Landslide surge height prediction method and system based on gating circulation unit

Abstract

The invention relates to the technical field of landslide surge and data driving, and discloses a landslide surge height prediction method and system based on a gating circulation unit, and the method comprises the steps: collecting the related data of a large-scale physical model test of a landslide surge region, carrying out the data standardization and normalization preprocessing of the feature data, and obtaining the data of the large-scale physical model test; the preprocessed feature data serve as a gated cycle unit GRU model to be input, and the surge height is obtained; building a gated loop unit GRU network in a Python compiling environment by using tensor flow; training a GRU (Gated Recirculation Unit) network model, and training the GRU network model by using the historical landslide data to adjust model parameters; and deploying the trained GRU model into an actual landslide surge prediction system for real-time monitoring. The system comprises a data preprocessing module, a model building module and a model training module. According to the invention, an efficient and accurate prediction system is formed.

Description

Technical Field

[0001] The invention relates to the field of landslide surge and data driven technology, and in particular to a landslide surge height prediction method and system based on a gated circulation unit. Background Art

[0002] In the field of landslide surge disaster prevention and control, the movement of soil and rock mass during landslide is a complex dynamic process, which leads to rapid changes in topography and then causes surges in the water body in the landslide area or its downstream. The accurate prediction of the surge height is of great significance for the early warning and prevention of landslide disasters. With the development of deep learning technology, gated recurrent unit (GRU) as an effective time series data processing method has shown broad application prospects in the prediction of landslide surge height. GRU can capture the time series characteristics of the landslide process and accurately predict the surge height caused by the landslide by learning the laws of historical landslide data. Compared with traditional methods, the landslide surge height prediction method based on GRU has the following advantages: first, the GRU model can effectively handle long-term dependencies and capture the dynamic change characteristics of the landslide process; second, GRU is trained in combination with actual test data, which can better reflect the physical characteristics of the landslide process and improve the prediction accuracy of the model; third, the GRU model can automatically learn and optimize parameters, avoiding the problem of parameter overfitting in traditional methods.

[0003] Prior art 1, Chinese patent, application number 202410069549.X discloses a method and system for coupling the near-field and far-field propagation evolution of landslide surge waves, wherein the method steps include: using particle discrete elements and smooth particle fluid dynamics to simulate the near-field motion of granular landslides; using the simulated near-field motion as the initial boundary condition to calculate the far-field motion of granular landslides; based on drag force and buoyancy, completing the near-field and far-field propagation evolution of granular landslides during movement and water entry. Although the propagation path, speed and impact range of landslide surges can be more accurately predicted through detailed numerical simulation of key areas, the topography, geology, hydrological conditions, etc. can also be considered in detail to improve the authenticity of landslide surge simulation. By simulating different scenarios, the potential impact of landslide surges on coastal structures, residential areas and key infrastructure can be evaluated to provide hazard analysis for rescue and planning. However, the method itself lacks the combination of mechanical theory, which easily leads to overfitting of parameters or even parameters without physical meaning, making it difficult to analyze the landslide surge process and related risk prevention and control.

[0004] Prior art 2, Chinese patent, application number 202410014296.6 discloses a method for judging the type of surge waves induced by landslides in waters and predicting wave heights, including establishing a dynamic model of slope instability, obtaining the landslide velocity u(t) and landslide displacement s(t) processes, calculating dimensionless parameters Fr and Sz, distinguishing the surge wave type, and calling the oscillation wave / single wave formula to calculate the wave height Am of the corresponding wave type. Although the landslide surge waves are classified and distinguished, and wave height prediction methods are proposed for different surge waves, reliable technical support is provided for the determination of relevant disaster prevention and mitigation measures and the formulation of emergency plans. However, it mainly relies on numerical simulation or on-site monitoring data, and there are problems with insufficient calibration of dynamic parameters and uncertainty processing.

[0005] Prior art three, Chinese patent, application number 202311122865.0 discloses a random sampling surge exceedance probability calculation method, including: according to the target area, analyzing the range of landslide parameter values ​​in the target area, processing the landslide input parameters required for constructing the numerical model, generating random variables, performing Latin hypercube sampling according to the random variables, generating random samples, inputting the random samples into the numerical model, extracting the surge propagation height corresponding to the random variables, determining the surge propagation height threshold according to the affected body in the target area, and calculating the exceedance probability of the surge propagation height exceeding the surge propagation height threshold. Although Latin hypercube sampling is performed on the selected random variables to generate random samples, and then the random samples are processed by the model, the exceedance probability of the surge propagation distance exceeding the threshold height is calculated, the uncertainty of the landslide input parameters is quantified, and the accuracy of the landslide-surge disaster prediction is improved. However, its method is limited by many assumptions and boundary condition processing, and cannot reflect the actual situation of landslide surges.

[0006] At present, the existing methods of prior art 1, prior art 2 and prior art 3 lack the combination of mechanical theory, which easily leads to parameter overfitting or even parameters without physical meaning, making it difficult to analyze the landslide surge process and related risk prevention and control. They mainly rely on numerical simulation or field monitoring data, and have insufficient calibration of dynamic parameters and uncertainty processing. They are limited by many assumptions and boundary condition processing, and cannot reflect the real situation of landslide surge. Therefore, the present invention proposes a landslide surge height prediction method based on GRU, which combines actual test data and deep learning algorithms to significantly improve the accuracy and reliability of the prediction, and provides more effective technical support for the risk prevention and control of the landslide surge disaster chain. Summary of the invention

[0007] The main purpose of the present invention is to provide a landslide surge height prediction method and system based on a gated circulation unit to solve the problem that the calibration of dynamic parameters and their uncertainty processing are insufficient in the prior art, and they are limited by many assumptions and boundary conditions and cannot reflect the actual situation of landslide surges.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A landslide surge height prediction method based on a gated cyclic unit, the landslide surge height prediction method based on a gated cyclic unit comprising:

[0010] Collect relevant data from large-scale physical model tests in the landslide surge area, and perform data standardization and normalization preprocessing on the characteristic data. The preprocessed characteristic data is used as the input of the gated recurrent unit (GRU) model to obtain the surge height.

[0011] Use Zhang Liuxun in the Python compilation environment to build a gated recurrent unit GRU network; introduce update gates and reset gates to control the flow of information. The update gate determines how much information from the previous state should be retained in the current state, while the reset gate controls how much previous state information should be forgotten;

[0012] Train the gated recurrent unit (GRU) network model and use historical landslide data to train the gated recurrent unit (GRU) network model to adjust the model parameters; deploy the trained gated recurrent unit (GRU) model to the actual landslide surge prediction system for real-time monitoring.

