Deep learning-based roadbed settlement prediction method, electronic equipment and medium

Abstract

The invention relates to the technical field of subgrade settlement prediction, in particular to a subgrade settlement prediction method based on deep learning, electronic equipment and a medium, the subgrade settlement prediction method provided by the invention adopts a subgrade settlement prediction model for prediction, and the model adopts the combination of a convolutional neural network and a bidirectional long-short term memory network. Spatio-temporal features are effectively captured, and the prediction precision and stability of the model are improved; by introducing a channel attention enhancement mechanism module, dynamic weighting is performed on different feature channels, the selection ability of the model to key features is improved, and the sensitivity of the model to subtle changes in data and the capture ability of the model to global features are enhanced; the model is trained through the genetic algorithm, the efficiency is improved, and the genetic algorithm can automatically find an optimal combination in a complex hyper-parameter space, so that the model can show good generalization ability on different roadbed settlement data sets, and the prediction precision of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of roadbed settlement prediction, and in particular to a roadbed settlement prediction method, electronic equipment and medium based on deep learning. Background Art

[0002] As an important transportation infrastructure, the roadbed is affected by a variety of complex environmental factors during its long-term operation. These factors include geological conditions, climate change, groundwater level fluctuations, traffic loads, etc. Long-term external effects may cause the roadbed to settle, which in turn affects the flatness and safety of the road. Roadbed settlement not only causes driving discomfort and damages vehicles, but may also cause more serious structural problems, affecting traffic safety and efficiency. Therefore, timely and accurate prediction of roadbed settlement is crucial to ensure the safe operation of roads and extend their service life. Deep learning-based methods can effectively process complex environmental data, provide high-precision settlement predictions, and provide a scientific basis for roadbed monitoring and maintenance.

[0003] The existing roadbed settlement prediction methods mainly include numerical simulation, mathematical statistics and deep learning. Although the numerical simulation method can provide a detailed model, it is difficult to obtain physical parameters, which often leads to modeling difficulties, and thus the prediction results lack robustness; the mathematical statistics method has the advantages of simplicity and speed, but with the increase of the number of prediction periods, its accuracy will decrease, and it has high requirements for the quality of historical data, which limits the wide application. In addition, this method lacks the ability of self-learning and cannot adapt to the dynamically changing environment; although the deep learning method has the ability of self-adaptation and self-learning, it faces challenges such as model complexity and long training time, which requires further optimization in practical applications. Based on the above, the existing methods still need to be improved in terms of accuracy, efficiency and adaptability.

[0004] Based on the above, the present application provides a roadbed settlement prediction method, electronic device and medium based on deep learning to solve the technical problems existing in the prior art. Summary of the invention

[0005] The purpose of the present invention is to provide a roadbed settlement prediction method, electronic equipment and medium based on deep learning. The specific technical scheme is as follows: A roadbed settlement prediction method based on deep learning includes the following steps: Obtain real-time data of the roadbed to be predicted; The final roadbed settlement prediction model is used to process the real-time roadbed data to obtain the prediction result of roadbed settlement; The final roadbed settlement prediction model is obtained by following the steps below: Step S1, constructing a data set: obtaining historical roadbed data, sorting them in order according to the acquisition time, and preprocessing the historical roadbed data; Step S2: construct input data and corresponding label data to form sample data of roadbed settlement; Step S3, constructing a roadbed settlement prediction model, wherein the roadbed settlement prediction model includes an input layer, a convolutional neural network layer, a channel attention enhancement mechanism module, a bidirectional long short-term memory network layer, a fully connected layer and an output layer; Step S4: training the roadbed settlement prediction model using a genetic algorithm to obtain optimal hyperparameters of the roadbed settlement prediction model; Step S5: Based on the optimal hyperparameters of the roadbed settlement prediction model, the roadbed settlement prediction model is iteratively trained using the roadbed settlement sample data to obtain a final roadbed settlement prediction model.

