Kalman filtering and improved CNN-LSTM-based karst cave data intelligent classification prediction method and system

Abstract

The invention discloses a karst cave data intelligent classification prediction method and system based on Kalman filtering and improved CNN-LSTM, and belongs to the technical field of artificial intelligence, and the method comprises the steps: collecting shield tunneling parameter data, and obtaining the feature information of a karst cave in front of a shield tunneling machine; performing data cleaning on the shield tunneling parameter data and the karst cave feature information; establishing a karst cave prediction model through a model structure of a convolutional neural network, a long-short-term memory network and a full-connection network; and deploying the karst cave prediction model into a shield tunneling system, inputting shield tunneling parameter data, and carrying out karst cave classification prediction. Denoising processing is carried out on the data, the influence of random noise on model training is reduced, and therefore the accuracy of the network model is improved. Through the hybrid model, the local feature extraction capability of the convolutional neural network, the time sequence prediction capability of the long and short-term memory network and the feature fusion and mapping advantages of the full-connection network are combined, so that the high-precision prediction of the karst cave is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly relates to an intelligent classification prediction method and system for karst cave data based on Kalman filtering and improved CNN-LSTM. Background Art

[0002] In karst areas, karst caves are a common geological hazard, which may cause accidents such as shield machine nose diving, collapse, and water inrush, seriously affecting construction safety and progress. The traditional physical detection methods for karst caves are costly and difficult to monitor in real time. The parameter data generated during shield tunneling can reflect the changes in geological conditions, but the existing big data analysis technologies cannot accurately predict karst caves. Therefore, there is an urgent need for a karst cave prediction method based on shield tunneling parameter data to achieve real-time and accurate karst cave early warning.

[0003] With the rapid development of big data analysis and neural network models, the close relationship between the large amount of data generated by the shield method and the construction conditions has been fully utilized to judge the situation of karst caves ahead. The neural network model can continuously optimize its own parameters using the real-time data generated by the shield and provide reliable early warnings during the construction process.

[0004] In data regression prediction and classification prediction, the most widely used is the BP neural network. The main feature of this network is the forward propagation of data signals and the backward propagation of errors. If the desired output is not obtained at the output layer, it will turn to backward propagation, and the network weights and thresholds are adjusted according to the prediction error, so that the prediction output of the BP neural network continuously approaches the desired output. However, the adjustment of weights and thresholds in the BP neural network model uses the negative gradient descent method, which has the disadvantages of slow convergence speed and getting stuck in local optimal solutions. The genetic algorithm optimized BP neural network is an optimization method for the ordinary BP neural network. The genetic algorithm is used to optimize and analyze the initial weights and thresholds of the BP neural network; the optimal initial weights and thresholds are found through selection, crossover, and mutation operations. However, this algorithm can only improve the prediction accuracy of the original BP neural network and cannot optimize the BP neural network with large prediction errors into a BP neural network that can accurately predict.

[0005] Convolutional neural network (CNN) + long short-term memory network (LSTM) + fully connected network (FCN) can extract multi-modal features from shield tunneling parameters and geological parameters. CNN is good at extracting local features of shield tunneling parameter changes. LSTM has significant superiority in analyzing the temporal changes of tunneling parameters and identifying data anomalies caused by karst caves. FNN can map the features extracted by CNN and LSTM to the output space and perform karst cave classification. However, when the shield machine faces complex geological conditions, a large amount of data noise will be generated in the shield tunneling parameters. Therefore, the Kalman filtering algorithm can be introduced into the neural network model to eliminate data noise and improve the accuracy of network model prediction. Summary of the Invention

[0006] To solve the above technical problems, an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM is proposed, including collecting shield tunneling parameter data to obtain karst cave feature information in front of the shield machine; performing data cleaning on the shield tunneling parameter data and the karst cave feature information; establishing a karst cave prediction model through a hybrid model structure of a convolutional neural network, a long short-term memory network, and a fully connected network; deploying the karst cave prediction model into the shield tunneling system, inputting the shield tunneling parameter data, and performing karst cave classification and prediction.

