A slope displacement prediction method and system based on a physical information neural network and a gated recurrent unit

Abstract

The present application relates to the technical field of geological disaster monitoring and early warning, and particularly relates to a slope displacement prediction method and system based on a physical information neural network and a gated recurrent unit. The method combines the physical information neural network and the gated recurrent unit, realizes the organic integration of data driving and physical constraints, embeds the slope mechanical control equation and the boundary condition into the neural network training process through the physical information neural network, and still maintains the physical rationality of the prediction result when the monitoring data is insufficient or has noise. Meanwhile, the gated recurrent unit effectively captures the nonlinear time sequence characteristics of the slope displacement under complex working conditions such as rainfall, water level and construction, and improves the modeling capability for the dynamic evolution process. The present application not only improves the precision and stability of the slope displacement prediction, but also enhances the generalization capability and engineering interpretability of the model under different working conditions, realizes the real-time early warning of the slope instability risk, and provides more reliable technical support for the prevention and control of geological disasters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological disaster monitoring and early warning, and in particular to a slope displacement prediction method and system based on a physical information neural network and a gated recurrent unit. BACKGROUND

[0002] Slope stability analysis and displacement prediction have always been the core problems in the fields of geotechnical engineering and disaster prevention. With the acceleration of urbanization, more and more roads, buildings and other infrastructure construction need to rely on slopes or inclined terrains. Once the slope loses stability, it often accompanies large-scale displacement of soil or rock mass, which may pose a serious threat to infrastructure and resident safety. Slope instability is usually characterized by progressive displacement accumulation. If the displacement development trend can be accurately predicted, potential risks can be identified in advance to guide prevention measures. Therefore, how to use existing monitoring data and theoretical models to accurately predict the spatio-temporal evolution process of slope displacement has always been the focus of the academic and engineering communities. Traditional slope stability early warning methods usually rely on field investigation and static analysis, but such methods are often limited by insufficient real-time monitoring data, untimely data collection and other factors, making it difficult to ensure the accuracy and timeliness of early warning, especially in complex terrain and dynamic environments, the prediction effect is more limited.

[0003] The current common prediction methods mainly fall into three categories: 1) numerical simulation methods (such as finite element, finite difference), which can combine geomechanical models but are highly dependent on rock and soil parameters and have high computational costs. For example, solving control equations using numerical tools such as finite element method (FEM) and finite difference method (FDM) can reflect the physical and mechanical mechanisms of slopes, but these methods are highly dependent on material parameters and boundary conditions. In actual engineering, due to the difficulty in obtaining parameters or the existence of uncertainties, the prediction effect is often limited. 2) Empirical formula and statistical method, simple and easy to implement, but lack of description ability for complex nonlinear mechanism; common ones are time series regression, grey prediction model, etc., which have the advantages of simple implementation and high computational efficiency, but these methods ignore the mechanical mechanism of slopes and are difficult to adapt to the displacement evolution process under complex working conditions. 3) In recent years, neural networks, support vector machines, LSTM / GRU, etc. have been widely used in slope displacement prediction, which can automatically learn the nonlinear laws in monitoring data. However, this kind of method is completely data-driven and lacks physical constraints, which may cause overfitting and lack of stability in extrapolation prediction.

[0004] To make up for the above shortcomings, this paper attempts to fuse physical constraints and data-driven methods. Among them, PINN can reduce the dependence on a large amount of data by introducing physical knowledge such as control equations and boundary conditions into the neural network; and GRU is an effective tool for time series modeling, which is suitable for processing long sequence and non-stationary data in slope displacement monitoring. Therefore, the method combining PINN and GRU is expected to balance physical rationality and prediction accuracy, and to provide a new hybrid modeling framework for slope displacement prediction. SUMMARY

[0005] To solve the problems of strong parameter dependence, low automation and data overfitting in existing slope displacement prediction, the present application provides a slope displacement prediction method based on physical information neural network and gated recurrent unit, comprising the following steps:

[0006] S1. A monitoring data matrix of the target slope is established by multi-source data acquisition, the monitoring data matrix is preprocessed, and the preprocessed monitoring data matrix is time series divided by a sliding window algorithm to obtain a multi-time series monitoring data set;

[0007] S2. A gated recurrent unit model is constructed, the multi-time series monitoring data set is input, displacement time trend feature extraction is performed, and slope displacement time series feature data is obtained;

[0008] S3. A physical information neural network is constructed, the multi-time series monitoring data set is input, the slope mechanical control algorithm is used as a physical constraint, displacement physical feature extraction is performed, and slope displacement physical feature data is obtained;

[0009] S4. The slope displacement time series feature data and the slope displacement physical feature data are fused, a joint loss function model is constructed for collaborative optimization of the feature fusion, and a slope displacement prediction result is obtained.