[0013] As a further improvement of the present invention, the relevant data include landslide volume, sliding speed, slope angle and characteristic data for simulating similar material properties of real geotechnical materials.

[0014] As a further improvement of the present invention, a gated recurrent unit GRU network is constructed, including:

[0015] Design the input layer, GRU layer, fully connected layer and output layer of the gated recurrent unit GRU network;

[0016] The mean square error is used as the loss function of the regression task, and the learning rate is dynamically adjusted according to the gradient changes;

[0017] When new data is input, the weights are updated through the back-propagation algorithm, and the functions of the reset gate and update gate are adjusted through feedback when different environmental conditions change.

[0018] As a further improvement of the present invention, the input layer is used to input feature vectors, the GRU layer captures long-term and short-term dependencies in time series data through a gate mechanism, the fully connected layer maps the output of the GRU layer to a new feature space, and combines the time series information extracted from the GRU layer to generate a more advanced feature representation, and the output layer maps the output of the fully connected layer to a specific prediction result, the height of the landslide surge.

[0019] As a further improvement of the present invention, the input layer sets the input feature vector x t for:

[0020] x t =[x t,1 ,x t,2 ,…,x t,n ]

[0021] Where n represents the number of features, including landslide related parameters, landslide volume, rate, slope; environmental information, rainfall, soil moisture, temperature, and data of historical landslide events;

[0022] The input data needs to be organized as a 3D tensor:

[0023]

[0024] In the formula, m is the number of samples, T is the time step, and n is the number of features in each time step;

[0025] The reset gate r of the GRU layer t It is an important gating mechanism used to control how much historical information should be forgotten when calculating a new hidden state. Its calculation formula is:

[0026] r t =σ(W r ·[h t-1 ,x t ]+b r )

[0027] Where W r represents the weight matrix of the reset gate, h t-1 is the hidden state of the previous time step, σ(·) is the sigmoid activation function with an output range between [0,1], and b r is the bias term for the reset gate;

[0028] Update gate z t It is used to control the degree to which the previous hidden state retains the current state. The calculation formula is:

[0029] z t =σ(W z ·[h t-1 ,x t]+b z )

[0030] Where W z is the weight matrix of the update gate, b z is the bias term of the update gate;

[0031] Candidate hidden states It is generated when deciding to update the current hidden state, and its calculation formula is:

[0032]

[0033] In the formula, ⊙ represents element-by-element multiplication, W is the weight matrix of the candidate state, and b is the bias term of the candidate state;

[0034] Current hidden state h t The update formula is:

[0035]

[0036] Multiple GRU layers are stacked, and the GRU units in the subsequent layers not only obtain information from the previous layer, but also combine the hidden states of the previous layers in their calculations, forming a residual connection mechanism, that is:

[0037]

[0038] In the formula, Represents the current hidden state of the lth layer, which is calculated based on the hidden state of the previous layer and the hidden states of the first two layers. Represents the current hidden state of the l-1th layer, providing information from the previous layer. Represents the current hidden state of layer l-2, providing information from earlier layers;

[0039] The fully connected layer converts the output h of the GRU layer into t Input the fully connected layer and perform nonlinear mapping through the ReLU activation function:

[0040] h fc =ReLU(W fc ·h t +b fc )

[0041] In the formula, h fc represents the output of the fully connected layer, that is, the activation value after the output from the GRU layer to the fully connected layer, W fc represents the weight matrix of the fully connected layer, which determines the contribution of the input features to the output features, b fc The bias term of the fully connected layer is used to adjust the output; the ReLU activation function maps negative values ​​to zero, that is, ReLU(x) = max(0,x), introducing nonlinear characteristics;

[0042] The output layer maps the features to the predicted values ​​through full connection and uses a linear activation function:

[0043] y pred =W out ·h fc +b out

[0044] In the formula, y pred is the final predicted value of surge height; W out The weight matrix of the output layer is used to map the features of the fully connected layer to the output values, b out Bias term for the output layer.

[0045] As a further improvement of the present invention, the loss function:

[0046]

[0047] Where, L is the mean square error calculated by the loss function, which measures the difference between the surge height predicted by the model and the true value, N is the total number of samples used to calculate the average loss, The true surge height of the i-th sample, The model predicts the swell height for the i-th sample.

[0048] As a further improvement of the present invention, the reset gate and the update gate adopt a dynamic coupling strategy, including:

[0049] Calculate the interaction between the reset gate and the update gate. When calculating the reset gate at the current moment, consider the combined information of the previous hidden state and the current input. Introduce the update gate at the previous moment, and refer to both the current input and the previous hidden state to determine the amount of information retained at the current moment.

[0050] In the process of generating candidate hidden states, historical information is integrated. Through comprehensive analysis of historical information and new inputs, the candidate states reflect the characteristics of the current moment.

[0051] The hidden state update process and its dynamic coupling, the update gate is responsible for determining how much information from the previous hidden state should be retained in the current hidden state.

[0052] As a further improvement of the present invention, the generation process of the candidate hidden state includes the interaction of a reset gate and an update gate, and the weight of the reset gate affects the contribution of the previous hidden state to the candidate hidden state.

[0053] As a further improvement of the present invention, through a dynamic coupling mechanism, the reset gate and the update gate jointly determine the flow and storage of information; when updating the hidden state at the current moment, the reset gate determines the extent to which the previous state needs to be forgotten.

[0054] To achieve the above object, the present invention also provides the following technical solutions:

[0055] A landslide surge height prediction system based on a gated cyclic unit is applied to the landslide surge height prediction method based on a gated cyclic unit. The landslide surge height prediction system based on a gated cyclic unit includes:

[0056] The data preprocessing module is used to collect relevant data of large-scale physical model tests in the landslide surge area, and perform data standardization and normalization preprocessing on the characteristic data. The preprocessed characteristic data is used as the input of the gated recurrent unit GRU model to obtain the surge height;

[0057] Model building module, used to build a gated recurrent unit (GRU) network using tensorflow in Python compilation environment; introduce update gate and reset gate to control the flow of information. The update gate determines how much information of the previous state should be retained in the current state, while the reset gate controls how much information of the previous state should be forgotten;

[0058] The model training module is used to train the gated recurrent unit (GRU) network model. The historical landslide data is used to train the gated recurrent unit (GRU) network model to adjust the model parameters. The trained gated recurrent unit (GRU) model is deployed to the actual landslide surge prediction system for real-time monitoring.