[0006] Furthermore, in step S1, the step of preprocessing the roadbed historical data includes: Step S1.1, digitizing the roadbed historical data; Step S1.2, processing missing values ​​and outliers; Step S1.3, normalize the data processed in step S1.2.

[0007] Furthermore, in step S2, the pre-processed roadbed historical data is intercepted and sorted to obtain input data and corresponding label data, specifically: extract the set window length The input data at least includes The settlement monitoring data at each moment; the label data is The settlement monitoring data at each moment, Represents the predicted time span of roadbed settlement.

[0008] Further, the step of using the roadbed settlement prediction model in step S3 to predict roadbed settlement includes: Step S3.1, input data is passed into the roadbed settlement prediction model through the input layer, and local time series features are extracted through the convolutional neural network layer; Step S3.2, inputting the local temporal features extracted in step S3.1 into the channel attention enhancement mechanism module, and performing weighted processing on the feature channels through the attention mechanism; Step S3.3, input the weighted feature vector into the bidirectional long short-term memory network, deeply mine the time series information, and obtain the feature representation that covers both the time dynamics and the correlation between features; finally, after being processed by the fully connected layer, the prediction result of the roadbed settlement is obtained through the output layer.

[0009] Furthermore, in step S3.1, the specific steps of extracting local temporal features through the convolutional neural network layer are: Step S3.1.1. Input data Input convolutional neural network; convolutional layer uses multiple convolution kernels When entering data Perform convolution operation on it to generate a preliminary feature map , ; in, is the bias term, ReLU is the activation function; Step S3.1.2: The preliminary feature map Through the pooling layer, the dimension is reduced to obtain the feature map containing key information , ; in, Represents a pooling operation.

[0010] Furthermore, step S3.2 specifically includes: Step S3.2.1: Input feature map A global average pooling operation is performed along the channel dimension to extract the global feature information of each channel, thereby generating a feature vector describing the global distribution. , each element in the feature vector Indicates channel The global average eigenvalue of is calculated as: ; in, and are the height and width of the input feature map, respectively. is the channel index of the feature map, and is the pixel coordinate of the feature map, The feature map In Location and Channel The value on Step S3.2.2: Process the feature vector through the fully connected layer , generate channel attention weights , the weight is normalized by the Sigmoid activation function, and the calculation formula is as follows: ; in, is the Sigmoid activation function, and are the weight matrix and bias vector of the fully connected layer, respectively. Indicates channel The attention weight of Step S3.2.3: Channel attention weights With feature map Channel-by-channel multiplication is performed to enhance important features and suppress minor features. The calculation formula is as follows: ; in, represents element-wise multiplication, Represents the enhanced feature map.

[0011] Furthermore, in step S3.3, the weighted feature vector is input into a bidirectional long short-term memory network, one LSTM layer processes the sequence forward according to the time step, and the other LSTM layer processes the sequence reversely according to the time step. The results of the forward processing and the reverse processing are connected at the last time step. After being processed by the fully connected layer, the predicted result of the roadbed settlement is obtained through the output layer.

[0012] Furthermore, step S4 specifically includes: Step S4.1, initialize the hyperparameters of the roadbed settlement prediction model and construct an initial population: determine the key hyperparameters of the roadbed settlement prediction model; use a genetic algorithm to randomly generate a set of hyperparameter combinations, construct an initial population and initialize it; Step S4.2: After the population is initialized, the fitness of each individual in the population is evaluated; if the termination condition is met, go directly to step S4.7; if the termination condition is not met, go to the next step; Step S4.3, perform a selection operation to select individuals with better fitness from the current population; Step S4.4, perform a crossover operation to generate new individuals; after generating the new individuals, perform a mutation operation on some individuals to randomly change some hyperparameter values ​​of the individuals; Step S4.5, after crossover and mutation, the new individuals are used to replace the individuals in the old population to form a new population; Step S4.6, repeat steps S4.2 to S4.5 until the termination condition is met; Step S4.7, select the individual with the highest fitness value as the optimal solution, and the hyperparameter combination corresponding to the individual is the optimal hyperparameter of the roadbed settlement prediction model.

