Shield tunneling attitude prediction method and system based on improved LSTM

Abstract

The invention discloses a shield tunneling attitude prediction method and system based on an improved LSTM. The method comprises the steps that a shield tunneling attitude prediction model is constructed, a training set is adopted to train the model, a test set is adopted to test the model, and the trained shield tunneling attitude prediction model is obtained; wherein the shield tunneling attitude prediction model comprises a time sequence convolutional network, an improved LSTM network and a multi-head attention mechanism layer which are connected in sequence; according to the improved LSTM network, a peep hole link is added on the basis of the LSTM network, the peep hole link connects the state of a previous unit with a forgetting gate and an input gate, the state of the previous unit is added into calculation of the forgetting gate and the input gate, and shield construction data to be predicted are obtained; and inputting to-be-predicted shield construction data into the trained shield tunneling attitude prediction model to obtain a tunneling attitude prediction result of the to-be-predicted shield construction data.

Description

Technical Field

[0001] The present invention relates to the technical field of shield construction, and in particular to a shield tunneling posture prediction method and system based on an improved LSTM. Background Art

[0002] With the continuous advancement of urbanization, underground space planning and construction have developed rapidly. The shield method has become the main construction method for tunnel construction due to its good adaptability, high safety, fast construction and low impact on the ground environment. During the shield tunneling process, the shield attitude is a parameter that needs to be strictly controlled. Deviations in the shield attitude often cause the tunneling axis of the shield machine to present a "snake-like" movement along the track design axis (Design Tunnel Axis, DTA), thereby affecting the quality of subsequent segment lining. In severe cases, it may even cause irreversible hazards such as segment cracking and excessive surface settlement. Therefore, it is necessary to predict the shield attitude deviation in advance during the tunneling process.

[0003] At present, the control of shield machine posture mainly relies on the driver's operating experience. The subjective factors are large and the operation is lagging. It is difficult to adapt to the complex underground construction environment in a timely and effective manner, and the correction accuracy and efficiency are difficult to guarantee. Therefore, it is urgent to conduct prediction research on shield posture under complex conditions to make shield construction more efficient.

[0004] With the rapid development of computer technology in recent years, artificial intelligence-based methods, such as machine learning and deep learning, have gradually been applied to the problem of shield machine posture prediction. However, the existing research methods still have some shortcomings: shield machine posture prediction currently mainly uses long short-term memory neural networks to capture the time dependency of input data, but fails to fully capture important local features and each gate control unit will lose the information processed in the previous sequence. At the same time, existing research does not fully explain the impact of changes in geological conditions on shield machine posture, and there is little research on the impact of different strata on shield tunneling posture. Summary of the invention

[0005] In order to solve the deficiencies of the prior art, the present invention provides a shield tunneling posture prediction method and system based on an improved LSTM, that is, the factors affecting the posture of the shield machine are summarized, and the formation parameters and tunneling parameters related to the posture of the shield machine are analyzed based on the data collected on the actual site, and the original data are pre-processed, and the input features are further screened from the preliminary screening data based on the Pearson correlation analysis. After the input features are denoised, the short-term local features in the sequence are captured by the Temporal Convolutional Network (TCN), the long-term correlation in the sequence is captured by the improved LSTM, and the features are weighted by the multi-head attention mechanism (MHA), so as to improve the accuracy of the shield posture prediction method.

[0006] On the one hand, a shield tunneling posture prediction method based on improved LSTM is provided, including:

[0007] Acquire historical shield construction data; preprocess the historical shield construction data; determine a training set and a test set based on the preprocessed data; the training set and the test set include: historical shield construction data of whether the shield tunneling posture is known to deviate or not;

[0008] Construct a shield tunneling posture prediction model, use a training set to train the model, use a test set to test the model, and obtain a trained shield tunneling posture prediction model; wherein, the shield tunneling posture prediction model includes: a sequentially connected temporal convolutional network, an improved LSTM network, and a multi-head attention mechanism layer; the improved LSTM network is based on the LSTM network, and a peephole link is added. The peephole link connects the previous unit state with the forget gate and the input gate respectively, and adds the previous unit state to the calculation of the forget gate and the input gate, so that the current state does not lose the information already obtained by the upper state, so that the output at the next moment is complete;

[0009] Obtain the shield construction data to be predicted; input the shield construction data to be predicted into the trained shield tunneling posture prediction model to obtain the tunneling posture prediction result of the shield construction data to be predicted.