[0010] Based on the above method, the present application also provides a slope displacement prediction system based on physical information neural network and gated recurrent unit, which is realized based on any one of the above methods, and comprises a data acquisition and preprocessing module for establishing a monitoring data matrix of a target slope by multi-source data acquisition, preprocessing the monitoring data matrix, and time series dividing the preprocessed monitoring data matrix by a sliding window algorithm to obtain a multi-time series monitoring data set;

[0011] A gated recurrent unit construction and displacement time series feature extraction module is used to construct a gated recurrent unit model, input the multi-time series monitoring data set, perform displacement time trend feature extraction, and obtain slope displacement time series feature data;

[0012] The physical information neural network construction and displacement physical feature extraction module is used for constructing a physical information neural network, taking a multi-time sequence monitoring data set as input, taking a slope mechanical control algorithm as a physical constraint, performing displacement physical feature extraction, and obtaining slope displacement physical feature data.

[0013] The feature fusion and prediction module is used for fusing the slope displacement time sequence feature data and the slope displacement physical feature data, constructing a joint loss function model to cooperatively optimize the feature fusion, and obtaining a slope displacement prediction result.

[0014] The present application has the beneficial effect that by combining the physical information neural network (PINN) with the gated recurrent unit (GRU), the organic fusion of data driving and physical constraints is realized in the slope displacement prediction, the slope mechanical control equation and the boundary condition are embedded into the neural network training process through the PINN, and the physical rationality of the prediction result can be maintained even in the case of insufficient monitoring data or noise; at the same time, the GRU effectively captures the nonlinear time sequence features of the slope displacement under complex working conditions such as rainfall, water level and construction, and improves the modeling ability of the dynamic evolution process. Through the coupling of the two, the present application not only improves the accuracy and stability of the slope displacement prediction, but also enhances the generalization ability and engineering interpretability of the model under different working conditions, realizes real-time early warning of the slope instability risk, and thus provides more reliable technical support for geological disaster prevention. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a slope displacement prediction method flowchart based on a physical information neural network and a gated recurrent unit according to an embodiment of the present application.

[0016] Figure 2 is a slope displacement prediction system structure diagram based on a physical information neural network and a gated recurrent unit according to an embodiment of the present application.

[0017] Figure 3 is a slope displacement prediction result diagram according to an embodiment of the present application.

[0018] Figure 4 is a terminal device structure diagram of a slope displacement prediction method based on a physical information neural network and a gated recurrent unit according to an embodiment of the present application.

[0019] Figure 5 is a computer readable storage medium structure diagram of a slope displacement prediction method based on a physical information neural network and a gated recurrent unit according to an embodiment of the present application.

[0020] In the figure, 200 - terminal device, 210 - memory, 211 - RAM, 212 - cache memory, 213 - ROM, 214 - programs / utilities, 215 - program modules, 220 - processor, 230 - bus, 240 - external device, 250 - I / O interface, 260 - network adapter, 300 - program product. DETAILED DESCRIPTION

[0021] For the purposes of the present application, the technical solutions and advantages will be clearer. The present application will be further described in conjunction with the accompanying drawings and examples. It should be understood that the specific examples described herein are only intended to explain the present application and are not intended to limit the present application. That is, the described examples are only a part of the examples of the present application, but not all examples. The components of the embodiments of the present application generally described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] Furthermore, the term "comprise", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or mechanical device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or mechanical device. Without more limitations, the element defined by the sentence "comprises a" does not exclude the presence of additional identical elements in the process, method, article or mechanical device comprising the element.