[0013] An electronic device comprises a memory and a processor, wherein the memory stores a computer program; the processor executes the computer program to implement the roadbed settlement prediction method based on deep learning as described above.

[0014] A medium is a readable storage medium, a computer program is stored on the readable storage medium, and a processor executes the computer program to implement the roadbed settlement prediction method based on deep learning as described above.

[0015] The application of the technical solution of the present invention has the following beneficial effects: (1) The present invention provides a roadbed settlement prediction method based on deep learning, which adopts a hybrid model combining a convolutional neural network (CNN) and a bidirectional long short-term memory network (Bi-LSTM), making full use of the advantages of CNN in local feature extraction and the long-term dependency capture ability of LSTM in processing time series data, thereby improving the modeling ability of complex spatiotemporal correlations in the roadbed settlement process. Compared with the traditional single deep learning model, the hybrid structure can more effectively capture spatiotemporal features, improve the prediction accuracy and stability of the model, and perform better in application scenarios with long time series and complex environments.

[0016] (2) The present invention introduces a channel-wise attention enhancement mechanism (CAEM) module to dynamically weight different feature channels, thereby improving the model's ability to select key features. The CAEM module captures the dependencies between channels, allowing the model to pay more attention to important features related to roadbed settlement prediction, thereby effectively improving the accuracy of feature extraction. Compared with models that rely solely on convolutional neural networks, the introduction of the CAEM module significantly enhances the model's sensitivity to subtle changes in the data and its ability to capture global features.

[0017] (3) In the present invention, the channel attention enhancement mechanism is combined with the bidirectional long short-term memory network (Bi-LSTM), which enables the model to more flexibly and accurately handle the key factors affecting sedimentation while capturing temporal information, significantly improving the accuracy and stability of the model prediction.

[0018] (4) The present invention uses a genetic algorithm to globally optimize hyperparameters. The genetic algorithm is used to search and select key parameters such as the number of convolution layers, convolution kernel step size, number of Bi-LSTM layers, and number of hidden layer units, thereby avoiding the limitations of manual parameter adjustment and improving efficiency. The genetic algorithm can automatically find the optimal combination in a complex hyperparameter space, so that the model can show good generalization ability on different roadbed settlement data sets. Compared with traditional fixed or manual parameter adjustment methods, the model optimized by the genetic algorithm not only has higher prediction accuracy, but also shows better robustness.

[0019] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 It is a flow chart for obtaining the roadbed settlement prediction model in the present invention; Figure 2 is a schematic diagram of a roadbed settlement prediction model in an embodiment of the present invention; Figure 3 It is a flow chart of using genetic algorithm to train the roadbed settlement prediction model; Figure 4 1 is a diagram of roadbed settlement prediction results of different models in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. However, the present invention can be implemented in many different ways as defined and covered. Example:

[0022] This embodiment provides a roadbed settlement prediction method based on deep learning, comprising the following steps: Obtain real-time data of the roadbed to be predicted; The final roadbed settlement prediction model is used to process the real-time roadbed data to obtain the prediction result of roadbed settlement; See also Figure 1 ,The acquisition of the final roadbed settlement prediction model includes the following steps: Step S1, constructing a data set: acquiring historical roadbed data, sorting them in order according to the acquisition time, and preprocessing the historical roadbed data; the historical roadbed data at least includes roadbed settlement monitoring data, and also includes relevant geological conditions and engineering parameters.

[0023] Preprocessing of historical roadbed data includes: Step S1.1, digitize the historical data of the roadbed: digitize the non-numerical data. For example, categorical variables (such as soil type) can be converted into numerical form using one-hot encoding or label encoding. For example, "sand" can be encoded as [1, 0], and "clay" can be encoded as [0, 1].