[0010] On the other hand, a shield tunneling posture prediction system based on improved LSTM is provided, including:

[0011] The acquisition module is configured to: acquire historical shield construction data; preprocess the historical shield construction data; determine a training set and a test set based on the preprocessed data; the training set and the test set include: historical shield construction data of whether the shield excavation posture is known to deviate or not;

[0012] A model training module is configured to: construct a shield tunneling posture prediction model, train the model with a training set, test the model with a test set, and obtain a trained shield tunneling posture prediction model; wherein the shield tunneling posture prediction model includes: a sequentially connected temporal convolutional network, an improved LSTM network, and a multi-head attention mechanism layer; the improved LSTM network is based on the LSTM network, and a peephole link is added. The peephole link connects the previous unit state with a forget gate and an input gate respectively, and adds the previous unit state to the calculation of the forget gate and the input gate, so that the current state does not lose the information obtained by the upper state, so that the output at the next moment is complete;

[0013] The prediction module is configured to: obtain the shield construction data to be predicted; input the shield construction data to be predicted into the trained shield excavation posture prediction model to obtain the excavation posture prediction result of the shield construction data to be predicted.

[0014] On the other hand, there is also provided an electronic device, comprising:

[0015] a memory for non-transitory storage of computer-readable instructions; and

[0016] a processor for executing the computer readable instructions,

[0017] When the computer-readable instructions are executed by the processor, the method described in the first aspect is executed.

[0018] On the other hand, a storage medium is provided, which non-temporarily stores computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the instructions of the method described in the first aspect are executed.

[0019] On the other hand, a computer program product is provided, comprising a computer program, wherein the computer program is used to implement the method described in the first aspect when running on one or more processors.

[0020] The above technical solution has the following advantages or beneficial effects:

[0021] (1) The preferred input parameter data set of the present invention introduces different formation parameters into the data set on the basis of using the shield machine excavation parameters related to the deviation of the shield posture. By comprehensively considering the shield machine excavation parameters and the formation parameters and performing data preprocessing on the input parameters, a more comprehensive and accurate prediction of the shield posture can be achieved.

[0022] (2) The present invention captures local features in the sequence through TCN. Since the data to be processed is time series data, traditional convolutional neural networks are difficult to handle the long-term dependencies of time series data during shield tunneling, resulting in a decrease in model performance. TCN expands the receptive field through dilated convolution, so that long-term dependencies can be captured and modeled more effectively. TCN stacks multiple convolutional layers to extract features of different scales, so that TCN can effectively extract local features of time series data.

[0023] (3) The present invention uses an improved LSTM to capture long-term correlations in a sequence through an improved long short-term memory neural network. Since the computing unit of the long short-term memory neural network will lose the information processed in the previous sequence, in order to improve the integrity of the information and the accuracy of the output at the next moment, a peephole link is added on the basis of the long short-term memory neural network, the state of the previous unit is linked to the gate control unit, and the state of the previous unit is also added to the gate control unit calculation, so that the current state will not lose the information already obtained by the upper state, making the output at the next moment more specific and complete; at the same time, MHA is introduced to perform weighted processing on the data of the improved LSTM layer, so that more weights can be allocated to important inputs, thereby improving the prediction accuracy of the shield machine posture. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings in the specification, which constitute 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 improper limitations on the present invention.

[0025] Figure 1 The present invention is a flow chart of the shield posture hybrid prediction method.

[0026] Figure 2 Schematic diagram of the temporal convolutional network structure of the present invention.

