step 6: Use the transpose( ) function to reshape the output of the TCN to facilitate its forwarding to the linear layer;

[0091] Step 7: Finally, the output of the last time step is selected as the final prediction result.

[0092] In one embodiment, the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ) as evaluation indicators, these indicators have been widely used in related research. RMSE amplifies large errors caused by squaring operations, making it sensitive to large errors between measured and predicted values. In contrast, MAE treats all errors equally and assigns the same weight to each error. Therefore, MAE and RMSE can effectively evaluate overall error and extreme errors, respectively. The closer the values ​​of MAE and RMSE are to 0, the smaller the model error. 2 Measures the fit between the predicted value and the observed value. The closer the value is to 1, the better the performance of the regression model. The above three evaluation indicators are calculated using the following formula:

[0093] Where m is the number of samples, P i Represents the i-th predicted value, O i represents the i-th measured value, is the average value of the measured data.

[0094] Example 2:

[0095] This example provides a measurement scheme for acquiring temperature and strain data, and performs data processing and analysis on the measured temperature and strain. The time-frequency characteristics of temperature and strain within the time period of interest, as well as the correlation between the two, are identified, and the non-Gaussian characteristics of the data are analyzed.

[0096] This example is based on a 1:10 scale model (60 meters) of the world's largest span arch bridge. This bridge is a top-decker, rigid-frame concrete arch bridge with a catenary-shaped main arch axis. The calculated span is 600 meters, the rise is 125 meters, the rise-to-span ratio f = 1 / 4.8, and the arch axis coefficient m = 1.9. The rigid frame of this scaled model is entirely made of Q420D material. The inner concrete of the pipe is C80 fine-grained concrete with good fluidity, and the outer concrete is C50 fine-grained concrete.

[0097] Specifically, the model has 17 strain test sections, and the positions of each test section and the distribution of cross-sectional measurement points are as follows: Figure 3 As shown, all models use fiber optic sensors. Fourteen strain measurement points and three temperature measurement points are deployed on each of test sections 2 through 16. Twelve additional strain measurement points are deployed on the steel pipes in sections 1 and 17 at the arch foot. The model has a total of 262 strain measurement points and 51 temperature measurement points.

[0098] This example selects the temperature data of the T3 measuring point and the strain data of the S2 measuring point at section 3 from July 4, 2023 to July 12, 2023 for data analysis. In order to facilitate the analysis of the correlation characteristics between temperature and strain, and to conform to the data characteristics of actual bridge health monitoring (the data of the health monitoring system is generally incremental), the data at the initial moment of the time period is used as the initial value, and the measured data minus the initial value is the increment within the time period. The data mentioned later are all incremental. The time history curves of 3-T3 and 3-S2 and the corresponding spectrum diagrams are shown as follows: Figure 4(a)-Figure 4(b) as well as Figure 5(a)-Figure 5(b) shown.

[0099] observe Figure 4(a)-Figure 4(b) as well as Figure 5(a)-Figure 5(b) It can be found that the strain and temperature have opposite trends, and the strain decreases when the temperature rises. Figure 4(a)-Figure 4(b) as well as Figure 5(a)-Figure 5(b) It can be found that the spectral characteristics of temperature and strain are very similar, with obvious daily periodic changes. (The period corresponding to the frequency of 1.158E-5 is approximately equal to 86356 seconds, which is approximately equal to 86400 seconds in a day.) After analyzing the time-frequency characteristics of the data, the correlation characteristics of temperature and strain were further studied. The correlation scatter plot of temperature and strain is shown in the figure below. Figure 6 and Figure 7(a)-Figure 7(b) shown.

[0100] Figure 6 This is the correlation analysis of 3-T3 and 3-S2 at section 3, that is, the temperature-strain correlation analysis of the same section. Figure 7(a)-(b) is the correlation analysis of temperature 15-T3 and 3-S2 at section 15, which is symmetrical about the dome, that is, the temperature and strain data come from different sections with a larger span in the longitudinal direction. Figure 6 and Figure 7(a)-Figure 7(b) It can be found that temperature and strain show an obvious negative correlation overall. Figure 6The scatter points in Figure 7(a)-(b) are more concentrated and the correlation is more significant. However, the scatter points in Figures 7(a)-(b) exhibit a clear loop characteristic. To more thoroughly study this characteristic, the data was downsampled and data from a single day was selected and displayed in Figure 7(b). As can be seen in Figure 7(b), the strain decreases when the temperature increases and increases when the temperature decreases, but the same temperature corresponds to different strains, with the maximum difference being 29.34με. The above analysis shows that the temperature distribution has a non-uniform characteristic, the temperature-strain correlation is more significant on the same cross section, and there is a more obvious time lag effect between the temperature and strain on different cross sections. In this case, the correlation between temperature and strain cannot be described by a simple linear relationship.