[0024] Step S1.2: To process missing values, linear interpolation can be used. For example, if the roadbed settlement value of a certain month is missing, linear interpolation can be performed through the values ​​of the previous and next months to obtain a reasonable estimate. To process outliers, the standard deviation method can be used to detect data points that deviate too much from the mean (for example, values ​​that exceed 3 times the standard deviation), and the outliers can be replaced with reasonable values ​​through the replacement method, such as the mean of the previous and next data.

[0025] Step S1.3: Normalize the data processed in step S1.2 to eliminate the differences between different data scales. The data normalization is processed using the Min-Max normalization method to convert the original data into a uniform range of [0, 1]. The conversion function is as follows: ; in, is the original data, and are the minimum and maximum values ​​in the original data, is the normalized data.

[0026] Step S2: construct input data and corresponding label data to form sample data of roadbed settlement: The pre-processed roadbed historical data is intercepted and sorted to obtain the input data and corresponding label data, specifically: extract the set window length The input data at least includes The settlement monitoring data at each moment; the label data is The settlement monitoring data at each moment, Indicates the predicted time span of roadbed settlement. Through the above method, sample data of roadbed settlement is formed to prepare data for model training.

[0027] Step S3: construct a roadbed settlement prediction model, see Figure 2 , the roadbed settlement prediction model includes an input layer, a convolutional neural network layer, a channel attention enhancement mechanism module, a bidirectional long short-term memory network layer, a fully connected layer and an output layer; The steps of predicting roadbed settlement using the roadbed settlement prediction model include: Step S3.1: Input data is passed into the roadbed settlement prediction model through the input layer, and local time series features are extracted through the convolutional neural network layer. Specifically: Step S3.1.1. Input data Input convolutional neural network; convolutional layer uses multiple convolution kernels When entering data Perform convolution operations on the input data to mine nonlinear local features and generate preliminary feature maps. , ; in, is the bias term, ReLU is the activation function; Step S3.1.2: Initial feature map Dimensionality reduction is performed through the pooling layer, using maximum pooling or average pooling to obtain a feature map containing key information , ; in, It is a pooling operation that is used to reduce the spatial dimension of the feature map, thereby retaining key information and reducing the amount of computation.

[0028] Step S3.2: Input the local temporal features extracted in step S3.1 into the channel attention enhancement mechanism module, and perform weighted processing on the feature channels through the attention mechanism, thereby improving the sensitivity of the model to important features and suppressing the interference of redundant features on the prediction results; the specific operations are as follows: Step S3.2.1: Input feature map A global average pooling operation is performed along the channel dimension to extract the global feature information of each channel, thereby generating a feature vector describing the global distribution. , each element in the feature vector Indicates channel The global average eigenvalue of is calculated as: ; in, and are the height and width of the input feature map, respectively. is the channel index of the feature map, and is the pixel coordinate of the feature map, The feature map In Location and Channel The value on Step S3.2.2: Process the feature vector through the fully connected layer , generate channel attention weights , the weight is normalized by the Sigmoid activation function, and the calculation formula is as follows: ; in, is the Sigmoid activation function, and are the weight matrix and bias vector of the fully connected layer, respectively. Indicates channel The attention weight of Step S3.2.3: Channel attention weights With feature map Channel-by-channel multiplication is performed to enhance important features and suppress minor features. The calculation formula is as follows: ; in, represents element-wise multiplication, Represents the enhanced feature map.

[0029] Step S3.3, input the weighted feature vector into the bidirectional long short-term memory network (Bidirectional Long Short-Term Memory, Bi-LSTM), one LSTM (Long Short-Term Memory Network) layer processes the sequence forward according to the time step, and the other LSTM layer processes the sequence reversely according to the time step; the results of the forward and reverse processing are connected at the last time step, and after being processed by the fully connected layer, the prediction result of the roadbed settlement is obtained through the output layer.