[0027] Figure 3 Schematic diagram of the improved LSTM unit structure of the present invention. DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0029] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0030] In this embodiment, all data is obtained in compliance with laws and regulations and based on the user's consent, and is used legally.

[0031] Embodiment 1

[0032] This embodiment provides a shield tunneling posture prediction method based on an improved LSTM;

[0033] like Figure 1 As shown in the figure, the shield tunneling posture prediction method based on the improved LSTM includes:

[0034] S101: Acquire historical shield construction data; preprocess the historical shield construction data; determine a training set and a test set based on the preprocessed data; the training set and the test set include: historical shield construction data of whether the shield tunneling posture is known to deviate or not;

[0035] S102: constructing a shield tunneling posture prediction model, training the model using a training set, and testing the model using a test set to obtain a trained shield tunneling posture prediction model;

[0036] The shield tunneling posture prediction model includes: a sequentially connected temporal convolutional network, an improved LSTM network and a multi-head attention mechanism layer; the improved LSTM network is based on the LSTM network, and a peephole link is added. The peephole link connects the previous unit state with the forget gate and the input gate respectively, and adds the previous unit state to the calculation of the forget gate and the input gate, so that the current state does not lose the information obtained by the upper state, so that the output at the next moment is complete;

[0037] S103: Acquire the shield construction data to be predicted; input the shield construction data to be predicted into the trained shield tunneling posture prediction model to obtain the tunneling posture prediction result of the shield construction data to be predicted.

[0038] It should be understood that the formation parameters and excavation parameters collected and screened on site are used as input data, based on TCN to capture short-term local features in the sequence, and a peephole link is added on the basis of the long short-term memory neural network, which is called the improved LSTM, which adds the state of the previous unit to the gated unit calculation, so that the current state will not lose the information already obtained by the upper state, making the output at the next moment more specific and complete, so as to capture the long-term correlation in the sequence. The introduction of MHA gives higher weights to important features, improves the accuracy of prediction, and finally guides the correction through the prediction results.

[0039] Furthermore, the S101: acquiring historical shield construction data, the historical shield construction data including: shield machine construction parameters, stratum parameters and tunnel geometry parameters;

[0040] The shield machine construction parameters include: the shield machine's cutter head torque, cutter head speed, total propulsion force, excavation speed, penetration rate, working chamber pressure, excavation chamber pressure, grouting pressure, propulsion cylinder thrust, propulsion cylinder stroke, shield tail grease amount and cutter head telescopic cylinder stroke;

[0041] The formation parameters include: natural gravity, cohesion, internal friction angle and lateral pressure coefficient;

[0042] The tunnel geometric parameters include: tunnel burial depth.

[0043] Furthermore, the S101: pre-processing the historical shield construction data includes: removing shutdown state data, outlier processing, data normalization, Pearson correlation analysis and SG filtering noise reduction performed in sequence.

[0044] Furthermore, the elimination of shutdown status data specifically includes:

[0045] The binary discrimination method is used to eliminate the shutdown state data. Only when the excavation speed, total propulsion force, cutter head torque and cutter head speed are not 0, it is considered to be in the excavation state, otherwise it is judged to be in the shutdown state. The shutdown state data is eliminated according to the following formula:

[0046]

[0047] f(t)=G(v)·G(F)·G(T)·G(n);

[0048]

[0049] Among them, v is the excavation speed, F is the total propulsion force, T is the cutter head torque, and n is the cutter head speed.

[0050] Furthermore, the outlier processing includes:

[0051] The outliers of parameters are detected and eliminated based on the box plot method. The box plot method has no requirements on the characteristics of the data itself and has a strong ability to screen outliers.

[0052] Furthermore, the data normalization includes:

[0053] The min-max normalization method is used to map all parameters to [0,1], the formula is as follows:

[0054]

[0055] Among them, x * is the normalized data, x is the original data, and x min is the minimum value of the data, x max is the maximum value of the data.