[0007] As a preferred solution of the intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to the present invention, wherein: the data cleaning includes filling the missing values in the shield tunneling parameter data by the interpolation method; using the Kalman filtering algorithm to denoise the filled data; performing normalization processing on the denoised data to convert the shield tunneling parameters with different dimensions to the same numerical range; dividing the normalized data into a training set and a test set in chronological order.

[0008] As a preferred solution of the intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to the present invention, wherein: the karst cave prediction model includes an input layer for receiving the shield tunneling parameter data; a convolutional layer for extracting local features of the shield tunneling parameters using a convolutional neural network; an LSTM layer for modeling the time series data of the tunneling parameters using a long short-term memory network to extract long-term dependencies; a fully connected layer for fusing the features extracted by the convolutional neural network and the long short-term memory network using a fully connected network; and an output layer for outputting the fusion result of the fully connected layer to generate a karst cave classification result.

[0009] As a preferred solution of the intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to the present invention, wherein: before deploying the karst cave prediction model into the shield tunneling system, it further includes training the karst cave prediction model using the training set; evaluating the prediction performance of the karst cave prediction model using at least one evaluation index, and the evaluation index includes but is not limited to accuracy, precision, mean square error, and mean absolute error; validating the trained karst cave prediction model using the test set and generating an error distribution map; adjusting the parameters of the karst cave prediction model according to the test results, including but not limited to selecting the number of network layers, the size of the convolutional kernel, the number of LSTM units, and the number of fully connected network layers.

[0010] As a preferred embodiment of the intelligent classification and prediction method for karst cave data based on Kalman filter and improved CNN-LSTM of the present invention, the convolutional layer includes a first convolutional block, a max pooling layer, and a second convolutional block; the first convolutional block and the second convolutional block are composed of a convolutional layer, a batch normalization layer, and a ReLU activation function, and the convolutional kernel size is 2; the number of convolutional kernels in the first convolutional block is 16; the number of convolutional kernels in the second convolutional block is 32; the first convolutional block slides on the input feature map through convolutional operations, extracts local spatial features, and generates a first feature map; the second convolutional block processes the first feature map from the first convolutional block, and through convolutional operations, extracts a second feature map.

[0011] As a preferred embodiment of the intelligent classification and prediction method for karst cave data based on Kalman filter and improved CNN-LSTM of the present invention, the LSTM layer has 64 hidden units and processes the time series data output by the convolutional layer; the fully connected layer includes a first fully connected layer with 128 neurons and a second fully connected layer with 64 neurons; the first fully connected layer performs a linear transformation on the input feature data and processes it through a non-linear activation function to generate a first output; the second fully connected layer receives the first output from the first fully connected layer, performs a linear transformation and non-linear activation processing to generate a second output for karst cave classification.

[0012] As a preferred embodiment of the intelligent classification and prediction method for karst cave data based on Kalman filter and improved CNN-LSTM of the present invention, the karst cave prediction model further includes that the input time series data is converted into a multi-dimensional data format suitable for convolutional neural network processing through a sequence folding layer, and the data of each time step is mapped to different feature dimensions in the multi-dimensional matrix; after being processed by the convolutional neural network, the sequence unfolding layer is used to convert the output data of the convolutional layer back into the time series format, and the data of each time step corresponds to a sample point.

[0013] Another object of the present invention is to provide an intelligent classification and prediction system for karst cave data based on Kalman filter and improved CNN-LSTM. The present invention solves the problem of how to predict the karst cave state in front of the shield machine based on a large amount of tunneling parameter data and geological parameters, so as to timely adjust the shield tunneling parameters and reduce the construction risk.

[0014] As a preferred solution of an intelligent classification and prediction system for karst cave data based on Kalman filtering and improved CNN-LSTM according to the present invention, it is characterized by including: a collection module for collecting shield tunneling parameter data to obtain karst cave feature information in front of the shield machine; a preprocessing module for performing data cleaning on the shield tunneling parameter data and the karst cave feature information; a prediction module for establishing a karst cave prediction model through a hybrid model structure of a convolutional neural network, a long short-term memory network, and a fully connected network; and a classification module for deploying the karst cave prediction model into the shield tunneling system, inputting the shield tunneling parameter data, and performing karst cave classification and prediction.