[0024] Example 1

[0025] As shown in Figure 1 Example 1 of the present application provides a slope displacement prediction method based on physical information neural network and gated recurrent unit, comprising the following steps:

[0026] S1. A monitoring data matrix of the target slope is established by multi-source data acquisition, the monitoring data matrix is preprocessed, and the preprocessed monitoring data matrix is time-sequentially divided by a sliding window algorithm to obtain a multi-time sequence monitoring data set;

[0027] S2. A gated recurrent unit model is constructed, the multi-time sequence monitoring data set is input, displacement time trend feature extraction is performed, and slope displacement time sequence feature data is obtained;

[0028] S3. Construct a physical information neural network, take a multi-time series monitoring dataset as input, and use the slope mechanical control algorithm as physical constraint to extract displacement physical features and obtain slope displacement physical feature data;

[0029] S4. The temporal characteristic data of slope displacement and the physical characteristic data of slope displacement are fused together, and a joint loss function model is constructed to optimize the feature fusion in a coordinated manner to obtain the slope displacement prediction results.

[0030] Furthermore, step S1 includes the following sub-steps:

[0031] S101. By acquiring multi-source data, obtain slope displacement curve data, spatial coordinate data and physical parameter data of the target slope area respectively, and establish a monitoring data matrix;

[0032] S102. Based on the Hampshire filtering algorithm, noise removal and time-series partitioning are performed on the monitoring data matrix to obtain the denoised and time-series partitioned monitoring data matrix.

[0033] S103. Based on the neighbor mean imputation algorithm, missing values ​​are processed in the monitoring data matrix after denoising and time series division to obtain a multi-time series monitoring dataset.

[0034] Specifically, the implementation flow of each sub-step in the above embodiments is as follows:

[0035] The collected raw data (displacement curves, spatial information, and corresponding physical parameter data) undergoes preprocessing, including denoising, handling missing values, and time alignment. Since GNSS instruments can introduce significant noise into the data during acquisition due to external factors, this noise needs to be removed before inputting it into the network for training. Here, the Hampel filtering algorithm is used for denoising and time-series partitioning to ensure effective removal of outliers. Specifically, the parameters are first set as follows: window length N is an odd number, such as 5 or 7, and threshold coefficient k is 3. Then, a sliding window algorithm is used to process each data point. Extract all data within the window and create an initial dataset. ,in For the i-th data point, calculate the median Median(x) of the data points in the current window based on the initial dataset, and calculate the absolute deviation | -Median(x) | (j∈window); Calculate the MAD value: MAD = 1.4826 Median(∣ -Median(x)∣); Determine outliers: if | -Median(x)∣>k MAD, then the outliers are marked; replace the outliers: replace with median or linear interpolation correction, obtain the denoised and time-sequential division monitoring data matrix, then perform missing value processing on the denoised and time-sequential division monitoring data matrix, in order to better train sequence features of the model, the missing values need to be processed, here the adjacent mean filling method is adopted, that is, the entire sequence is traversed in a loop, if the current node has missing values, the mean of the previous node and the next node is used as the value of the current node, and a multi-time sequence monitoring data set is obtained.

[0036] Specifically, the sliding window (Sliding Window) is a commonly used time series processing method, and the core idea is to define a fixed length window (window size) on the original time series, and the window slides forward with time; each time the window is slid, the subsequence covered by the window is taken as input to predict one or more future values after the window. This method can capture the local time sequence characteristics of the data, and can also cut the long time series into supervised learning samples that can be processed by the training model.

[0037] Further, the gated recurrent unit model in step S2 includes an input layer, a gated recurrent unit layer, a long short-term memory network layer, an attention mechanism, a fully connected layer, a regularization layer, and an output layer.

[0038] Further, step S2 includes the following sub-steps:

[0039] S201. Based on the input layer, the multi-time sequence monitoring data set is taken as input, and the edge slope displacement time sequence training sample set is generated through the input layer of the gated recurrent unit model;

[0040] S202. Based on the gated recurrent unit layer, the edge slope displacement time sequence training sample set is input into the gated recurrent unit layer, and the short-term time sequence dynamic feature of the edge slope displacement time sequence training sample set is extracted;

[0041] S203. Based on the long short-term memory network layer, the extracted dynamic features are input into the long short-term memory network layer, and the long-term dependence relationship and nonlinear dynamic characteristics are captured;

[0042] S204. Based on the attention mechanism, the importance of different time step features is adaptively allocated according to the output obtained in step S203, and an enhanced time sequence feature representation is obtained;

[0043] S205. Based on the fully connected layer, the enhanced features are input into the multi-layer fully connected layer for high-order feature extraction, and a regularization layer is constructed through a regularization mechanism to perform regularization processing on the extracted high-order features;

[0044] S206. Based on the output layer, the multi-dimensional displacement prediction sequence of the future time period is output, and the edge slope displacement time sequence feature data is obtained.