[0030] Step S4: train the roadbed settlement prediction model using a genetic algorithm to obtain the optimal hyperparameters of the roadbed settlement prediction model; Figure 3 , specifically: Step S4.1, initialize the hyperparameters of the roadbed settlement prediction model and construct the initial population: determine the key hyperparameters of the roadbed settlement prediction model, including the number of convolution layers, convolution kernel step size, LSTM time step size, number of hidden units, number of hidden layers, and number of LSTM layers; use a genetic algorithm to randomly generate a set of hyperparameter combinations to form an initial population and initialize them; each hyperparameter combination is an "individual" in the population, and each individual is composed of multiple different hyperparameters. The value range of each hyperparameter is pre-set during initialization. For example, the number of convolution layers can be selected between 1 and 5, and the number of hidden units is randomly selected between 5 and 50.

[0031] Step S4.2: After the population is initialized, the fitness of each individual in the population is evaluated; if the termination condition is met, go directly to step S4.7; if the termination condition is not met, go to the next step; In this embodiment, the mean square error (MSE) is used as the fitness function to evaluate the quality of individuals in the model. The smaller the mean square error, the higher the fitness value, indicating that the individual is better, and the excellent individuals will be retained for the next round of evolution. The mean square error calculation formula is: ; in, is the mean square error, is the number of samples, For the The actual value of the samples, For the The predicted value of a sample.

[0032] Step S4.3, perform a selection operation. The purpose of the selection operation is to select individuals with better fitness from the current population so that they have a greater probability of entering the next generation. The selection operation uses a roulette wheel to select individuals. The probability of being selected and inherited to the next generation for: ; in, is the group size, For individuals The fitness of is the sum of all individual fitnesses.

[0033] Step S4.4: After the selection operation is completed, a crossover operation is performed, and the selected parent individuals generate new child individuals through crossover. The crossover operation uses linear interpolation crossover to generate a random mixing factor with a value range of 0 to 1. In two parent individuals and Interpolate between them to generate new offspring and , the interpolation formula is as follows: ; in, is a randomly generated mixing factor, and is the selected parent individual, and The generated offspring individuals.

[0034] After generating new individuals, mutation operations are performed on some individuals to randomly change some hyperparameter values ​​of the individuals. The specific steps are as follows: First, select the individuals that need to be mutated, usually by setting a mutation probability (such as 0.01 to 0.1) to randomly decide whether to mutate the individual; then, randomly select the hyperparameters to be mutated from the selected individuals to ensure that these parameters are within their reasonable range; then, according to the preset mutation strategy (such as random perturbation, gene reset or Gaussian perturbation, etc.), adjust the selected hyperparameters to ensure that the mutation amplitude is moderate to avoid excessive negative impact on individual performance; finally, update the hyperparameters of the individuals and re-incorporate them into the population to maintain the diversity of the population. By performing this mutation operation regularly, the exploration ability of the search space can be effectively improved, thereby promoting the discovery of the global optimal solution.

[0035] Step S4.5: After crossover and mutation, new individuals are used to replace individuals in the old population to form a new population.

[0036] Step S4.6, repeat steps S4.2-S4.5 until the termination condition is met, such as reaching the set maximum number of iterations or the population fitness value converges.

[0037] Step S4.7, select the individual with the highest fitness value as the optimal solution, and the hyperparameter combination corresponding to the individual is the optimal hyperparameter of the roadbed settlement prediction model.

[0038] Step S5: Based on the optimal hyperparameters of the roadbed settlement prediction model, the roadbed settlement prediction model is iteratively trained using the roadbed settlement sample data to obtain a final roadbed settlement prediction model.

[0039] In order to verify the effectiveness of the method of the present invention, a comparative test was conducted using the prediction model of the present invention and several commonly used neural network prediction models.

[0040] 150 sets of monitoring data were selected from the actual project, and the monitoring data was divided into 145 sets of sample data for prediction analysis, with 5 monitoring data intervals as 1 step length. The 6th monitoring data interval was used as the output value. The sample data was divided into training set and test set according to the ratio of 8:2. Four models, BP, SVM, LSTM, and Bi-LSTM, were used to conduct horizontal comparative analysis with the method of the present invention. The weights and thresholds of each model were iteratively trained using the training set to obtain the final model. The prediction and comparative analysis of each model was performed using the test set. The prediction results are compared in Figure 4 ,The comparison of prediction accuracy is shown in Table 1.