[0056] Furthermore, the Pearson correlation analysis includes: selecting the optimal input parameter from the parameters, and the formula is as follows:

[0057]

[0058] In the formula, r xy is the Pearson correlation coefficient of the selected feature, x i and i are the i-th values ​​of x and y respectively, and are the average values ​​of x and y respectively. xy The value range is [-1,1], through r xy To measure the correlation between different features, when the r between two features xy The closer it is to 1 or -1, the stronger the correlation between the features, and the closer it is to 0, the weaker the correlation. Positive values ​​indicate positive correlation, and negative values ​​indicate negative correlation. In order to avoid data redundancy, some input parameters with correlation coefficients greater than 0.7 are filtered out to improve the model prediction effect.

[0059] Furthermore, the SG filtering noise reduction uses the Savitzky-Golay (SG) filtering method to perform data noise reduction processing. The SG filtering uses the least squares method to perform weighted averaging on the moving window to implement a polynomial fitting method. The data reconstructed by the SG filtering can well retain local features and the data is smoother.

[0060] Furthermore, determining the training set and the test set based on the preprocessed data includes:

[0061] The sliding window method is used to reconstruct the data; the step size s and prediction time period t of the shield tunneling posture prediction model are determined, and the preprocessed data are continuously read to generate time series data. The data set is divided into 70% as a training set, 10% as a validation set, and the last 20% as a test set.

[0062] The shield machine construction parameters, formation parameters and tunnel geometry parameters are obtained through the sensors of each subsystem of the shield machine and the tunnel geological survey report, and the input and output parameters related to the deviation of the shield posture are screened out through investigation.

[0063] 80% of the feature data of the input sequence data set after data preprocessing is used to train the model. Specifically, TCN is used to capture the short-term local features in the sequence, the improved LSTM is used to capture the long-term correlation in the sequence, and MHA is introduced to weight the features, finally obtaining the shield posture prediction model.

[0064] To test the accuracy of the deep learning model after training, 20% of the feature data of the sequence data set after data preprocessing was used to test the model, and the accuracy of the model prediction was judged through regression analysis indicators.

[0065] Furthermore, the S102: constructing a shield tunneling posture prediction model, training the model with a training set, and testing the model with a test set to obtain a trained shield tunneling posture prediction model, includes:

[0066] The shield tunneling posture prediction model is trained using the training set. When the loss function value of the shield tunneling posture prediction model no longer decreases, the training is stopped to obtain the shield tunneling posture prediction model after preliminary training.

[0067] During the training process, the input parameters of the shield tunneling posture prediction model include: shield machine parameters, formation parameters and tunnel geometry parameters;

[0068] The shield machine parameters include: cutter head torque, cutter head rotation speed, total thrust, excavation speed, penetration rate, working chamber pressure, excavation chamber pressure, grouting pressure, thrust of the propulsion cylinder, propulsion cylinder stroke, shield tail grease amount and cutter head telescopic cylinder stroke; the formation parameters include: natural gravity, cohesion, internal friction angle, lateral pressure coefficient; the tunnel geometric parameters include: tunnel burial depth.

[0069] The output parameters of the shield tunneling posture prediction model include: pitch angle, roll angle, shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation;

[0070] The test set is then input into the shield tunneling posture prediction model after preliminary training. When the prediction accuracy exceeds the set threshold, it means that the shield tunneling posture prediction model after preliminary training is the shield tunneling posture prediction model after final training; if the prediction accuracy is less than the set threshold, the training set is replaced and retrained.

[0071] Furthermore, the prediction accuracy was measured using the root mean square error (RMSE), mean absolute error (MAE) and goodness of fit (R 2 ) is evaluated, and the specific formula is as follows:

[0072]

[0073] Where: represents the shield tunneling posture deviation value predicted by the model; y i Indicates the actual shield tunneling posture deviation data; represents the average value of the real data; f i Represents the calculated predicted value; the smaller the RMSE and MAE values ​​are, the better the R 2The closer the value is to 1, the better the prediction performance of the model.