[0101] The high-order statistical characteristics are mainly judged by skewness and kurtosis. When the skewness is not equal to 0 or the kurtosis is not equal to 3, the data can be considered non-Gaussian. This embodiment uses the excess kurtosis to evaluate the kurtosis. When the kurtosis is not equal to 0, the data is considered non-Gaussian. For the probability density function, it is more intuitive to compare it with the standard Gaussian distribution. The probability density function is as follows: Figure 8(a)-Figure 8(b) shown.

[0102] The calculated kurtosis values ​​for the temperature and strain data were -0.43 and -0.47, respectively, both less than 0, indicating that the probability density distributions for temperature and strain have light tails. This characteristic is consistent with the features shown in Figures 8(a)-(b). Compared to the standard normal distribution, the data contain fewer extreme values. Regarding skewness, the skewness of temperature is 0.065, showing a positive skewness, with a slightly right-leaning distribution. The skewness of strain is -0.081, showing a negative skewness, with a slightly left-leaning distribution, consistent with Figures 8(a)-(b). In addition, three methods were used to test the data for normality. The test results are shown in Table 1. The normality test results of all three methods indicate that the temperature and strain data do not significantly come from a normal distribution population at the 0.05 level. In summary, it can be concluded that the temperature and strain data of this scheme are non-Gaussian.

[0103] Table 1 Results of normality test

[0104]

[0105] Specifically, regarding the construction of the data set, data from July 4, 2023 to July 20, 2023 were selected for research, a total of 17 days, with a sampling time of 1 minute and a data length of 24480. 3-T3 and its temperature difference with 15-T3 were used as the input data of the model, and 3-S2 was used as the output data of the model, such as Figure 9(a)-Figure 9(b) shown.

[0106] Figure 9(a) shows that the temperature difference and 3-T3 show similar trends. Comparing the two subgraphs in Figure 9(a) and (b), we can see that the strain fluctuates primarily within the range [-140, 40], a much larger range than the temperature data. During neural network model training, features with larger values ​​have a disproportionate impact on the model, while features with smaller values ​​may be ignored. If the input data distribution varies significantly, this can lead to dramatic gradient changes, making the gradient descent algorithm unstable during training or even unable to converge. Furthermore, many operations in the training process involve numerical calculations, and a large input data range can make these operations complex and time-consuming. Therefore, before constructing a training dataset, the data should be standardized or normalized. Standardization or normalization can scale all features to the same numerical range, eliminating dimensional differences, making the model more convergent, and stabilizing the data distribution. This helps the gradient descent algorithm find the optimal solution more quickly, achieving more stable performance while improving model efficiency. This embodiment uses the minimum-maximum normalization method to scale the data to [0, 1]. This method is a linear transformation and is applicable to situations where the input data range is known (maximum and minimum values ​​are known). The specific formula is as follows:

[0107]

[0108] Where x min 、x max are the minimum and maximum values ​​of the data, and x′ is the normalized value. As can be seen from the formula, it conforms to the form of y=kx+b and is a linear transformation. Its inverse transformation is:

[0109] x=x′·(x max -x min )+x min (twenty one)

[0110] The temperature data and strain data used are typical time series data. After normalization, the sliding window method is selected to construct the data set. The sliding window method can extract subsequences (windows) of fixed length from continuous time series data. Each window contains information from a period of time in the past, which can effectively capture the time dependency of the data. Taking into account the sampling frequency of the monitoring data, this embodiment sets the window size to 1440, that is, the data of one window can cover a whole day's data, which enhances the ability to capture periodic features. In addition, through the sliding window, multiple overlapping samples can be created from the original data, thereby increasing the size of the training set. More training samples can help the model learn richer features, which is beneficial to improving the training effect of the model. Therefore, the sliding window step size is set to 1. The specific sliding window sampling process is as follows: Figure 10As shown in the figure, the temperature sliding window data sequence is used as input and the strain data with a sequence length of 1 is used as output. The sampling time of the strain corresponds to the sampling time of the 1440th temperature data in the input temperature data sequence. Subtracting the window size, 23040 pairs of input and output data are finally obtained, where the input temperature data size is (23040, 1440, 2) and the output strain data size is (23040, 1).

[0111] Specifically, to accurately evaluate the fit of the trained neural network, the dataset was divided into a training set (80%, 19,104 pairs) and a validation set (20%, 4,776 pairs). When training the neural network model parameters, only the training set data was used; when evaluating the network model, the validation set data was used to verify whether the network model was overfitting or underfitting.