[0045] Specifically, the specific implementation process of each sub-step of the above embodiment is as follows:

[0046] 1) Input layer design: define the input structure of the network, and the shape of the input tensor is (36, 3), where 36 represents the time step, corresponding to 36 continuous time steps of monitoring data, and 3 represents the multi-dimensional input dimension, which represents the displacement of GNSS monitoring data x, y, z in three directions. The input is used for modeling the prediction of the future displacement sequence;

[0047] 2) Gated recurrent unit layer construction: a layer of gated recurrent unit (GRU) network is added after the input layer, the number of neurons is set to 32, and the hyperbolic tangent function (tanh) is used as the activation function to capture the short-term dynamic characteristics in the time series.

[0048] 3) Long short-term memory network layer construction: a long short-term memory (LSTM) network layer is introduced after the gated recurrent unit layer, the number of neurons is set to 96, and the tanh function and the Sigmoid function are used as the activation function, which is used to further model the long-term dependence relationship and nonlinear dynamic characteristics in the time series;

[0049] 4) Attention mechanism design: an attention mechanism (Attention) is introduced based on the output of the long short-term memory network layer, and a trainable parameter is set to weight the input features and the parameter, and the tanh activation function is used to obtain the score of each node., and then converted into a weight distribution through a flexible maximum transfer function (softmax), so as to realize the adaptive adjustment of the importance of the features, and obtain the enhanced time sequence feature representation;

[0050] 5) Additional layer design: after the attention mechanism output, two layers of dense layers are set, each containing 128 neurons, which are used to further extract high-order features; at the same time, a dropout regularization layer is introduced, and the dropout rate coefficient is set to 0.25, in order to prevent model overfitting and improve the generalization ability;

[0051] 6) Output layer design: the final output layer is a fully connected layer, and the output shape is set to (12, 3), which corresponds to the three-dimensional sequence data of the predicted future 12 time steps, and the dimension represents the displacement of GNSS monitoring data x, y, z in three directions, completing the prediction task of the slope displacement trend.

[0052] Further, the physical information neural network of step S3 includes an input layer, a hidden layer, and an output layer.

[0053] Further, step S3 includes the following sub-steps:

[0054] S301. Based on the input layer, a multi-time sequence monitoring data set is taken as input to establish a slope displacement physical training sample set in a space-time domain;

[0055] S302. Based on the hidden layer, a multi-layer perception machine structure is used to extract nonlinear features of the slope displacement physical training sample set and maintain continuous differentiability;

[0056] S303. The slope mechanical control algorithm is taken as a physical condition constraint, the displacement boundary condition and the stress boundary condition are taken as boundary condition constraints, and a loss function is constructed according to a physical residual error, a boundary constraint residual error and a data fitting error to optimize parameters of the physical information neural network;

[0057] S304. Based on the output layer, a space-time consistent displacement prediction result is output to obtain slope displacement physical feature data.

[0058] Specifically, the specific implementation process of each sub-step of the above embodiment is as follows:

[0059] 1) Input layer design: define the input tensor structure of the physical information neural network (PINN), the input is a four-tuple (x, y, z, t), where (x, y, z) represents the spatial coordinates, t represents the time step, and the input is used to predict the slope displacement field and related physical quantities in the space-time domain ;

[0060] 2) Hidden layer construction: a multi-layer perception machine (MLP) structure is used, 6-10 hidden layers are set, the number of neurons in each layer is 64-256, and the tanh activation function is selected to ensure that the network has good nonlinear fitting ability and continuous differentiability;

[0061] 3) Physical constraint embedding: in the network training process, the slope mechanical control equation is embedded in the loss function, for example, the quasi-static equation is expressed as: ; ; ; wherein, represents a physical constraint residual error, represents a stress tensor, represents a constitutive relationship function that maps strain to stress, represents a strain tensor, represents the gradient of the displacement field, represents the transpose of the displacement gradient, represents the stress divergence, reflecting the internal stress balance; and represent the soil density and gravity acceleration vector respectively;