[0041] Table 1 Comparison of different neural network prediction models

[0042] In Table 1, MAE: Mean Absolute Error in English and Mean Absolute Error in Chinese. The range is [0,+∞). When the predicted value is exactly the same as the true value, it is equal to 0, that is, a perfect model. The larger the error, the larger the value.

[0043] RMSE: Root Mean Square Error in English and root mean square error in Chinese. The range is [0, +∞). When the predicted value is exactly the same as the true value, it is equal to 0, that is, a perfect model. The larger the error, the larger the value.

[0044] MAPE: Mean Absolute Percentage Error in English and Mean Absolute Percentage Error in Chinese. Its range is [0,+∞). A MAPE of 0% indicates a perfect model, and a MAPE greater than 100% indicates a poor model.

[0045] from Figure 1 It can be seen from Table 1 that, compared with the traditional neural network prediction model, the prediction model provided by the present invention has higher prediction accuracy and the prediction results are closer to the measured values.

[0046] The present invention also provides an electronic device corresponding to the above embodiment, which may be a processing device for a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of the above embodiment.

[0047] The device of this embodiment includes a memory, a processor, and a computer program stored in the memory; the processor executes the computer program in the memory to implement the steps of the method of the above embodiment.

[0048] In some implementations, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0049] In some other implementations, the processor may be a central processing unit (CPU), a digital signal processor (DSP), or other general-purpose processors of various types, which are not limited herein.

[0050] The present invention also provides a medium corresponding to the above embodiment, which is a readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method of the above embodiment are implemented.

[0051] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.

[0052] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0053] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A roadbed settlement prediction method based on deep learning, characterized in that: The following steps are involved: Obtain real-time data of the roadbed to be predicted; The final roadbed settlement prediction model is used to process the real-time roadbed data to obtain the prediction result of roadbed settlement; The final roadbed settlement prediction model is obtained by following the steps below: Step S1, constructing a data set: obtaining historical roadbed data, sorting them in order according to the acquisition time, and preprocessing the historical roadbed data; Step S2: construct input data and corresponding label data to form sample data of roadbed settlement; Step S3, constructing a roadbed settlement prediction model, wherein the roadbed settlement prediction model includes an input layer, a convolutional neural network layer, a channel attention enhancement mechanism module, a bidirectional long short-term memory network layer, a fully connected layer and an output layer; Step S4: training the roadbed settlement prediction model using a genetic algorithm to obtain optimal hyperparameters of the roadbed settlement prediction model; Step S5: Based on the optimal hyperparameters of the roadbed settlement prediction model, the roadbed settlement prediction model is iteratively trained using the roadbed settlement sample data to obtain a final roadbed settlement prediction model.

2. The method for predicting roadbed settlement based on deep learning according to claim 1, characterized in that: In step S1, the step of preprocessing the roadbed historical data includes: Step S1.1, digitizing the roadbed historical data; Step S1.2, processing missing values ​​and outliers; Step S1.3, normalize the data processed in step S1.

2.

3. The method for predicting roadbed settlement based on deep learning according to claim 1, characterized in that: In step S2, the pre-processed roadbed historical data is intercepted and sorted to obtain input data and corresponding label data, specifically: extract the set window length The input data shall include at least The settlement monitoring data at each moment; the label data is The settlement monitoring data at each moment, Represents the predicted time span of roadbed settlement.