[0074] Furthermore, if Figure 2 As shown, the time series convolution network TCN extracts local features of the converted time series data through TCN. TCN is a strict time constraint model that does not consider future information. It is a unidirectional structure and has better performance in extracting time series features. It can extract implicit and deep data features by convolving the input data layer by layer and fuse them layer by layer to generate high-dimensional abstract features. TCN extracts features of different scales by stacking multiple convolutional layers, thereby capturing features of different time ranges in the sequence data. Therefore, TCN can be used to extract local features of time series data, which is very suitable for time series data. This is an advantage over CNN.

[0075] The temporal convolutional network is specifically:

[0076] The convolutional layer performs causal convolution operations. For a convolutional layer, the mathematical expression is:

[0077]

[0078] Among them, y(t) is the output of the convolution layer, K is the size of the convolution kernel, and W k is the weight of the kth convolution kernel, x t-k is the input at time tk.

[0079] The convolution operation of the TCN model uses dilated convolution, which expands the receptive field by inserting zero elements between the convolution kernels. The mathematical expression of dilated convolution is:

[0080]

[0081] Among them, x t-d·k are the elements of the input sequence and d is the dilation rate.

[0082] Furthermore, if Figure 3 As shown, the improved LSTM network is specifically:

[0083] Forget gate: The model forgets some information from the cell state and retains important information;

[0084] f t =σ(W f ·[C t-1 ,h t-1 ,x t ]+b f )

[0085] Among them, f t is the output vector of the forget gate, Wf is the weight matrix, C t-1 is the unit state at the previous moment, h t-1 is the output at the previous moment, x t is the current input parameter, b f is the bias vector, and σ represents the sigmoid activation function.

[0086] Input gate: extracts valid information from the current input and updates the entry into the unit state through filtering;

[0087] i t =σ(W i ·[C t-1 ,h t-1 ,x t ]+b i )

[0088] C t ′=tanh(W i [C t-1 ,h t-1 ,x t ]+b i )

[0089] C t =f t ·C t-1 +i t ·C t '

[0090] Among them, i t is the state of the input gate, W i is the weight matrix, b i is the weight matrix, C t ′ represents the candidate value, C t is the new unit state value;

[0091] Output gate: determines the output value at the current moment;

[0092] O t =σ(W o ·[h t-1 ,x t ]+b o )

[0093] h t =O t tanh(C t )

[0094] Among them, O t is the initial output of the output gate, W o is the weight matrix, b o is the weight matrix, h t is the final output vector.

[0095] The improved LSTM neural network captures the long-term dependencies of data features. By adding a peephole link on the basis of the LSTM network, the state of the previous unit is connected with the forget gate and the input gate, and the state of the previous unit is added to the calculation of the forget gate and the input gate, so that the current state will not lose the information already obtained by the upper state, making the output at the next moment more specific and complete.

[0096] Further, the S103: obtains the shield construction data to be predicted; inputs the shield construction data to be predicted into the trained shield tunneling posture prediction model to obtain the tunneling posture prediction result of the shield construction data to be predicted, including: predicting whether the shield posture deviates, based on the established shield posture prediction model, inputting the shield machine parameters, stratum parameters and tunnel geometry parameters related to the shield posture, and outputting the pitch angle, roll angle, shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation, shield tail vertical deviation through the model, judging whether the shield machine will have an excessive tunneling posture deviation in the future according to the output parameters, so as to achieve effective prediction.

[0097] Furthermore, the multi-head attention mechanism layer performs weighted processing on the improved LSTM data. The multi-head attention mechanism layer is an optimization of the self-attention mechanism, integrating different knowledge generated from the same attention pooling, enhancing the representation ability of the model, and extracting more useful information. The specific calculation process of the multi-head attention mechanism layer MHA is as follows:

[0098] The input data is processed through the embedding layer to obtain a1, a2, a3…a T ;

[0099] a i =W·x i

[0100] Use multiple groups of W Q , W K , W V The three weight matrices are respectively i Multiply, we get multiple groups q i , k i , v i , a group of q i ,k i ,v i It is a "head" of the attention mechanism;