[0059] The present invention solves the problems of insufficient feature extraction and low model prediction accuracy in the prior art when predicting the height of landslide surge. The implementation steps are: obtaining feature data related to landslide surge; using TensorFlow to build a GRU network in a Python compilation environment; controlling the flow of information by introducing an update gate and a reset gate. The update gate determines how much information of the previous state should be retained in the current state, while the reset gate controls how much previous state information should be forgotten; training the GRU network and using historical landslide data for model training; using the trained GRU network to predict the height of landslide surge. The present invention constructs a landslide surge height prediction model based on GRU, which can effectively capture the long-term and short-term time dependencies in the landslide surge process and improve the prediction accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the steps of an embodiment of a method for predicting landslide surge height based on a gated circulation unit of the present invention;

[0061] Figure 2 It is a schematic diagram of an embodiment of a method for predicting landslide surge height based on a gated circulation unit of the present invention;

[0062] Figure 3 A schematic diagram of a gated circulation unit of an embodiment of a method for predicting landslide surge height based on a gated circulation unit according to the present invention;

[0063] Figure 4 It is a schematic flow chart of the steps of preprocessing the characteristic data by standardizing and normalizing the characteristic data in one embodiment of the landslide surge height prediction method based on the gated cyclic unit of the present invention;

[0064] Figure 5 A schematic flow chart of the steps of building a gated recurrent unit (GRU) network for an embodiment of a landslide surge height prediction method based on a gated recurrent unit according to the present invention;

[0065] Figure 6 This is a schematic diagram of the steps of a gated recurrent unit (GRU) network model of an embodiment of a landslide surge height prediction method based on a gated recurrent unit of the present invention;

[0066] Figure 7 It is a functional module schematic diagram of an embodiment of a landslide surge height prediction system based on a gated circulation unit of the present invention;

[0067] Figure 8 It is a structural schematic diagram of an embodiment of an electronic device of the present invention;

[0068] Fig. 9 A schematic diagram of the structure of a storage medium according to an embodiment of the present invention;

[0069] Fig.10 It is a comparison chart of the physical model test data of the present invention and the predicted value of the gated cycle unit method;

[0070] Fig.11 This is a graph of K-fold cross validation results of the present invention (K=5). DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] The terms "first", "second" and "third" in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second" and "third" can explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship, movement, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication also changes accordingly. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0073] Reference to an "embodiment" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0074] like Figure 1 As shown, this embodiment provides an embodiment of a landslide surge height prediction method based on a gated cyclic unit. In this embodiment, the landslide surge height prediction method based on a gated cyclic unit specifically includes the following steps:

[0075] Step S1: collecting relevant data of large-scale physical model tests in the landslide surge area, the relevant data including characteristic data such as landslide volume, sliding speed, slope angle and similar material properties used to simulate real geotechnical materials, and preprocessing the characteristic data by data standardization and normalization. The preprocessed characteristic data is used as the input of the gated recurrent unit GRU model to obtain the surge height;

[0076] Step S2: Use TensorFlow in the Python compilation environment to build a gated recurrent unit (GRU) network; introduce update gates and reset gates to control the flow of information. The update gate determines how much information of the previous state should be retained in the current state, while the reset gate controls how much information of the previous state should be forgotten.

[0077] Step S3: training the gated recurrent unit (GRU) network model, using historical landslide data to train the gated recurrent unit (GRU) network model to adjust model parameters; deploying the trained gated recurrent unit (GRU) model to an actual landslide surge prediction system for real-time monitoring.

[0078] Preferably, in step S1 of this embodiment, data collection and preprocessing focus on collecting relevant characteristic data in the landslide surge area, including landslide volume, sliding speed, slope angle and similar material properties; the collected characteristic data are standardized and normalized to eliminate the influence of different characteristic dimensions and ranges, and ensure the stability and accuracy of model training. Significance achieved: By preprocessing the characteristic data, the quality of data information input to the GRU model is ensured to be higher, which helps to improve the prediction accuracy of the model; standardization and normalization can speed up the convergence of the model, reduce training time, and avoid the problem of gradient disappearance or explosion caused by eigenvalue differences. Step S2 builds a GRU network, uses the TensorFlow framework to build a GRU network in a Python environment, and controls the flow of information by introducing update gates and reset gates. This enables the model to dynamically adjust the state according to historical information; the update gate can decide how much information of the previous state to retain according to the current input data, and the reset gate can effectively clear old information. Significance achieved: The introduction of the gating mechanism enables the GRU network to better capture the long-term and short-term dependencies in time series data. This is an improvement on the traditional RNN, making the network more efficient when processing time series data; through dynamic information management, the model can adapt to changes in input data more flexibly, thereby improving the prediction ability of complex systems (such as landslide surges). Step S3 trains the network and real-time monitoring, uses historical landslide data to train the GRU model, and adjusts the model parameters to optimize the prediction effect; deploys the trained GRU model to the actual landslide surge prediction system for real-time data monitoring and prediction. Significance achieved: Through the training process, the model can learn the complex relationship between the landslide surge height and historical data, so as to provide more accurate results in actual predictions; real-time monitoring and prediction can provide timely information to relevant agencies, help make faster decisions, and reduce the safety risks that may be caused by landslide surges. In addition, it can also help optimize the emergency response mechanism to ensure people’s safety and property protection (for specific principles, please refer to the attached Figure 2 , gated recurrent unit, see attached Figure 3 ).

[0079] In summary, this embodiment forms an efficient and accurate prediction system through steps such as data collection and preprocessing, network construction, model training and real-time application; it helps to build an effective natural disaster prediction and early warning system under complex environmental conditions.