[0015] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM as described above.

[0016] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM as described above.

[0017] The beneficial effects of the present invention: The Kalman filtering algorithm is used in data preprocessing to perform denoising on the data, reducing the influence of random noise on model training, thereby improving the accuracy of the network model.

[0018] Through the CNN+LSTM+FCN hybrid model, by combining the local feature extraction ability of the convolutional neural network (CNN), the time series prediction ability of the long short-term memory network (LSTM), and the feature fusion and mapping advantages of the fully connected network (FCN), high-precision prediction of karst caves is achieved.

[0019] This model can effectively fuse the local change characteristics of shield tunneling parameters and the time series of shield tunneling parameters, capture the local morphological anomalies and dynamic change trends of karst caves, and significantly improve the prediction accuracy (up to more than 90%).

[0020] During shield tunneling, the tunneling parameters can be analyzed in real time, and karst cave prediction and dynamic early warning can be carried out. In engineering, the traditional detection cost can be reduced by about 30%, and the construction risk can be reduced, providing an intelligent solution for karst cave prediction under complex geological conditions and having important engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0022] Figure 1 This is the overall flowchart of an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM provided by an embodiment of the present invention.

[0023] Figure 2 This is the schematic diagram of the Kalman filtering and improved CNN-LSTM network structure of an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM provided by an embodiment of the present invention.

[0024] Figure 3 This is the comparison chart of the prediction results of the training set of an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM provided by an embodiment of the present invention.

[0025] Figure 4 This is the comparison chart of the prediction results of the test set of an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM provided by an embodiment of the present invention. Specific Embodiments

[0026] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0027] Example 1, referring to Figure 1 - Figure 2 , which is the first embodiment of the present invention. This embodiment provides an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM, including:

[0028] S101: Collect shield tunneling parameter data to obtain the karst cave feature information in front of the shield machine.

[0029] In step S101, the shield tunneling parameter data at least includes shield thrust, cutter head torque, shield tunneling speed, soil chamber pressure, synchronous grouting pressure, synchronous grouting volume, and geological parameters.

[0030] Characteristics information of karst caves; including whether there are karst caves, the filling degree of karst caves, and whether the karst caves are within the tunnel range.

[0031] S102: Clean the shield tunneling parameter data and the characteristics information of karst caves.

[0032] In step S102, collect the above shield tunneling parameter data and the corresponding geological or karst cave characteristics, use the Kalman filter algorithm (KF) for denoising, and perform data preprocessing.

[0033] In a preferred embodiment of the present invention, the interpolation method is used to fill in the missing values; the Kalman filter algorithm is used to remove noise.

[0034] Specifically, first define the Kalman filter function simple_kalman_filter in matlab, and then use the code filtered_res = res to call the Kalman filter function to perform denoising processing on the data.

[0035] Among the parameters of the Kalman filter algorithm, the state transition matrix is set to 1, the observation matrix is set to 1, the process noise covariance is set to 0.1, and the observation noise covariance is set to 0.5.

[0036] In an alternative embodiment of the present invention, the noise processing steps of the Kalman filter algorithm are initialization, prediction, and then data update. Specifically, when the actual data of the shield tunneling parameters of the sample is transmitted, first predict the current state, and then correct it according to the observed value to obtain the optimal estimate and remove the noise in the data.

[0037] During the tunneling process of the shield machine, due to large environmental changes, data errors may occur. The Kalman filter can estimate the shield tunneling parameters at the current moment and the data at the previous moment in real time, correct the shield machine sensor data effectively, remove the data noise, and retain the true change trend of the shield tunneling parameters; the essence of using the improved CNN-LSTM algorithm for classifying and predicting karst caves is to use the short-term large-scale changes in the shield tunneling parameters at a certain moment to distinguish karst caves. The Kalman filter can greatly reduce the short-term large-scale changes in the shield tunneling parameters caused by other factors, thereby improving the prediction accuracy.

[0038] Divide the data set into a training set (70%) and a test set (30%) in chronological order.