[0101] q i =W Q ·a i ;k i =W K ·a i ;vi =W V ·a i ;

[0102] q i Respectively with k1, k2, k3…k T Vector dot product, we get α 1,1 ,α 1,2 ,α 1,3 …α 1,T ;Will

[0103] α 1,1 ,α 1,2 ,α 1,3 …α 1,T Input to the softmax layer and get the attention weight values ​​between 0 and 1:

[0104] Then Respectively with v1, v2, v3…v in the corresponding positions T By multiplying and summing, we get the output corresponding to the input x1

[0105]

[0106] Will Splice and then use linear transformation to get the results b1, b i Refers to the result of the attention mechanism operation between the i-th position information and all position information in the input sequence.

[0107]

[0108] The present invention combines a temporal convolutional network and an improved LSTM hybrid prediction model of shield machine posture, inputs screened shield machine construction parameters, formation parameters and tunnel geometry parameters, preprocesses the original data, and then inputs the input data into a preset shield machine tunneling posture prediction model to obtain a shield machine tunneling posture prediction result.

[0109] Embodiment 2

[0110] This embodiment provides a shield tunneling posture prediction system based on an improved LSTM;

[0111] The shield tunneling posture prediction system based on improved LSTM includes:

[0112] The acquisition module is configured to: acquire historical shield construction data; preprocess the historical shield construction data; determine a training set and a test set based on the preprocessed data; the training set and the test set include: historical shield construction data of whether the shield excavation posture is known to deviate or not;

[0113] A model training module is configured to: construct a shield tunneling posture prediction model, train the model with a training set, and test the model with a test set to obtain a trained shield tunneling posture prediction model; wherein the shield tunneling posture prediction model includes: a sequentially connected temporal convolutional network, an improved LSTM network, and a multi-head attention mechanism layer; the improved LSTM network is based on the LSTM network, and a peephole link is added. The peephole link connects the previous unit state with a forget gate and an input gate, and adds the previous unit state to the calculation of the forget gate and the input gate, so that the current state does not lose the information already obtained by the upper state, so that the output at the next moment is complete;

[0114] The prediction module is configured to: obtain the shield construction data to be predicted; input the shield construction data to be predicted into the trained shield excavation posture prediction model to obtain the excavation posture prediction result of the shield construction data to be predicted.

[0115] It should be noted that the acquisition module, model training module and prediction module correspond to steps S101 to S103 in Embodiment 1, and the examples and application scenarios implemented by the modules are the same as those of the corresponding steps, but are not limited to the contents disclosed in Embodiment 1. It should be noted that the modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0116] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0117] The proposed system can be implemented in other ways. For example, the system embodiment described above is only illustrative, and the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0118] Embodiment 3

[0119] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory so that the electronic device executes the method described in the above embodiment one.

[0120] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0121] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0122] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0123] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0124] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0125] Embodiment 4

[0126] This embodiment further provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the method described in the first embodiment is completed.

[0127] 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. The shield tunneling posture prediction method based on improved LSTM is characterized by: include: Obtain historical shield construction data; Preprocess historical shield construction data; Based on the preprocessed data, determine the training set and test set; The training set and the test set include: historical shield construction data of whether the shield tunneling posture is known to deviate or not; Construct a shield tunneling posture prediction model, use a training set to train the model, use a test set to test the model, and obtain a trained shield tunneling posture prediction model; wherein, the shield tunneling posture prediction model includes: a sequentially connected temporal convolutional network, an improved LSTM network, and a multi-head attention mechanism layer; the improved LSTM network is based on the LSTM network, and a peephole link is added. The peephole link connects the previous unit state with the forget gate and the input gate, and the previous unit state is added to the calculation of the forget gate and the input gate, so that the current state does not lose the information already obtained by the upper state, so that the output at the next moment is complete; Obtain the shield construction data to be predicted; input the shield construction data to be predicted into the trained shield tunneling posture prediction model to obtain the tunneling posture prediction result of the shield construction data to be predicted.