[0112] In a specific embodiment, the computer hardware configuration used for Phy-SETCN model construction is as follows: NVIDIA RTX 3060 GPU, 16GB memory, and 6GB video memory. The deep learning framework PyTorch is used for neural network training and verification, version PyTorch 2.4.0, and CUDA 12.4 and cuDNN 12.5 are used for GPU acceleration, and the programming language is Python 3.10.0. During the training process, the model hyperparameters are determined by the grid search method, and the Adam optimizer with good optimization effect is used. The learning rate is 0.00015, the batch size is set to 256, the number of training steps is 100, and SmoothL1Loss is used to calculate the error between the true value and the predicted value. Dropout is 0. Specific parameter values ​​are shown in Table 2 below.

[0113] Table 2 Model hyperparameter values

[0114]

[0115] Specifically, in order to verify the practicality and superiority of the SETCN method proposed by SENet to improve TCN, the effects of establishing temperature-strain correlation models using benchmark models such as TCN, LSTM and CNN were compared. The specific training process of each model is as follows: Figure 11 shown.

[0116] Figure 11The convergence process of the loss curves of each model is shown. The trends of the training loss curves of each model are similar, all rapidly decreasing and tending to be stable in the first 25 epochs. The validation losses of SETCN and TCN rapidly decrease and tend to be stable in the first 25 epochs, with minimum values of 0.000113 and 0.000139 respectively. While the validation losses of LSTM and CNN only tend to be stable at the 50th epoch, with minimum values of 0.000171 and 0.000654 respectively. It shows that the proposed method is superior to other models in both convergence speed and stability. Further, the mapped strain of the validation set is compared with the measured strain and shown in Figure 12 as follows.

[0117] From Figure 12 it can be observed that the strains mapped by SETCN, TCN and LSTM generally coincide with the changes of the measured strain, successfully capturing the strain changes of each day. Compared with the results of TCN and LSTM, the SETCN results are closer to the measured strain at the peaks and valleys, and the mapped strain is smoother with fewer burrs. The CNN results are relatively consistent with the measured strain in the first 2000 time points, but perform poorly in subsequent mappings, with an increasing deviation from the measured strain, especially obvious at the minimum values near time points 3000 and 4500. To sum up, the proposed method has stronger feature capture ability, faster convergence speed and higher mapping accuracy. To quantify the differences between different methods, the specific evaluation index values and radial bar charts are shown in Table 3 and Figure 13 as follows.

[0118] Table 3 Comparison table of model accuracies

[0119]

[0120]

[0121] It can be found from Table 3 that under the two evaluation indexes of MAE and RMSE, the performance ranking of each model is CNN < LSTM < TCN < SETCN. The performance of CNN is significantly behind that of other models, indicating that CNN is less suitable for processing long-time series data. As for R 2 the results of each model do not differ much. By observing the specific values in Table 3, it can be found that although the time consumption of SETCN is slightly higher than that of TCN, compared with TCN, the MAE and RMSE of SETCN are reduced by 9.40% and 8.92% respectively. R 2The improvement is 0.62%. It can be considered that the proposed method has a significant improvement in both accuracy and fitting degree, and the increase in runtime caused by the more complex model structure is within an acceptable range. Although LSTM and CNN take less time, their model accuracy is lower. Compared with LSTM, SETCN's MAE and RMSE are reduced by 17.70% and 18.21% respectively, R 2 Compared with CNN, SETCN’s MAE and RMSE are reduced by 53.98% and 57.33% respectively. 2 An increase of 14.47%.

[0122] The theoretical stress value calculated by the physical formula is subtracted from the measured stress value to build a residual-driven model embedded in the physical mechanism - Phy-SETCN, and the loss curves of Phy-SETCN and SETCN are shown. It can be observed that the training and verification loss of Phy-SETCN decrease faster and smoother than that of SETCN. The minimum values ​​of training loss and verification loss of Phy-SETCN are 0.0000740 and 0.0000745, respectively, which are 5.88% and 34.66% lower than those of SETCN. The mapping strain of the verification set of the two methods is further compared with the measured strain. Figure 14 Specific indicators are shown in Table 4.

[0123] Table 4 Model performance comparison

[0124] MAE RMSE <![CDATA[R 2 ]]> Time SETCN 1.590 1.981 0.973 66.288 Phy-SETCN 1.220 1.537 0.984 67.698 Improvement -23.27％ -22.41％ 1.13％ 2.13％

[0125] Analysis shows that Phy-SETCN outperforms SETCN on the validation set, especially in the part outlined by the red dotted line. The stress mapped by Phy-SETCN is closer to the measured stress, and the curve trend is more consistent. The index values ​​in Table 4 also confirm the superiority of Phy-SETCN performance. Although R 2 It only improves by 1.13%, but its accuracy is significantly improved. Compared with SETCN, the MAE and RMSE of Phy-SETCN are reduced by 23.27% and 22.41% respectively, and the running time is only increased by 2.13%.