[0039] In another alternative embodiment of the present invention, parameters with different dimensions are placed in the range of [0,1]. The maximum-minimum method is used to normalize the original data, and the normalization formula is as follows:

[0040] x k =(x k -x min ) / (xmax -x min )

[0041] where x min is the minimum value in the data sequence; x max is the maximum value in the sequence.

[0042] Convert the shield tunneling parameter data into a three-dimensional data format (feature number, time step, number of samples). For example, assume that each sample contains 1 time step, and each time step has 7 features (shield thrust, cutterhead torque, shield tunneling speed, soil chamber pressure, synchronous grouting pressure, synchronous grouting volume, formation lithology), then the data shape is (7, 1, number of samples).

[0043] It should be noted that in the data shape, 7 represents the above 7 features of shield thrust, cutterhead torque, shield tunneling speed, soil chamber pressure, synchronous grouting pressure, synchronous grouting volume, and formation lithology; 1 represents the time step. Specifically, when the shield machine is tunneling, a series of shield tunneling parameters will be generated for each ring of segments, that is, the above 7 features. One sample in the dataset is a row of input data, that is, a time step and 7 feature values.

[0044] It should be noted that the purpose of step S102 is to convert the individual data in the dataset into a format suitable for input to the neural network model, that is, to convert the individual data into an array format, which conforms to the format in the input layer of the network. In the dataset of this article, one sample has 7 features, and here it is to convert the data of 7 features into an array format.

[0045] S103: Establish a karst cave prediction model through a hybrid model structure of a convolutional neural network, a long short-term memory network, and a fully connected network. The model structure is as Figure 2 shown.

[0046] In step S103, the method of this embodiment preferably adopts a hybrid model structure of two-layer convolutional neural network (CNN) + long short-term memory network (LSTM) + two-layer fully connected network (FCN).

[0047] It should be noted that existing CNN, LSTM, FCN and other individual neural network models can be used to predict the existence and classification of karst caves, but the accuracy is not high. The hybrid structure model is adopted to give full play to the different advantages of CNN, LSTM and FCN to improve the prediction accuracy of the model. CNN can extract the features of large local changes in the shield tunneling parameters, which can exactly reflect the short-term large changes in the shield tunneling parameters when the shield encounters a karst cave. LTSM can adjust the retention and forgetting of information data, so as to remember long-term dependencies and has advantages in time series prediction. In this paper, the input value - shield tunneling parameters change with the tunneling time of the shield machine, so it has advantages in the classification prediction of karst caves. FCN can extract the features extracted by the two neural networks of CNN and LSTM and map them to the output - karst cave classification. The essential role of FCN is to fuse the two features.

[0048] Furthermore, when the shield machine encounters a karst cave, the shield tunneling parameters will have a large local mutation. Two-layer CNN neural network can more accurately extract the features of such local mutations in the shield tunneling parameter changes. Among them, the convolutional layer network structure is: the first convolutional block, the max pooling layer, and the second convolutional block.

[0049] The first convolutional block contains a convolutional layer, batch normalization, and ReLu activation. The convolutional kernel size is 2, the number of convolutional kernels is 16, and the ReLU function is used as the activation function.

[0050] The stride of the max pooling layer is [2, 1].

[0051] The second convolutional block contains a convolutional layer, batch normalization, and ReLu activation. The convolutional kernel size is 2, the number of convolutional kernels is 32, and the ReLU function is used as the activation function.

[0052] It should be noted that the above convolutional kernel size, stride, and number of convolutional kernels are obtained through multiple debugging on the basis of conventional values to improve the prediction accuracy of the neural network model.

[0053] In a preferred embodiment of the present invention, the convolutional neural network (CNN) part realizes feature extraction through two convolutional blocks. After the input data is preprocessed, the original feature matrix is converted into a 7×1×1 "pseudo-image" structure suitable for convolutional operations.

[0054] The first convolutional block consists of a convolutional layer, batch normalization, and ReLU activation. The convolutional layer uses a kernel size of 2×1. Set 16 feature maps and keep the output size unchanged to ensure that the information at all feature positions is effectively captured. The batch normalization layer normalizes the feature maps, alleviates the problem of gradient disappearance and accelerates the training convergence, and the ReLU activation function introduces non-linear expression ability.