[0080] The landslide surge height prediction method based on the gated cyclic unit in this embodiment has the following obvious advantages: (1) It processes the monitoring data of the landslide surge physical model test, combines the landslide surge control factors and the deep learning algorithm to perform a dynamic evaluation of the surge height, and the method can effectively handle long-term dependencies, and can maintain efficient operation even in the face of a large amount of monitoring data. (2) It can establish a deep learning model based on the deep learning algorithm and actual test data, significantly improving the accuracy of the surge height prediction. (3) It combines the deep learning theory and the mathematical model of the actual large-scale physical model test data to achieve accurate prediction of the landslide surge height, forming a new landslide disaster prevention and control evaluation method, and providing new ideas for related projects. This embodiment can quantitatively analyze the changes in mechanical parameters during the landslide process, deepen the understanding of the landslide surge process in the project, and achieve better prediction results; compared with the existing numerical simulation and simple statistical methods, it combines deep learning and mechanical mechanisms, realizes the dual innovation of data-driven and theory-driven, significantly improves the accuracy and reliability of landslide surge height prediction, and provides more reliable and efficient technical support for landslide disaster prevention and control.

[0081] Furthermore, if Figure 4 As shown, the process of preprocessing the feature data for data standardization and normalization in step S1 specifically includes the following steps:

[0082] Step S11: Calculate the mean and standard deviation of each feature data point by point, and use the statistical information of the mean and standard deviation to readjust each data point so that the feature data can be compared on the same scale; for the value range of each feature data, scale the feature data to the interval [0,1], find the maximum and minimum values ​​of each feature, and adjust the feature data by linear mapping;

[0083] Step S12: Obtain the normalized and adjusted feature data, sort all feature data according to timestamps, and use a dynamic segmentation strategy to divide the feature data into a training set, a validation set, and a test set, using 70% of the feature data for the training set, 15% of the feature data for the validation set, and 15% of the feature data for the test set;

[0084] Step S13: From the starting position of the feature data, place the window at the current starting point, and collect 10 data points in the window as a sample; move the window 1 data point each time, and repeat the above steps until enough samples are generated; for each sample generated by the sliding window, determine the corresponding label, that is, the surge height at the next moment.

[0085] Preferably, in step S11 of this embodiment, the feature data is standardized and normalized by calculating the mean and standard deviation of each feature and adjusting the data using these statistical information so that the feature data can be compared on the same scale, eliminating the deviation caused by different value ranges between different features, and improving the effectiveness of the model; scaling the feature data to the interval [0,1] so that each feature is in the same range, avoiding the influence of the size of the feature value on the model learning process, especially when using distance-based algorithms (such as KNN). Significance: Standardization and normalization processing make the optimization algorithm more efficient when updating parameters, reduce the volatility of the model during training, and help speed up convergence; standardized data makes the model more generalizable, thereby improving the prediction accuracy of new data, especially when processing high-dimensional features. Step S12: Data sorting and dynamic segmentation: sort the feature data according to the timestamp to ensure that the training, validation and test sets are continuous in time; the key step in processing time series data avoids the problem of data leakage, that is, the test data is used in the training process; the dynamic segmentation strategy is adopted to divide 70% of the data into training sets, 15% into validation sets, and 15% into test sets, ensuring that each data set is representative and covers the change trend of the entire data set. Significance: By confirming the order of time series data, the validation set and the test set can accurately reflect the performance of the model in the real environment, which improves the credibility of the model evaluation; ensure that the training set can learn sufficient and diverse data features, which helps it perform well on the validation set and the test set, and further improves the practicality and reliability of the model. Step S13: Sliding window sample generation: through the sliding window method, continuous data points in the window are extracted from the time series data to generate new training samples; make data utilization more efficient, ensure that the model obtains enough samples to learn the trend of dynamic changes; give each sample generated by the sliding window a clear label, that is, the surge height at the next moment, to ensure that the model can learn the direct relationship between data and target output. Significance: By using the sliding window method, the model can better understand the changes in feature values ​​at different times, thereby effectively capturing the dynamic characteristics of the time series; by creating more samples, especially when data is scarce, the size of the model's training set can be significantly increased, making the model more capable of learning and reasoning.

[0086] In summary, this embodiment achieves standardization and normalization of feature data, ensuring the improvement of data quality and the effectiveness of model training. Secondly, by sorting and dynamically segmenting the data, the rationality and security of the time series data are ensured, and the reliability of model evaluation is enhanced. Finally, effective samples are generated through the sliding window technology, which enhances the diversity and accuracy of model learning. Overall, these processing steps complement each other, help improve the prediction performance and generalization ability of the model, and lay a solid foundation for the accurate prediction of landslide surge height.

[0087] Furthermore, the process of dividing the training set, the validation set and the test set by the dynamic segmentation strategy in step S12 specifically includes the following steps:

[0088] Step S121: When the wind speed or tidal bore reaches the standard, the critical event threshold is set to automatically trigger data segmentation, and training, verification and test sets are generated based on the data samples before and after the event; define the fluidity of time (i.e., the continuity of fluctuations), create a flow sample pool in each time period, and fill it according to the continuous data fluctuations. The deep learning framework is used to realize automatic loading of data from the flow sample pool;

[0089] Step S122: setting a fuzzy range for the division of a specific time period, including the data of the time period, and also including the data of several time points before and after it; by introducing fuzzy logic, more context information of adjacent time periods is received during training;

[0090] Step S123: The division of the training set and the validation set is determined by feature selection according to the actual influence of the input features; the time periods corresponding to the features with high importance scores are given priority in the training set, while the time periods corresponding to the features with lower scores are given priority in the validation set.

[0091] Preferably, in step S121 of this embodiment, the key event threshold division and the creation of the mobile sample pool can automatically identify the time period with significant impact on environmental prediction by setting the key event threshold of wind speed or tidal bore, and include its data in the training set; reduce manual intervention and improve the real-time performance of the segmentation process; define time fluidity, create a mobile sample pool based on data volatility, and ensure that context information between different time periods can be recognized and utilized by the system; the mobile sample pool will contain known and practical data samples, which is convenient for rapid loading and model training. Significance: The automatic segmentation mechanism ensures the timeliness and relevance of model training, so that the model can quickly adapt to sudden changes in the environment; improve the timeliness of training data, and ensure that the model can obtain real-time data when high-risk events (such as extreme weather) occur, thereby improving the prediction ability. The introduction of fuzzy time period division in step S122 sets the fuzzy range of a specific time period, so that the training set contains data within the time period, and can also integrate the data of the time points before and after it; can better capture the potential trends and patterns in the time series; through the setting of the fuzzy range, the model can receive context information of adjacent time periods during training, and enhance the model's comprehensive understanding of short-term fluctuations and long-term trends. Significance: By integrating data from adjacent time periods, the model improves its comprehensive understanding of time series characteristics, especially in the case of extreme climate change, and can make more accurate predictions; it reduces the reliance on hard time nodes, making the model more flexible in the face of nonlinear changes, which helps to improve the robustness of the model. Step S123 Feature selection and division basis: Through feature selection technology, the system can identify and weigh the actual influence of input features, and select data samples that are critical to the prediction results; divide the training set and validation set according to the importance of the features to ensure that the training data contains higher quality feature information; assign features with high importance scores to the training set, and send features with low scores to the validation set, which can help the model focus on the most influential features, thereby improving learning efficiency. Significance: Division based on feature importance ensures that model training is more efficient, effectively utilizes data resources, and avoids possible feature noise interference; improves the interpretability of the model, and can clearly identify which factors are most critical in the prediction, which helps with result analysis and subsequent decision making.