2. The shield tunneling posture prediction method based on improved LSTM as claimed in claim 1 is characterized in that: Acquire historical shield construction data, wherein the historical shield construction data includes: shield machine construction parameters, stratum parameters and tunnel geometry parameters; The shield machine construction parameters include: the shield machine's cutter head torque, cutter head speed, total propulsion force, excavation speed, penetration rate, working chamber pressure, excavation chamber pressure, grouting pressure, propulsion cylinder thrust, propulsion cylinder stroke, shield tail grease amount and cutter head telescopic cylinder stroke; The formation parameters include: natural gravity, cohesion, internal friction angle and lateral pressure coefficient; The tunnel geometric parameters include: tunnel burial depth.

3. The shield tunneling posture prediction method based on improved LSTM as claimed in claim 1 is characterized in that: Preprocessing of historical shield construction data, including: elimination of shutdown data, outlier processing, data normalization, Pearson correlation analysis and SG filtering noise reduction; The elimination of the stop state data specifically includes: using a binary discrimination method to eliminate the stop state data, only when the excavation speed, the total propulsion force, the cutter head torque and the cutter head speed are not 0, it is considered to be in the excavation state, otherwise it is determined to be in the stop state, and the stop state data is eliminated according to the following formula: f(t)=G(v)*G(F)*G(T)*G(n); Among them, v is the excavation speed, F is the total propulsion force, T is the cutter head torque, and n is the cutter head speed; The outlier processing includes: detecting and eliminating parameter outliers based on a box plot method.

4. The shield tunneling posture prediction method based on improved LSTM as claimed in claim 3 is characterized in that: The Pearson correlation analysis includes: selecting the optimal input parameters from the parameters, and the formula is as follows: In the formula, r xy is the Pearson correlation coefficient of the selected feature, x i and i are the i-th values ​​of x and y respectively, and are the average values ​​of x and y respectively; r xy The value range is [-1,1], through r xy To measure the correlation between different features, when the r between two features xy The closer it is to 1 or -1, the stronger the correlation between the features, and the closer it is to 0, the weaker the correlation; positive values ​​indicate positive correlation, and negative values ​​indicate negative correlation; The SG filtering noise reduction is to use the Savitzky-Golay filtering method to perform data noise reduction processing. The SG filtering uses the least squares method to perform weighted averaging on the moving window; The step of determining a training set and a test set based on the preprocessed data includes: The sliding window method is used to reconstruct the data; the step size s and prediction time period t of the shield tunneling posture prediction model are determined, and the preprocessed data are continuously read to generate time series data; the data set is divided into 70% as a training set, 10% as a validation set, and the last 20% as a test set.

5. The shield tunneling posture prediction method based on improved LSTM as claimed in claim 1 is characterized in that: Construct a shield tunneling posture prediction model, use the training set to train the model, use the test set to test the model, and obtain the trained shield tunneling posture prediction model, including: The shield tunneling posture prediction model is trained using the training set. When the loss function value of the shield tunneling posture prediction model no longer decreases, the training is stopped to obtain the shield tunneling posture prediction model after preliminary training. During the training process, the input parameters of the shield tunneling posture prediction model include: shield machine parameters, formation parameters and tunnel geometry parameters; The shield machine parameters include: cutter head torque, cutter head speed, total propulsion force, excavation speed, penetration rate, working chamber pressure, excavation chamber pressure, grouting pressure, propulsion cylinder thrust, propulsion cylinder stroke, shield tail grease amount and cutter head telescopic cylinder stroke; the formation parameters include: natural gravity, cohesion, internal friction angle, lateral pressure coefficient; the tunnel geometric parameters include: tunnel burial depth; The output parameters of the shield tunneling posture prediction model include: pitch angle, roll angle, shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation; The test set is then input into the shield tunneling posture prediction model after preliminary training. When the prediction accuracy exceeds the set threshold, it means that the shield tunneling posture prediction model after preliminary training is the shield tunneling posture prediction model after final training; if the prediction accuracy is less than the set threshold, the training set is replaced and retrained.