[0055] The max pooling layer then downsamples the feature dimensions using a 2×1 window, reducing computational effort while preserving key information, and compressing the output dimensions to half the original number of features. This design is particularly important in cave data, highlighting significant patterns of sudden changes in formations or parameters.

[0056] The second convolutional block further deepens feature extraction, retaining the 2×1 convolution kernel but increasing the number of feature maps to 32. This adjustment enhances the model's ability to discern complex feature combinations, such as those caused by co-variation of multiple parameters or nonlinear interactions. Batch normalization and ReLU activations then further optimize the feature distribution and strengthen nonlinear transformations, enabling the model to extract higher-order, abstract features from the reduced-dimensional data. The output feature maps now contain the twice-refined local features of the shield tunneling parameters.

[0057] The beneficial effect of this preferred technical solution is that the first convolution block is for basic local pattern extraction, using a 2×1 convolution kernel to slide on the feature dimension to capture the local linear relationship between adjacent parameters. For example, the coordinated change relationship between shield thrust and cutter head torque can be obtained. The second convolution block is for high-order abstract feature extraction. After the feature dimension is halved in the maximum pooling layer, the second convolution block is used to mine the complex nonlinear relationship across features. Two convolution blocks are equivalent to one convolution block, which enables the network model to change from extracting simple linear patterns to complex nonlinear patterns. The LSTM layer has 64 hidden units, which processes the time series data output by the convolution layer to capture the long-term change pattern of shield tunneling parameters when the shield machine is excavating in the stratum.

[0058] The two-layer convolutional neural network includes the first fully connected layer containing 128 neurons and the second fully connected layer containing 64 neurons.

[0059] It should be noted that the two connection layers can increase the nonlinearity of the network, while multiple fully connected layers can gradually reduce the dimension, retain more useful features, and improve the accuracy of the shield machine in predicting the cave characteristic information according to the changes in shield tunneling parameters.

[0060] In an optional embodiment of the present invention, another implementation is possible, in which two fully connected layers are placed after the LSTM layer. Fully connected layer 1 expands the 64-dimensional temporal features output by the LSTM layer to 128 dimensions to accommodate more complex feature combinations, such as the cumulative effect of cutterhead torque over time and the coordinated changes in soil bin pressure. A nonlinear mapping f(x) = max(0, x) is then introduced to address the problem of misidentification caused by similar shield tunneling parameters in different strata. A dropout layer is then introduced to randomly discard 50% of the neuron outputs during network training, forcing the network to not rely on a single feature.

[0061] The fully connected layer 2 compresses the 128-dimensional features output by the fully connected layer 1 to 64 dimensions, removing the information irrelevant to the karst cave type and retaining the key classification features. Subsequently, a Dropout layer is introduced to further improve the generalization ability, so that even if there are missing features in the data, the network model can still make stable predictions through other features.

[0062] In this alternative embodiment, the fully connected layer 1 increases the model complexity to capture subtle pattern differences, and the fully connected layer 2 prevents overfitting and prevents memorizing noise with limited data. Additionally, the 64-dimensional temporal features output by the LSTM layer may contain a large number of weakly correlated signals, and the two fully connected layers screen out strongly correlated features through non-linear combinations.

[0063] Furthermore, in the connection of the convolutional neural network (CNN), long short-term memory network (LSTM), and fully connected network (FCN), sequence folding layers and sequence unfolding layers are adopted for data format conversion. In Matlab, the sequenceFoldingLayer command is used to establish a sequence folding layer to convert the data dimension of the input layer into an image format to adapt to the data input of the convolutional layer; the sequenceUnfoldingLayer command is used to establish a sequence unfolding layer to convert the data dimension of the convolutional layer into a one-dimensional time series format to adapt to the data input of the long short-term memory network layer.

[0064] It should be noted that the sequenceUnfoldingLayer command and the sequenceFoldingLayer command play a role in data conversion of the sequence folding layer in Matlab, converting the sequence data into a spatial format suitable for the input of the convolutional layer. It can be directly described as establishing a sequence unfolding layer.