[0092] In summary, the dynamic segmentation strategy of this embodiment ensures the flexibility, timeliness and efficiency of data set division; the application of this strategy in a dynamic environment can effectively improve the prediction accuracy and adaptability of the model, especially when dealing with complex climate change or environmental data, it can better capture and learn motion characteristics; by reducing human intervention and enhancing automated processing capabilities, this method ensures high-quality input of data sets in the deep learning framework, thereby promoting progress in research and application.

[0093] Furthermore, if Figure 5As shown, the process of building a gated recurrent unit GRU network in step S2 specifically includes the following steps:

[0094] Step S21: designing the input layer, GRU layer, fully connected layer and output layer of the gated recurrent unit GRU network; the input layer is used to input feature vectors, the GRU layer captures long-term and short-term dependencies in time series data through gate mechanisms (reset gates and update gates), the fully connected layer maps the output of the GRU layer to a new feature space, and combines the time series information extracted from the GRU layer to generate a more advanced feature representation, and the output layer maps the output of the fully connected layer to specific prediction results, such as the height of landslide surges;

[0095] Among them, the input layer sets the input feature vector x t for:

[0096] x t =[x t,1 ,x t,2 ,…,x t,n ]

[0097] Where n represents the number of features, including landslide-related parameters (such as landslide volume, velocity, and slope), environmental information (such as rainfall, soil moisture, and temperature), and data on historical landslide events;

[0098] The input data needs to be organized as a 3D tensor:

[0099]

[0100] In the formula, m is the number of samples, T is the time step, and n is the number of features in each time step;

[0101] The reset gate r of the GRU layer t It is an important gating mechanism used to control how much historical information should be forgotten when calculating a new hidden state. Its calculation formula is:

[0102] r t =σ(W r ·[h t-1 ,x t ]+b r )

[0103] Where W r represents the weight matrix of the reset gate, h t-1 is the hidden state of the previous time step, σ(·) is the sigmoid activation function with an output range between [0,1], and b r is the bias term for the reset gate;

[0104] Update gate z tIt is used to control the degree to which the previous hidden state retains the current state. The calculation formula is:

[0105] z t =σ(W z ·[h t-1 ,x t ]+b z )

[0106] Where W z is the weight matrix of the update gate, b z is the bias term of the update gate;

[0107] Candidate hidden states It is generated when deciding to update the current hidden state, and its calculation formula is:

[0108]

[0109] In the formula, ⊙ represents element-by-element multiplication, W is the weight matrix of the candidate state, and b is the bias term of the candidate state;

[0110] Current hidden state h t The update formula is:

[0111]

[0112] Multiple GRU layers are stacked, and the GRU units in the subsequent layers not only obtain information from the previous layer, but also combine the hidden states of the previous layers in their calculations to form a residual connection mechanism (the inspiration of ResNet), that is:

[0113]

[0114] In the formula, Represents the current hidden state of the lth layer, which is calculated based on the hidden state of the previous layer and the hidden states of the first two layers. Represents the current hidden state of the l-1th layer, providing information from the previous layer. Represents the current hidden state of layer l-2, providing information from earlier layers;

[0115] The fully connected layer converts the output h of the GRU layer into t Input the fully connected layer and perform nonlinear mapping through the ReLU activation function:

[0116] h fc =ReLU(W fc ·h t +b fc )

[0117] In the formula, h fcrepresents the output of the fully connected layer, that is, the activation value after the output from the GRU layer to the fully connected layer, W fc represents the weight matrix of the fully connected layer, which determines the contribution of the input features to the output features, b fc The bias term of the fully connected layer is used to adjust the output; the ReLU activation function, which maps negative values ​​to zero (i.e. (ReLU(x) = max(0,x))), introduces nonlinear characteristics;

[0118] The output layer maps the features to the predicted values ​​through full connection and uses a linear activation function:

[0119] y pred =W out ·h fc +b out

[0120] In the formula, y pred is the final predicted value of surge height; W out The weight matrix of the output layer is used to map the features of the fully connected layer to the output values, b out The bias term of the output layer;

[0121] Step S22: Use mean square error as the loss function of the regression task and dynamically adjust the learning rate according to the gradient change;

[0122]

[0123] Where, L is the mean square error (MSE) calculated by the loss function, which measures the difference between the surge height predicted by the model and the true value, N is the total number of samples used to calculate the average loss, The true surge height of the i-th sample, The model predicts the swell height of the i-th sample;

[0124] Step S23: When new data is input, the weights are updated through the back propagation algorithm, and the functions of the reset gate and the update gate are adjusted through feedback when different environmental conditions change.

[0125] Preferably, in step S21 of this embodiment, the GRU network structure is designed, the input layer: receives the feature vector related to the landslide, and the GRU layer captures the long-term and short-term dependencies in the time series data by resetting the gate and updating the gate; the full connection maps the GRU output to a new feature space to generate a more advanced feature representation; the output layer is finally mapped to a specific prediction result, such as the height of the landslide surge, effectively processing the time series data, using the GRU gating mechanism to prevent the gradient from disappearing, so that the model can learn historical information and time series features; through the mapping of the fully connected layer, the model can better understand and summarize the complex relationship of the input features. Significance: Accurate prediction of the height of the landslide surge is achieved, which enhances the early warning capability of natural disasters and helps reduce risks. Step S22 sets the loss function, using the mean square error MSE as the loss function of the regression task, and dynamically adjusts the learning rate according to the gradient change. The MSE loss function provides a quantitative standard to evaluate the difference between the prediction and the true value, so that the model can be optimized according to the actual performance; dynamic learning rate adjustment helps to improve the convergence speed and stability of the training. Significance: By optimizing the loss function, the prediction accuracy of the model is improved, thereby increasing the reliability and adaptability of the model. Step S23 weight update and gating mechanism feedback: when new data is input, the weight of the network is updated through the back propagation algorithm; when different environmental conditions change, the functions of the reset gate and update gate are adjusted through feedback; the back propagation algorithm enables the model parameters to be automatically optimized, thereby improving the adaptability to new data; the dynamic adjustment of the gating mechanism ensures that the GRU network can still effectively capture the temporal characteristics of the data under different environmental conditions. Significance: Improves the model's ability to resist interference in the face of changing environments, making it more sensitive to real-time input data and having stronger generalization capabilities.