[0062] 4) Boundary condition design: Introduce boundary constraints in the loss function, including displacement boundary conditions and stress boundary conditions, to ensure that the model prediction results meet the physical consistency of the geometric domain boundary, which is expressed as: ; wherein, represents the boundary condition residual, represents the physical information neural network predicted slope displacement vector, represents the normalized observed displacement data, represents the boundary normal vector, a unit vector pointing to the boundary surface, used to convert stress to force on the boundary surface, represents the load vector on the boundary;

[0063] 5) Loss function definition: Comprehensive consideration of partial differential equation (PDE) residual, boundary condition residual, data fitting error and regularization term, define the total loss function, which is expressed as: ; wherein, represents the physical constraint residual, represents the boundary condition residual, represents the normalized observed displacement data, represents the network parameters, represents the physical constraint residual weight parameter, represents the boundary condition residual weight parameter, represents the data fitting residual weight parameter, represents the regularization weight parameter, the above weight parameters can be dynamically adjusted according to the training;

[0064] 6) Output layer design: The final output layer is a fully connected layer, which outputs the physical information neural network predicted displacement vector , and can optionally output auxiliary variables (such as stress field or material reduction parameters), thereby providing physically consistent solutions for slope displacement prediction and stability evaluation.

[0065] Further, step S4 includes the following sub-steps:

[0066] S401. Map the slope displacement time series feature data into the boundary conditions, external loads and material parameters in the physical information neural network varying with time;

[0067] S402. Take the displacement increment in the slope displacement time series feature data as the correction parameter to correct the slope displacement physical feature data;

[0068] S403. Construct a joint loss function containing physical residual, boundary condition residual, data fitting error and time series correction error, and adjust the gated recurrent unit model parameters, physical information neural network parameters and joint loss function weight coefficients until the joint loss function converges, output the final slope displacement prediction result.

[0069] Specifically, the specific implementation process of each sub-step of the above embodiment is as follows:

[0070] 1) Parameter-driven coupling: mapping the slope displacement time series feature data predicted by the gated recurrent unit model (GRU) into the time-varying boundary condition, external load or material parameter input of the physical information neural network (PINN). That is, the GRU output is taken as part of the PINN input, forming a time-dependent physical constraint prediction, which is expressed as: ; wherein is a placeholder, representing the input variable of the PINN network in the space-time domain (such as spatial coordinates x, y, z or other unspecified inputs, determined according to the actual prediction scenario), represents the slope displacement time series feature data, represents the time point of the current slope displacement time series feature data, in this way, the PINN can dynamically respond to external driving and material parameter changes under physical constraints;

[0071] 2) Residual correction coupling: the displacement increment output by the gated recurrent unit model (GRU) is used to correct the prediction results of the PINN to obtain the final predicted displacement, which is expressed as: ; wherein, represents the final fused slope displacement increment prediction sequence of the future H time steps, represents the displacement increment of the future H time steps predicted based on the physical information neural network (PINN), represents the displacement increment of the future H time steps predicted based on the gated recurrent unit model (GRU), this step can enhance the ability to capture complex nonlinear time series behavior and improve prediction accuracy;

[0072] 3) Joint training and loss function: define the PINN-GRU joint optimization loss function, integrate the physical residual, boundary constraint residual, data fitting error and GRU correction error, which is expressed as: ; wherein, represents the joint optimization loss function, and represent adjustable weight coefficients, set to 0.5, 0.5, represents the loss function of the physical information neural network (PINN), represents the loss function of the gated recurrent unit model (GRU), which is specifically an L1 loss function (mean absolute error) for evaluating the absolute error between the predicted value and the true value; through joint optimization, the PINN and GRU work together in the training process to achieve physically consistent and high-precision prediction of slope displacement;

[0073] 4) The multi-time sequence monitoring data set is divided into 80% training data and 20% test data, the training data is normalized separately, and then the data is preprocessed by using a sliding window method, the window size is defined as 12, which can maximize the division of the training set, and the division result is shown in Table 1:

[0074] Table 1: Division result of multi-time sequence monitoring data set

[0075]

[0076] The first 36 data in the data set are divided as input, and the 37th to the 48th data are divided as output test data, then each time one data is iterated forward, until the data is completely divided, the model parameters need to be continuously optimized during training, until the final prediction effect is obtained, and the current trained model is the final result.