[0126] To reflect the model's performance when facing new and unseen data, data other than the training set and validation set is selected as the test set to evaluate the model's generalization ability. Here, the data of a week in March 2024 is used as an example. The strain obtained by mapping is consistent with the change trend of the measured strain, with obvious daily cycle characteristics. In addition, it can be found in the red dotted box that the model can well capture the stress changes caused by atypical temperature changes. The MAE of the test set Phy-SETCN is 5.426, the RMSE is 6.261, and the R2 The above analysis confirms that Phy-SETCN has good generalization ability. The trained model can be applied to new and unseen data, avoiding the loss of computing resources and time caused by repeated training.

[0127] Specifically, a temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge embeds physical mechanisms into a neural network to establish a temperature-strain mapping model based on Phy-SETCN. This method also considers cross-sectional temperature and longitudinal non-uniformity as temperature distribution and physical information, effectively improving the model's convergence stability, generalization ability, and mapping accuracy. This provides a solution. It mainly includes:

[0128] (1) The temperature distribution is significantly non-uniform. The correlation between temperature and strain within the same cross section is more significant, while there is a significant time lag between the temperature and strain between different cross sections. This indicates that the relationship between temperature and strain cannot be simply described by a linear model, and the measured temperature and strain data both exhibit significant non-Gaussian characteristics.

[0129] (2) SENet effectively depicts the dependencies between feature channels and recalibrates channel weights to emphasize key features and suppress useless features. This method significantly enhances the feature expression capability of TCN and solves the problem that the TCN receptive field focuses on early information and ignores recent information. Experimental results show that SETCN has the advantages of MAE, RMSE and R 2 The indicators are all better than the comparison models. Compared with TCN, the MAE and RMSE of SETCN are reduced by 9.40% and 8.92% respectively.

[0130] (3) By using the theoretical temperature-induced response calculation formula for a rigid skeleton arch bridge proposed in this paper, the physical mechanism was embedded in a neural network to build a residual-driven mapping model, further improving the model's mapping accuracy and interpretability. Compared with SETCN, Phy-SETCN achieved a 23.27% and 22.41% reduction in MAE and RMSE, respectively, while only increasing runtime by 2.13%.

[0131] (4) The present invention not only excels in mapping accuracy but also demonstrates good generalization capabilities. The trained model can be effectively applied to new and unseen data, avoiding the loss of computing resources and time caused by repeated training.

[0132] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0133] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge, characterized in that: include: S100: Acquire multi-channel temperature data of a specified cross section and strain data of corresponding measuring points in a bridge health monitoring system; S200: Preprocess the multi-channel temperature data and the strain data of the corresponding measuring points to generate a data set, and divide it into a training set and a validation set; S300: Build and train the Phy-SETCN model based on the training set, and verify the generalization ability of the Phy-SETCN model through the validation set, using MAE, RMSE, R 2 Evaluate the strain data mapped by the Phy-SETCN model; S400: Real-time detection of bridge temperature-induced strain and abnormal response warning using the Phy-SETCN model.

2. The temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge according to claim 1 is characterized in that: S100 includes: The temperature increment and strain increment are calculated based on the initial bridge monitoring data. The input is multi-channel temperature increment data, and the output is the residual between the theoretical strain and the measured strain.

3. The temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge according to claim 2 is characterized in that: The output is the residual between the theoretical strain and the measured strain, including: Based on the calculation formula of additional internal force of arch bridge under the action of temperature, the theoretical strain of the scaled model arch under the action of temperature is derived, and the residual between the theoretical strain and the measured strain is used as the model output data.

4. The temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge according to claim 1 is characterized in that: S200 includes: Construct input and output data sets through sliding windows; Perform minimum and maximum normalization processing on input and output data to eliminate dimensional differences; Split the dataset into training and validation sets.

5. The temperature-response intelligent mapping method for a long-span rigid skeleton arch bridge according to claim 1 is characterized in that: The Phy-SETCN model integrates the SENet module and the TCN module, and introduces physical information construction residual as a driver. The specific operation process is as follows: S310: Use the permute(·) function to adjust the input data; S320: After passing through the adaptive average pooling layer, the spatial dimensions of the feature map are compressed and the shape is adjusted to match the input of the fully connected layer; S330: Define two fully connected layers as the excitation operation, use the Sigmoid activation function, output the importance coefficient of each channel, and adjust the shape to match the shape of the original feature map; S340: The output of the SENet module is dimensionally adjusted through the transpose(·) function to facilitate its input into the TCN module; S350: The TCN module performs convolution operations on the input time series and uses residual connections to propagate feature information across layers to capture nonlinear data features. S360: Use the transpose(·) function to reshape the output of the TCN module to forward it to the linear layer; S370: Select the output of the last time step as the final prediction result.

Images