[0126] In summary, the steps of building the GRU network in this embodiment form a complete closed loop from feature input to model output, and achieve accurate prediction of landslide surge height. In practical applications, it can be used in many fields such as natural disaster monitoring, environmental protection, and engineering safety, which not only improves the ability of scientific prediction, but also helps to take timely response measures in emergencies, thereby reducing the losses caused by disasters.

[0127] Furthermore, the process of resetting the gate and updating the gate using the dynamic coupling strategy in step S21 specifically includes the following steps:

[0128] Step S211: Calculate the interaction between the reset gate and the update gate. When calculating the reset gate at the current moment, consider the combined information of the previous hidden state and the current input. Introduce the update gate at the previous moment, and refer to both the current input and the previous hidden state to determine the amount of information retained at the current moment.

[0129] Step S212: Integrate historical information in the process of generating candidate hidden states. The generation process of candidate hidden states includes the interaction of reset gates and update gates. The weight of the reset gate affects the contribution of the previous hidden state to the candidate hidden state. Through comprehensive analysis of historical information and new inputs, the candidate state reflects the characteristics of the current moment.

[0130] Step S213: Hidden state update process and its dynamic coupling. Through the dynamic coupling mechanism, the reset gate and the update gate jointly determine the flow and storage of information. When updating the hidden state at the current moment, the reset gate determines the extent to which the previous state needs to be forgotten. The update gate is responsible for determining how much information from the previous hidden state should be retained in the current hidden state. According to the dynamic coupling adjustment, when the update gate is higher, the past information is retained; and when the update gate is lower, the influence of historical information is suppressed and the introduction of new information is allowed.

[0131] Preferably, step S211 of this embodiment calculates the interaction between the reset gate and the update gate. By combining the previous hidden state and the current input, the reset gate and the update gate enable the model to dynamically adjust the information flow according to the context; the operation increases the flexibility of the model, allowing it to adaptively change its reliance on historical information according to the current input; the introduction of the influence of the update gate at the previous moment on the reset gate enables the model to intelligently retain or forget information when processing continuous time series; for example, in some cases, the model may need to pay more attention to historical data, while in other cases it needs to accept new information more quickly. Significance achieved: Through this step, the model can adjust the use of historical information in different input scenarios, thereby improving its adaptability to complex time series data patterns; this mechanism makes the model's predictions more accurate because it can dynamically balance the weight between old and new information. In step S212, historical information is integrated in the process of generating candidate hidden states. In the process of generating candidate hidden states, the interaction between the reset gate and the update gate means that the previous state and the current input information are dynamically fused through the weight of the reset gate, which improves the accuracy and representativeness of the candidate state; by integrating historical information with the current input, the candidate state can more comprehensively reflect the characteristics of the current moment, and has better context understanding ability, so as to better capture the changes in the time series. The significance achieved: when the model is updating the state, it can more carefully capture the short-term and long-term dependencies in the time series, and improve the model's ability to process complex data; the accurate generation of candidate states makes information transmission in the recurrent neural network more efficient, reduces unnecessary information interference, and improves the efficiency of the model. Step S213 hidden state update process and its dynamic coupling effect, the dynamic coupling mechanism used by the reset gate enables the model to determine the degree of information retention of the previous state at the current moment, thereby effectively managing the flow and storage of information; through dynamic coupling, the update gate is responsible for determining how much historical information needs to be retained, so that the model can adapt to the current input and flexibly introduce or suppress the influence of historical information. Significance achieved: The interaction between the reset gate and the update gate improves the model's memory ability when processing information, allowing the model to forget and retain information more reasonably, thereby improving the understanding of time series; this dynamic update method makes the model more adaptable to the environment and can still maintain high accuracy and robustness under changing input conditions.

[0132] In summary, the reset gate and update gate of this embodiment adopt the dynamic coupling strategy process, which realizes the intelligent management of information flow and storage through three key steps. These steps individually and as a whole improve the flexibility, adaptability and prediction ability of the model, so that GRU has better performance when processing complex time series tasks. The dynamic coupling strategy not only optimizes the structural design of the model, but also significantly improves its performance in practical applications.

[0133] Furthermore, the process of comprehensive analysis of historical information and new input in step S212 specifically includes the following steps:

[0134] Step S2121: If the value of the reset gate is close to 0, it is considered that the historical information will be forgotten and the current calculation depends more on the new input rather than the previous state; if the value of the reset gate is close to 1, the historical state is retained;

[0135] Step S2122: Process the previous hidden state, weight some parts of the previous hidden state by resetting the gate, and obtain modified historical information; integrate the historical information according to the current situation by resetting the gate;

[0136] Step S2123: The current input is introduced. In the generation of the candidate hidden state, the current input is combined with the modified historical information to jointly form a new candidate state, which reflects the characteristics of the current moment and is a comprehensive processing of the historical information and the new input.

[0137] Preferably, the value of the reset gate in step S2121 of this embodiment is affected. When the value of the reset gate is close to 0, the system effectively ignores the past hidden state and mainly relies on the current input data; the network can quickly adapt to sudden events or input scenarios with large changes; when the value of the reset gate is close to 1, the system can retain more historical information. In this case, the model will rely more on previous knowledge and incorporate the characteristics of the previous state into the current calculation. The significance achieved: The dynamic forgetting and retention mechanism makes the model more flexible in adaptability and resource utilization, and can flexibly adjust the dependence on historical information according to the characteristics of the data. In sequence data processing, the performance of the model is effectively improved, and the stability and accuracy of the model when processing long sequence data are enhanced. Step S2122 weighted processing of historical information, the weighted processing step of the previous hidden state, screens and integrates historical information through the reset gate to form modified historical information; it can remove irrelevant or unimportant information and only retain historical information that is helpful to the current task; the model can selectively retain and utilize previous states in various contexts, thereby providing a state representation that is more in line with the current input for subsequent calculations. Significance achieved: Through this targeted processing method, the model can more accurately reflect the characteristics of the input data when generating candidate states; avoid the redundancy and noise of historical information, make the candidate states more informative, and thus promote the model to better understand the current input and make corresponding decisions. Step S2123 combines the current input with historical information, and the current input with the modified historical information to form a new candidate state, so that the candidate state can fully reflect the characteristics of the current moment; not only integrates historical information, but also ensures the immediacy and relevance of the new input; the candidate state contains important information from the past and is directly affected by the current input, thereby achieving a dynamic state update. Significance achieved: The effective combination of historical and current information enables the model to better capture time dependencies in sequence learning tasks. The model has a stronger ability to understand context, so that it can make predictions and judgments more accurately; this comprehensive processing method is the key to the model learning the essence of time series data, thereby improving the overall performance of tasks such as language modeling and time series prediction.