[0077] The final model test result is shown in Figure 3 , three directions are predicted 12 data each time, until the complete curve is predicted, it can be seen that the predicted curve is basically fitted with the actual curve. In addition, the average error and the root mean square error of each node are 1.345 and 1.567 respectively, the unit is mm, and both are less than 2mm, so it can be seen that the method has a significant effect on the slope early warning ability.

[0078] Embodiment 2

[0079] As shown in Figure 2 , on the basis of embodiment 1, the embodiment 2 of the application proposes a slope displacement prediction system based on a physical information neural network and a gating recurrent unit, which is realized by a slope displacement prediction method based on a physical information neural network and a gating recurrent unit.

[0080] Specifically, the system comprises a data acquisition and preprocessing module, which is used for establishing a monitoring data matrix of a target slope through multi-source data acquisition, preprocessing the monitoring data matrix, and dividing the preprocessed monitoring data matrix in time sequence through a sliding window algorithm to obtain a multi-time sequence monitoring data set;

[0081] A gating recurrent unit construction and displacement time sequence feature extraction module is used for constructing a gating recurrent unit model, taking the multi-time sequence monitoring data set as input, extracting displacement time trend features, and obtaining slope displacement time sequence feature data;

[0082] A physical information neural network construction and displacement physical feature extraction module is used for constructing a physical information neural network, taking the multi-time sequence monitoring data set as input, taking a slope mechanics control algorithm as a physical constraint, extracting displacement physical features, and obtaining slope displacement physical feature data;

[0083] The feature fusion and prediction module is configured to fuse the time series feature data of the slope displacement with the physical feature data of the slope displacement, construct a joint loss function model to cooperatively optimize the fused features, and obtain a slope displacement prediction result.

[0084] Specifically, the workflow of the system is as follows:

[0085] First, the data acquisition and preprocessing module is used to collect multi-source data of the target slope, obtain displacement curve data, spatial coordinate data and related physical parameters of the slope, and arrange the raw data into a monitoring data matrix. Then, the monitoring data matrix is preprocessed, including noise removal and missing value filling, and the processed data is divided into time series using a sliding window algorithm to obtain a multi-time series monitoring data set, which provides standardized input for subsequent model training. Next, the multi-time series monitoring data set is input into the gated recurrent unit model. First, the time series training samples are generated through the input layer, then the short-term dynamic features are extracted in the gated recurrent unit layer, and then the long-term dependency and nonlinear dynamic characteristics are captured through the long short-term memory network layer. Subsequently, the attention mechanism is used to adaptively assign the importance of each time step feature, and the high-order features are further extracted through the full connection layer and the regularization mechanism. Finally, the multi-dimensional displacement prediction sequence of the future time period is output, and the time series feature data of the slope displacement is obtained. At the same time, the multi-time series monitoring data set is also input into the physical information neural network. First, the slope displacement physical training sample set is constructed in the input layer, then the nonlinear features are extracted through the multilayer perceptron hidden layer while maintaining the continuity and differentiability, and at the same time, the slope mechanical control algorithm is embedded as a physical constraint in the training process. The loss function is constructed in combination with the displacement and stress boundary conditions to optimize the network parameters, and finally the spatiotemporal consistent displacement prediction result is generated in the output layer, and the physical feature data of the slope displacement is obtained. Finally, in the feature fusion and prediction module, the time series features output by the gated recurrent unit and the physical features output by the physical information neural network are fused, and the fused features are cooperatively optimized by constructing a joint loss function. The parameter-driven and residual correction coupling are realized, the external load and boundary condition changes are dynamically responded, and the model parameters are constantly updated, so that the model meets the requirements of physical constraints and data-driven at the same time, thereby outputting the final slope displacement prediction result, providing a reliable basis for slope stability analysis and early warning.

[0086] Embodiment 3

[0087] As shown in Figure 4 Based on embodiment 1, the terminal device of the slope displacement prediction method based on the physical information neural network and the gated recurrent unit is proposed in embodiment 3. The terminal device 200 includes at least one memory 210, at least one processor 220, and a bus 230 connecting different platform systems.

[0088] The memory 210 can include a readable medium in the form of volatile memory, such as RAM 211 and / or cache memory 212, and can further include ROM 213.