[0065] It should be noted that the existing method can predict hidden large karst caves based on an RBF neural network optimized by a genetic algorithm. Specifically, 1) organize the dataset of hidden large karst caves, where the data includes lithology thickness, surrounding rock fragmentation degree, topography, geological structure, precipitation temperature, and groundwater; 2) perform normalization processing on the dataset; 3) divide the dataset into a training dataset and a test dataset, use the training dataset to train the RBF neural network, and optimize the structural parameters of the RBF neural network using a genetic algorithm; 4) test the trained RBF neural network with the test dataset; 5) use the RBF neural network that passes the test for actual prediction.

[0066] However, after the fusion of the genetic algorithm and the RBF neural network, the training complexity increases significantly. The RBF neural network adjusts parameters through gradient descent iteration, and the genetic algorithm also requires multiple iterations. The combination of the two will result in double iteration, leading to too long calculation time and unable to guarantee the real-time performance of karst cave prediction in engineering practice. At the same time, the parameter tuning of this method is cumbersome. The parameters in the genetic algorithm include the selection of parameters such as population size, crossover probability, and mutation probability, which require repeated training of the network for debugging. Coupled with the parameters of the RBF neural network including the number of hidden layer nodes, width of the basis function, etc., which need to cooperate with the genetic algorithm, it further increases the difficulty of parameter tuning.

[0067] In practical engineering, due to the real-time performance and multi-noise characteristics of karst cave prediction, this method is difficult to implement in practice.

[0068] S104: Deploy the karst cave prediction model into the shield tunneling system, input the shield tunneling parameter data, and conduct karst cave classification prediction.

[0069] It should be noted that the prediction results include whether there is a karst cave and the description of the karst cave state. The prediction results can help the shield machine operator judge whether there is a karst cave ahead and the classification type of the karst cave by observing the changes in the shield tunneling parameters.

[0070] Example 2, referring to Figure 3 - Figure 4 This is the second embodiment of the present invention, which provides an intelligent classification prediction method for karst cave data based on Kalman filter and improved CNN-LSTM. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0071] S201: Predict whether there is a karst cave ahead according to the shield tunneling parameter data. Input the shield tunneling parameters (shield thrust, cutterhead torque, shield tunneling speed, soil chamber pressure, synchronous grouting pressure, synchronous grouting volume, formation lithology), and output different classifications (there is a karst cave / there is no karst cave, filling degree of the karst cave, the karst cave is within the tunnel range / outside the tunnel range).

[0072] First, it is necessary to collect the time series data of shield thrust, cutterhead torque, shield tunneling speed, soil chamber pressure, synchronous grouting pressure, and synchronous grouting volume changing with the segment ring number from the shield control system. Secondly, combine physical geological exploration to obtain geological parameters, as well as the presence of karst caves (0 no karst cave / 1 there is a karst cave), output of karst cave filling conditions (0 no filling / 1 half filling / 2 full filling), and whether the karst cave is within the tunnel range (0 outside the tunnel range / 1 within the tunnel range) at the corresponding position of each segment. Part of the collected data is shown in Table 1 below.

[0073] Table 1 Data Details

[0074]

[0075] The last three columns of the dataset output are classified into one category through permutation and combination. For example, if the last three columns of the output are [1, 0, 0], it is denoted as category 1, and if they are [1, 0, 1], it is denoted as category 2. Therefore, the last column of the dataset can become the output classification prediction data, as shown in Table 2.

[0076] Table 2 Prediction Details

[0077]

[0078] S202: Collect the above shield tunneling parameter data and the corresponding geological or karst cave characteristics, and perform preprocessing.

[0079] Based on S201, the dataset is divided into a training set (70%) and a test set (30%); subsequently, the mapminmax() function is used in Matlab to normalize the dataset so that the data is between [0, 1].

[0080] S203: Establish a hybrid model structure of a convolutional neural network (CNN) + long short-term memory network (LSTM) + fully connected network (FCN) based on the shield tunneling parameter data. Input layer: Receive the shield tunneling parameter data; CNN layer: Extract local features: the change pattern of shield tunneling parameters in a short time; LSTM layer: Capture long-term dependencies: the trend of tunneling parameter changes; FCN layer: Map the data features to the output.