[0138] In summary, this embodiment constructs a flexible and adaptable model framework by dynamically integrating historical information with current input, which enables the model to make full use of historical experience and real-time information when processing complex sequence data, effectively improves the performance of deep learning models in practical applications, and promotes the improvement of the robustness and accuracy of the model in various data environments.

[0139] Furthermore, if Figure 6 As shown, the process of training the gated recurrent unit GRU network model in step S3 specifically includes the following steps:

[0140] Step S31: extracting a batch of data samples from the training set for training the gated recurrent unit GRU network model, selecting a loss function and an optimization algorithm to adjust the parameters of the gated recurrent unit GRU network model to minimize the prediction error;

[0141] Step S32: In each iteration, the input data is provided to the gated recurrent unit GRU network model for forward propagation to calculate the prediction result; at the random sampling probability u (in the uniform distribution [0,1]), whether to accept the current parameter update is determined according to the acceptance probability of the gated recurrent unit GRU network model parameters;

[0142] If u is less than the acceptance probability, accept the current parameter update (i.e., m i =m * );

[0143] Otherwise, keep the model parameters of the previous step (i.e., m i =m (i-1) );

[0144] Step S33: Repeat the above parameter updating steps until the loss function converges to the minimum value or reaches the preset number of training times; use the validation set data to evaluate the performance of the gated recurrent unit GRU network model, and adjust the hyperparameters according to the performance of the gated recurrent unit GRU network model on the validation set; perform a generalization test by changing the data sets of the training set and the validation set; use the test set data to evaluate the trained gated recurrent unit GRU network model.

[0145] Preferably, in step S31 of this embodiment, data sample extraction and parameter adjustment, a batch of data samples are extracted from the training set to ensure the diversity and representativeness of the input data, which is crucial for training the model; loss functions (such as mean square error, cross entropy, etc.) and optimization algorithms (such as Adam, SGD, etc.) are selected, and the learning objectives of the model and its optimization method are defined through tools. Significance: The extraction of data samples enables the model to learn the potential patterns of the data, while improving the training efficiency through batch processing; the selection of appropriate loss functions and optimization algorithms can guide the model to converge in a more effective way, reduce prediction errors, and thus improve the accuracy and generalization ability of the model. Step S32 forward propagation and parameter update decision, in each iteration, the input data is forward propagated through the GRU model, and the model evaluates the current model performance by calculating the error between the network output and the true label; according to the probability u of random extraction and the probability of acceptance (based on the evaluation of the current parameters), it is decided whether to update the parameters, and a random search mechanism is introduced to optimize the model parameters. Significance: The forward propagation process is the core of model training. Through this step, the model can learn and adjust its weights to better fit the training data; the introduction of a randomized parameter update mechanism (such as the MCMC method) can explore the parameter space, avoid falling into a local optimal solution, improve the model's exploration ability, make parameter updates more flexible, and enhance the robustness of the model. Step S33 iteration and model evaluation, repeat the update steps so that the model gradually converges, and stop training when the loss function reaches the minimum value or reaches the preset number of training times; use the validation set to evaluate the model performance, monitor the model's training effect in real time, and provide feedback for hyperparameter tuning. Significance: Through multiple iterations, the model can be gradually optimized to improve its performance on the training data while controlling the risk of overfitting; the use of the validation set ensures the generalization ability of the model. Through hyperparameter adjustment, the model can be further optimized to improve its performance on unseen data; generalization testing provides additional signals to verify the performance of the model under various data conditions (different distributions, different features, etc.), which helps to enhance the confidence and reliability of the model.

[0146] In summary, this embodiment achieves the following key goals through the organic combination of the above steps in the process of training the gated recurrent unit GRU network model: through appropriate data and loss functions, the model can extract meaningful patterns and features from the data; through precise parameter update strategies, the performance of the model on the test set is improved to ensure that the model not only has good performance on the training set, but also is reliable in real applications; through strict verification and generalization tests, the robustness of the model is ensured and the performance in various practical situations is improved. It ensures that the training process of the model is scientific and efficient, so that the final GRU model can perform well in the task of landslide surge height prediction.

[0147] Furthermore, the process of minimizing the prediction error in step S31 specifically includes the following steps:

[0148] Step S311: After a batch of data samples in the training set is selected, the batch of data samples is input into the GRU model for processing; each sample is a sequence of fixed length, the sequence is represented as a vector, and word embedding is applied to generate a corresponding vector representation for each time step;

[0149] Step S312: The update gate and the candidate state are combined to generate a new hidden state. The update gate determines how much of the previous state to retain and how much new information to introduce. The last hidden state is used to generate the final output.

[0150] Step S313: Use the loss function to evaluate the performance of the model and calculate the current loss value. The loss function calculates the error between the current model output and the actual label;

[0151] Among them, the adaptive weighted cross entropy loss function expression is:

[0152]

[0153] In the formula, L represents the loss value, y true One-hot encoded vector of the true label, y pred The probability distribution predicted by the model, n is the total number of samples, C is the total number of categories, w i The adaptive weight of the i-th sample is dynamically adjusted according to the characteristics of the sample. λ controls the weight hyperparameter of the regularization term. Regularization(w): Regularization function, which can be L1 or L2 regularization of the weight to prevent overfitting.