[0089] The memory 210 further stores a computer program, which can be executed by the processor 220, so that the processor 220 executes any one of the above-mentioned slope displacement prediction methods based on a physical information neural network and a gating recurrent unit in the embodiments of the present application. The specific implementation manners and the achieved technical effects are consistent with those described in the embodiments of the above-mentioned method and some contents will not be described herein. The memory 210 can further include programs / utilities 214 having a set of (at least one) program modules 215, and such program modules include but are not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof can include the implementation of a network environment.

[0090] Correspondingly, the processor 220 can execute the above-mentioned computer program and can execute the programs / utilities 214.

[0091] The bus 230 can represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures.

[0092] The terminal device 200 can also communicate with one or more external devices 240, such as a keyboard, a pointing device, a Bluetooth device, or the like, and can also communicate with one or more devices that enable interaction with the terminal device 200, such that the terminal device 200 can communicate with one or more other computing devices using any device (such as a router, a modem, and the like) that enables communication between the terminal device 200 and the one or more other computing devices. Such communication can be carried out through the I / O interface 250. Furthermore, the terminal device 200 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 260. The network adapter 260 can communicate with other modules of the terminal device 200 through the bus 230. It should be understood that although not shown in the figure, other hardware and / or software modules can be used in conjunction with the terminal device 200, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0093] Embodiment 4

[0094] As Figure 5As shown, on the basis of Embodiment 1, the present embodiment proposes a computer readable storage medium of a slope displacement prediction method based on a physical information neural network and a gated recurrent unit, and the computer readable storage medium stores instructions, and the instructions are executed by a processor to implement any one of the above-mentioned slope displacement prediction methods based on a physical information neural network and a gated recurrent unit. The specific implementation manner and the technical effects achieved are consistent with those described in the embodiments of the above-mentioned method, and part of the content will not be repeated.

[0095] The program product 300 for implementing the above-mentioned method provided by the present embodiment can adopt a portable compact disc read-only memory (CD-ROM) and include a program code, and can run on a terminal device such as a personal computer. However, the program product 300 of the present embodiment is not limited to this, and in the present embodiment, the readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or apparatus. The program product 300 can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0096] Computer-readable storage media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. Program code for performing operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on a user computing device, partially on a user device, as a standalone software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing devices can be connected to user computing devices via any type of network, including local area networks (LANs) or wide area networks (WANs), or they can be connected to external computing devices (e.g., via the Internet using an Internet service provider).

[0097] This invention is described from the perspectives of its intended use, effectiveness, progress, and novelty. Its practical and progressive features meet the functional enhancement and use requirements emphasized by the Patent Law. The above description and drawings are merely preferred embodiments of this application and are not intended to limit this application. Therefore, all structures, devices, features, etc., that are similar to or identical to those of this application, i.e., all equivalent substitutions or modifications made in accordance with the scope of this patent application, shall fall within the scope of protection of this patent application.

[0098] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A slope displacement prediction method based on physical information neural network and gated recurrent unit, characterized in that, Includes the following steps: S1. Establish a monitoring data matrix for the target slope by collecting data from multiple sources, preprocess the monitoring data matrix, and divide the preprocessed monitoring data matrix into time series using a sliding window algorithm to obtain a multi-time series monitoring dataset; S2. Construct a gated cyclic unit model, take the multi-time series monitoring dataset as input, extract displacement time trend features, and obtain slope displacement time series feature data; S3. Construct a physical information neural network, take a multi-time series monitoring dataset as input, and use the slope mechanical control algorithm as physical constraint to extract displacement physical features and obtain slope displacement physical feature data; S4. The temporal characteristic data of slope displacement and the physical characteristic data of slope displacement are fused together, and a joint loss function model is constructed to optimize the feature fusion in a coordinated manner to obtain the slope displacement prediction results. Step S4 includes the following sub-steps: S401. Map the temporal characteristic data of slope displacement to the time-varying boundary conditions, external loads and material parameters in the physical information neural network; S402. Use the displacement increment in the slope displacement time series characteristic data as a correction parameter to perform data correction on the slope displacement physical characteristic data; S403. Construct a joint loss function that includes physical residuals, boundary condition residuals, data fitting errors, and time-series correction errors. Adjust the parameters of the gated cyclic unit model, the parameters of the physical information neural network, and the weight coefficients of the joint loss function until the joint loss function converges, and output the final slope displacement prediction result.