6. The shield tunneling posture prediction method based on improved LSTM as claimed in claim 5 is characterized in that: The prediction accuracy is evaluated using root mean square error, mean absolute error, and goodness of fit. The specific formula is as follows: Where: represents the shield tunneling posture deviation value predicted by the model; y i Indicates the actual shield tunneling posture deviation data; represents the average value of the real data; f i Represents the calculated predicted value; the smaller the RMSE and MAE values ​​are, the better the R 2 The closer the value is to 1, the better the prediction performance of the model.

7. The shield tunneling posture prediction method based on improved LSTM as claimed in claim 1 is characterized in that: The improved LSTM network is specifically: Forget gate: The model forgets some information from the cell state and retains important information; f t =σ(W f ·[C t-1 ,h t-1 ,x t ]+b f ); Among them, f t is the output vector of the forget gate, W f is the weight matrix, C t-1 is the unit state at the previous moment, h t-1 is the output at the previous moment, x t is the current input parameter, b f is the bias vector, σ represents the sigmoid activation function; Input gate: extracts valid information from the current input and updates the entry into the unit state through filtering; i t =σ(W i ·[C t-1 ,h t-1 ,x t ]+b i ); C t ′=tanh(W i [C t-1 ,h t-1 ,x t ]+b i ); C t =f t ·C t-1 +i t ·C t ′; Among them, i t is the state of the input gate, W i is the weight matrix, b i is the weight matrix, C t ′ represents the candidate value, C t is the new unit state value; Output gate: determines the output value at the current moment; The t =σ(W o ·[h t-1 ,x t ]+b o ) h t =O t ·tanh(C t ) Among them, O t is the initial output of the output gate, W o is the weight matrix, b o is the weight matrix, h t is the final output vector; The shield construction data to be predicted is obtained; the shield construction data to be predicted is input into the trained shield tunneling posture prediction model to obtain the tunneling posture prediction result of the shield construction data to be predicted, including: predicting whether the shield posture is deviated, based on the established shield posture prediction model, inputting the shield machine parameters, stratum parameters and tunnel geometry parameters related to the shield posture, and outputting the pitch angle, roll angle, shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation through the model, judging whether the tunneling posture deviation of the shield machine will exceed the limit in the future according to the output parameters, so as to achieve effective prediction; the multi-head attention mechanism layer performs weighted processing on the data of the improved LSTM.

8. The shield tunneling posture prediction system based on improved LSTM is characterized by: include: An acquisition module is configured to: acquire historical shield construction data; Preprocess historical shield construction data; Based on the preprocessed data, determine the training set and test set; The training set and the test set include: historical shield construction data of whether the shield tunneling posture is known to deviate or not; A model training module is configured to: construct a shield tunneling posture prediction model, train the model with a training set, and test the model with a test set to obtain a trained shield tunneling posture prediction model; wherein the shield tunneling posture prediction model includes: a sequentially connected temporal convolutional network, an improved LSTM network, and a multi-head attention mechanism layer; the improved LSTM network is based on the LSTM network, and a peephole link is added. The peephole link connects the previous unit state with a forget gate and an input gate, and adds the previous unit state to the calculation of the forget gate and the input gate, so that the current state does not lose the information already obtained by the upper state, so that the output at the next moment is complete; The prediction module is configured to: obtain the shield construction data to be predicted; input the shield construction data to be predicted into the trained shield excavation posture prediction model to obtain the excavation posture prediction result of the shield construction data to be predicted.

9. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; as well as a processor for executing the computer readable instructions, When the computer-readable instructions are executed by the processor, the method described in any one of claims 1 to 7 is executed.

10. A storage medium, characterized in that: The computer-readable instructions are non-transitory stored, wherein when the non-transitory computer-readable instructions are executed by a computer, the instructions of the method according to any one of claims 1 to 7 are executed.

Images