[0081] The CNN layer uses a one-dimensional convolutional layer in the time dimension to extract local features. The convolutional layer contains two convolutional blocks, and the convolutional layer network structure is: the first convolutional block, the max pooling layer, and the second convolutional block.

[0082] Convolutional block 1 contains a convolutional layer, batch normalization, and ReLu activation. The convolutional kernel size is 2, the number of convolutional kernels is 16, and the activation function uses the ReLU function. The max pooling layer has a stride of [2, 1]. Convolutional block 2 contains a convolutional layer, batch normalization, and ReLu activation. The convolutional kernel size is 2, the number of convolutional kernels is 32, and the activation function uses the ReLU function. The LSTM layer has 64 hidden units.

[0083] The hybrid model structure of convolutional neural network (CNN) + long short-term memory network (LSTM) + fully connected network (FCN) contains 3 fully connected layers. The first fully connected layer contains 128 neurons, the second fully connected layer contains 64 neurons, and the third fully connected layer outputs the category. The sequence folding layer and the sequence unfolding layer are adopted in the network model for data format conversion.

[0084] S204: Use the training set to train the model; use the validation set to monitor the model performance to prevent overfitting.

[0085] In Matlab, call net = trainNetwork(p_train, t_train, lgraph, options) to train the network model. In the network training parameters, select the Adam optimizer, the maximum number of training times is 300, the initial learning rate is 0.001, the strength parameter of L2 regularization is 0.005, and the learning rate decay factor is 0.1. After 50 training times, the learning rate is 0.001 * 0.1, and the model training results are as Figure 3 shown.

[0086] It should be noted that the network training command uses the data in the training set to train the functions in the network model, and obtains a network model with optimized weights and network structure, which can be directly used for sample prediction of the test set.

[0087] S205: Use accuracy, precision, mean square error, and mean absolute error as evaluation indicators; use the test set to analyze the prediction results of the model and generate an error distribution diagram.

[0088] In Matlab, input the codes for accuracy, precision, mean square error, and mean absolute error, and output the results during the code running. The test set prediction results are as Figure 4 shown.

[0089] S206: According to the prediction result error, adjust the machine learning parameters, increase or decrease the number of network layers, adjust the convolutional kernel size, the number of LSTM units, and the number of fully connected network layers.

[0090] For more complex data, the convolutional layer can extract more features by selecting different convolutional kernel sizes or increasing the number of convolutional layers. At the same time, increasing the number of hidden units in the LSTM layer can improve the prediction performance of the network model.

[0091] S207: Deploy the trained model to the shield tunneling system, read the shield tunneling parameter data in real time, and perform karst cave prediction.

[0092] Embodiment 3 is the third embodiment of the present invention. The difference from the previous two embodiments is:

[0093] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0094] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a predefined sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device.

[0095] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part with one or more wirings (electronic device), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), fiber optic device, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as necessary, and then storing it in a computer memory.

[0096] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logic functions on data signals, application-specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0097] Embodiment 4, the fourth embodiment of the present invention, which provides an intelligent classification and prediction system for karst cave data based on Kalman filtering and improved CNN-LSTM, including.

[0098] An acquisition module for acquiring shield tunneling parameter data to obtain karst cave feature information in front of the shield machine;

[0099] A preprocessing module for cleaning the shield tunneling parameter data and karst cave feature information;

[0100] A prediction module for establishing a karst cave prediction model through a hybrid model structure of a convolutional neural network, a long short-term memory network, and a fully connected network;

[0101] A classification module for deploying the karst cave prediction model into the shield tunneling system, inputting the shield tunneling parameter data, and performing karst cave classification and prediction.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM, characterized in that: including, collecting shield tunneling parameter data to obtain karst cave characteristic information in front of the shield machine; performing data cleaning on the shield tunneling parameter data and the karst cave characteristic information; establishing a karst cave prediction model through a hybrid model structure of a convolutional neural network, a long short-term memory network, and a fully connected network; deploying the karst cave prediction model into the shield tunneling system, inputting the shield tunneling parameter data, and performing karst cave classification prediction.