4. The method for predicting roadbed settlement based on deep learning according to claim 1, characterized in that: The steps of using the roadbed settlement prediction model in step S3 to predict roadbed settlement include: Step S3.1, input data is passed into the roadbed settlement prediction model through the input layer, and local time series features are extracted through the convolutional neural network layer; Step S3.2, inputting the local temporal features extracted in step S3.1 into the channel attention enhancement mechanism module, and performing weighted processing on the feature channels through the attention mechanism; Step S3.3, input the weighted feature vector into the bidirectional long short-term memory network, deeply mine the time series information, and obtain the feature representation that covers both the time dynamics and the correlation between features; finally, after being processed by the fully connected layer, the prediction result of the roadbed settlement is obtained through the output layer.

5. A roadbed settlement prediction method based on deep learning according to claim 4, characterized in that: In step S3.1, the specific steps of extracting local temporal features through the convolutional neural network layer are: Step S3.1.

1. Input data Input convolutional neural network; convolutional layer uses multiple convolution kernels When entering data Perform convolution operation on it to generate a preliminary feature map , ； in, is the bias term, ReLU is the activation function; Step S3.1.2: The preliminary feature map Through the pooling layer, the dimension is reduced to obtain the feature map containing key information , ； in, Represents a pooling operation.

6. A roadbed settlement prediction method based on deep learning according to claim 5, characterized in that: Step S3.2 specifically includes: Step S3.2.1: Input feature map A global average pooling operation is performed along the channel dimension to extract the global feature information of each channel, thereby generating a feature vector describing the global distribution. , each element in the feature vector Indicates channel The global average eigenvalue of is calculated as: ； in, and are the height and width of the input feature map, respectively. is the channel index of the feature map, and is the pixel coordinate of the feature map, The feature map In Location and Channel The value on Step S3.2.2: Process the feature vector through the fully connected layer , generate channel attention weights , the weight is normalized by the Sigmoid activation function, and the calculation formula is as follows: ； in, is the Sigmoid activation function, and are the weight matrix and bias vector of the fully connected layer, respectively. Indicates channel The attention weight of Step S3.2.3: Channel attention weights With feature map Channel-by-channel multiplication is performed to enhance important features and suppress minor features. The calculation formula is as follows: ； in, represents element-wise multiplication, Represents the enhanced feature map.

7. A roadbed settlement prediction method based on deep learning according to claim 4, characterized in that: In step S3.3, the weighted feature vector is input into a bidirectional long short-term memory network. One LSTM layer processes the sequence forward according to the time step, and the other LSTM layer processes the sequence reversely according to the time step. The results of the forward processing and the reverse processing are connected at the last time step. After being processed by the fully connected layer, the predicted result of the roadbed settlement is obtained through the output layer.

8. The method for predicting roadbed settlement based on deep learning according to claim 1, characterized in that: Step S4 specifically includes: Step S4.1, initialize the hyperparameters of the roadbed settlement prediction model and construct an initial population: determine the key hyperparameters of the roadbed settlement prediction model; use a genetic algorithm to randomly generate a set of hyperparameter combinations, construct an initial population and initialize it; Step S4.2: After the population is initialized, the fitness of each individual in the population is evaluated; if the termination condition is met, go directly to step S4.7; If the termination condition is not met, proceed to the next step; Step S4.3, perform a selection operation to select individuals with better fitness from the current population; Step S4.4, perform a crossover operation to generate new individuals; after generating the new individuals, perform a mutation operation on some individuals to randomly change some hyperparameter values ​​of the individuals; Step S4.5, after crossover and mutation, the new individuals are used to replace the individuals in the old population to form a new population; Step S4.6, repeat steps S4.2 to S4.5 until the termination condition is met; Step S4.7, select the individual with the highest fitness value as the optimal solution, and the hyperparameter combination corresponding to the individual is the optimal hyperparameter of the roadbed settlement prediction model.

9. An electronic device, characterized in that: The device includes a memory and a processor, wherein the memory stores a computer program; the processor executes the computer program to implement the roadbed settlement prediction method based on deep learning as described in any one of claims 1 to 8.

10. A medium, characterized in that The medium is a readable storage medium, and a computer program is stored on the readable storage medium. The processor executes the computer program to implement the roadbed settlement prediction method based on deep learning as described in any one of claims 1 to 8.

Images