2. The slope displacement prediction method based on physical information neural network and gated recurrent unit according to claim 1, characterized in that, Step S1 includes the following sub-steps: S101. By acquiring multi-source data, obtain slope displacement curve data, spatial coordinate data and physical parameter data of the target slope area respectively, and establish a monitoring data matrix; S102. Based on the Hampshire filtering algorithm, noise removal and time-series partitioning are performed on the monitoring data matrix to obtain the denoised and time-series partitioned monitoring data matrix. S103. Based on the neighbor mean imputation algorithm, missing values ​​are processed in the monitoring data matrix after denoising and time series division to obtain a multi-time series monitoring dataset.

3. The slope displacement prediction method based on physical information neural network and gated recurrent unit according to claim 2, characterized in that, The gated recurrent unit model described in step S2 includes an input layer, a gated recurrent unit layer, a long short-term memory network layer, an attention mechanism, a fully connected layer, a regularization layer, and an output layer.

4. The slope displacement prediction method based on physical information neural network and gated recurrent unit according to claim 3, characterized in that, Step S2 includes the following sub-steps: S201. Based on the input layer, multiple time-series monitoring datasets are used as input, and a time-series training sample set of slope displacement is generated through the input layer of the gated recurrent unit model; S202. Based on the gated recurrent unit layer, the slope displacement time series training sample set is input into the gated recurrent unit layer to extract short-term time series dynamic features from the slope displacement time series training sample set; S203. Based on the Long Short-Term Memory (LSTM) network layer, the extracted dynamic features are input into the LSM network layer to capture long-term dependencies and nonlinear dynamic characteristics; S204. Based on the attention mechanism, according to the output obtained in step S203, the importance of features at different time steps is adaptively assigned to obtain an enhanced temporal feature representation; S205. Based on the fully connected layer, the enhanced features are input into multiple fully connected layers for high-order feature extraction, and a regularization layer is constructed through a regularization mechanism to perform regularization processing on the extracted high-order features; S206. Based on the output layer, output the multidimensional displacement prediction sequence for future time periods to obtain the temporal characteristic data of slope displacement.

5. The slope displacement prediction method based on physical information neural network and gated recurrent unit according to claim 2, characterized in that, The physical information neural network described in step S3 includes an input layer, a hidden layer, and an output layer.

6. The slope displacement prediction method based on physical information neural network and gated recurrent unit according to claim 5, characterized in that, Step S3 includes the following sub-steps: S301. Based on the input layer, a physical training sample set for slope displacement is established in the spatial and temporal domain by taking multiple time-series monitoring datasets as input. S302. Based on the hidden layer, the nonlinear features of the slope displacement physical training sample set are extracted through a multilayer perceptron structure while maintaining continuous differentiability. S303. Using the slope mechanics control algorithm as the physical condition constraint, displacement boundary condition and stress boundary condition as the boundary condition constraint, and constructing a loss function based on the physical residual, boundary constraint residual and data fitting error, the parameters of the physical information neural network are optimized. S304. Based on the output layer, output the spatiotemporally consistent displacement prediction results to obtain the physical characteristic data of slope displacement.

7. A slope displacement prediction system based on a physical information neural network and a gated loop unit, wherein the system is implemented based on the slope displacement prediction method based on a physical information neural network and a gated loop unit as described in any one of claims 1-6, characterized in that, include: The data acquisition and preprocessing module is used to establish a monitoring data matrix of the target slope through multi-source data acquisition, preprocess the monitoring data matrix, and divide the preprocessed monitoring data matrix into time series using a sliding window algorithm to obtain a multi-time series monitoring dataset. The gated cyclic unit construction and displacement time series feature extraction module is used to construct the gated cyclic unit model, take the multi-time series monitoring dataset as input, extract the displacement time trend features, and obtain the slope displacement time series feature data; The module for constructing a physical information neural network and extracting physical displacement features is used to construct a physical information neural network. It takes a multi-time series monitoring dataset as input and a slope mechanical control algorithm as physical constraint to extract physical displacement features and obtain slope displacement physical feature data. The feature fusion and prediction module is used to fuse the temporal feature data and physical feature data of slope displacement, and to construct a joint loss function model to optimize the feature fusion and obtain the slope displacement prediction results.

Images