2. The intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to claim 1, wherein: The data cleaning includes, filling in the missing values in the shield tunneling parameter data through interpolation; using the Kalman filter algorithm to denoise the filled data; performing normalization processing on the denoised data to convert shield tunneling parameters with different dimensions into the same numerical range; dividing the normalized data into a training set and a test set in chronological order.

3. The intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to claim 2, characterized in that: The karst cave prediction model includes, an input layer that receives shield tunneling parameter data; a convolutional layer that uses a convolutional neural network to extract local features of the shield tunneling parameters; an LSTM layer that uses a long short-term memory network to model the time series data of the tunneling parameters and extract long-term dependencies; a fully connected layer that uses a fully connected network to fuse the features extracted by the convolutional neural network and the long short-term memory network; an output layer that outputs the fusion result of the fully connected layer and generates a karst cave classification result.

4. The intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to claim 3, characterized in that: Before deploying the karst cave prediction model into the shield tunneling system, it also includes, training the karst cave prediction model using the training set; evaluating the prediction performance of the karst cave prediction model using at least one evaluation index, where the evaluation index includes but is not limited to accuracy, precision, mean square error, and mean absolute error; validating the trained karst cave prediction model using the test set and generating an error distribution map; adjusting the parameters of the karst cave prediction model according to the test results, including but not limited to selecting the number of network layers, the size of the convolutional kernel, the number of LSTM units, and the number of fully connected network layers.

5. The intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to claim 4, characterized in that: The convolutional layer includes a first convolutional block, a max pooling layer, and a second convolutional block; The first convolutional block and the second convolutional block are composed of a convolutional layer, a batch normalization layer, and a ReLU activation function, with a convolutional kernel size of 2; the number of convolutional kernels in the first convolutional block is 16; the number of convolutional kernels in the second convolutional block is 32; The first convolutional block slides on the input feature map through convolutional operations, extracts local spatial features, and generates a first feature map; The second convolutional block processes the first feature map from the first convolutional block, and extracts a second feature map through convolutional operations.

6. The intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to claim 5, characterized in that: The LSTM layer has 64 hidden units and processes the time series data output by the convolutional layer; The fully connected layer includes a first fully connected layer with 128 neurons and a second fully connected layer with 64 neurons; The first fully connected layer performs a linear transformation on the input feature data and processes it through a non-linear activation function to generate a first output; The second fully connected layer receives the first output from the first fully connected layer, performs linear transformation and non-linear activation processing, generates a second output, and performs karst cave classification.

7. The intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM according to claim 6, characterized in that: The karst cave prediction model also includes, converting the input time series data into a multi-dimensional data format suitable for processing by a convolutional neural network through a sequence folding layer, and mapping the data at each time step to different feature dimensions in the multi-dimensional matrix; After being processed by a convolutional neural network, a sequence unfolding layer is used to convert the output data of the convolutional layer back into a time series format, and the data for each time step corresponds to a sample point.

8. An intelligent classification and prediction system for karst cave data based on Kalman filtering and improved CNN-LSTM, which applies an intelligent classification and prediction method for karst cave data based on Kalman filtering and improved CNN-LSTM as described in any one of claims 1 to 7, characterized in that, It includes: An acquisition module, which is used to acquire shield tunneling parameter data and obtain the karst cave feature information in front of the shield machine; A preprocessing module, which is used to clean the shield tunneling parameter data and the karst cave feature information; A prediction module, which is used to establish a karst cave prediction model through a hybrid model structure of a convolutional neural network, a long short-term memory network, and a fully connected network; A classification module, which is used to deploy the karst cave prediction model into the shield tunneling system, input the shield tunneling parameter data, and perform karst cave classification prediction.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for intelligent classification and prediction of karst cave data based on Kalman filtering and improved CNN-LSTM according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for intelligent classification and prediction of karst cave data based on Kalman filtering and improved CNN-LSTM according to any one of claims 1 to 7.

Images