[0154] Preferably, step S311 of this embodiment inputs data to the GRU model. By selecting a batch of data samples in the training set and converting these samples into sequences of fixed length, the model can accept structured input; enabling GRU to effectively process time series or sequence data (such as text); applying word embedding technology (such as Word2Vec, GloVe or BERT) to generate corresponding vector representations for each time step, these vectors help to express the semantic features of the vocabulary, thereby providing the model with richer context information; through vectorization, the input data processed by the model is more efficient, which helps to capture the relationship and semantics between words. Significance achieved: By representing time series data as vector input, the GRU model can capture the interdependence between different time steps in the sequence; it lays the foundation for subsequent information processing and enables the model to better understand the dynamic changes of data; word embedding converts discrete text data into vectors in a high-dimensional continuous space, so that the model can mathematically effectively perform linear combinations and nonlinear transformations, enhancing the expressive power of the model. In step S312, the update gate is combined with the candidate state to generate a new hidden state. Through the update gate mechanism, the model can selectively retain information from the previous hidden state and introduce new information at the same time; enabling GRU to dynamically adjust the retention and discarding of information flow according to the specific needs of the current input; by calculating the current candidate state and combining the output of the update gate, the model generates a new hidden state; when processing different types of inputs, GRU can adaptively adjust its internal state. Significance achieved: GRU handles long-term dependency problems through its unique gating design, overcoming the shortcomings of traditional RNNs that are prone to gradient vanishing or exploding on long sequences; therefore, GRU can retain and utilize important information in a longer context range; the dynamically updated hidden state can synthesize and integrate information in multiple time steps, thereby generating a deep understanding of sequence data, which is crucial for many sequence prediction tasks (such as language modeling, sentiment analysis, etc.). Step S313 uses the loss function to evaluate the model performance. The model uses the loss function to evaluate the error between its prediction results and the actual label. Common loss functions such as mean square error (MSE) and cross entropy loss fully reflect the performance and prediction accuracy of the model. The loss value provides a feedback signal for the model during the training process, so that the subsequent gradient update can effectively reduce the error. Significance achieved: The calculation of the loss function guides the direction and degree of model optimization. Through the back propagation process, the decrease in the loss value will directly drive the adjustment of the model parameters, and ultimately make the model perform better on the target task. The loss function is not only used for gradient updates during training, but also can be used to monitor the performance of the model, including overfitting and underfitting. It provides a basis for further model adjustments (such as hyperparameter tuning, regularization method selection, etc.).

[0155] In summary, this embodiment creates a closed loop of efficient data input processing, dynamic information integration, and prediction performance evaluation. Through these technical means, the GRU model can maintain flexibility when facing complex time series data, improve the accuracy and generalization ability of the model, and thus achieve the goal of minimizing prediction errors.

[0156] Furthermore, the process of performing a generalization test by changing the data sets of the training set and the validation set in step S33 specifically includes the following steps:

[0157] Step S331: After the initial training, record the performance of each sample on the validation set, calculate the misclassification rate of each sample, the cumulative loss value, and calculate its F1-score on the validation set; assign weights to the samples according to the misclassification rate of the samples, set a threshold, and increase the weight of samples with a misclassification rate greater than 0.2 by 1.5 times;

[0158] Step S332: In the next round of training, the weights are directly applied to the loss function; adversarial samples are generated, the gradient of the loss function with respect to the input is calculated, and several adversarial samples are generated; the generated adversarial samples are incorporated into the validation set, and the accuracy and robustness of the model are re-evaluated;

[0159] Step S333: separate the main task (such as time series prediction) from the related additional tasks (such as trend detection or anomaly detection); design an independent sub-network for each task, and share the underlying feature extraction layer of GRU; adopt alternating training during the training process, that is, regularly update the shared layer of all tasks, and update the additional tasks every 5 epochs;

[0160] Step S334: Evaluate the performance of each task after each epoch and draw the loss change curve between different tasks; use TensorBoard to monitor the training process and record the loss, learning rate change and evaluation indicators of each epoch; if the validation set loss does not improve in three consecutive epochs, reduce the learning rate.

[0161] Preferably, in step S331 of this embodiment, sample performance recording and weight allocation are performed by evaluating the performance of the validation set samples, recording the misclassification rate and cumulative loss of each sample, so that the model can identify which samples contribute less to model training and which samples are borderline cases that are difficult to classify; dynamic sample recalibration is performed based on the misclassification rate, focusing on those samples that trouble the model and increasing their importance in training. Significance: The weight allocation method can guide the model to learn more effectively, significantly improve its performance when facing complex samples, thereby expanding the generalization ability of the model; realize the model's focus on unbalanced samples or borderline samples, improve the practicality of the model in real scenarios, and reduce the risk of overfitting. Step S332 is adversarial sample generation and re-evaluation, using the strategy of generating adversarial samples to test the robustness of the model in the face of slight perturbations; during the training process, rely on adversarial samples to further evaluate the sensitivity of the model; the introduction of adversarial samples can reveal the potential weaknesses and deficiencies of the model, making the model more resilient when encountering noise and attacks in the real world. Significance: By evaluating the performance of the model on adversarial samples, the model's adaptability to uncertainty and abnormal situations can be enhanced, and its stability and reliability in complex environments can be improved; the model's anti-interference ability is effectively enhanced, which means that it can better handle more difficult and complex data scenarios in practical applications. Step S333 Multi-task learning architecture design separates the main task from the additional task to ensure that each task can be optimized independently, while sharing the underlying features at the level, so that the training efficiency can be improved with the help of complementary learning between different tasks; the alternating training strategy allows the shared layer to maintain consistency, while providing timely updates for additional tasks and strengthening the synergy between tasks. Significance: The multi-task learning strategy enhances the flexibility of the model, especially when faced with new tasks or data sets, the model can quickly adapt and effectively transfer existing experience; in the same training process, by sharing features, redundant calculations can be avoided, the training speed can be increased, and better performance can be achieved with limited resources. Step S334 monitors and dynamically adjusts the training process. By real-time monitoring of the training process and recording the loss, learning rate and other evaluation indicators of each epoch, it can help to timely understand the learning status of the model and monitor the training progress; set the dynamic adjustment rules of the learning rate to ensure that the model can be adjusted in time under poor training conditions (such as overfitting) to maintain model performance. Significance: Ensures the high efficiency of training and avoids ineffective waste of computing time and resources. By dynamically adjusting the learning rate, it can be adaptively adjusted according to the actual training situation to increase the chance of successful convergence; the monitoring system provides a feedback mechanism for the model, making model training more controllable, improving the ability to identify potential problems in the training process, and enhancing the transparency of the training process.

[0162] In summary, in the entire generalization test process, the organic combination of various steps in this embodiment promotes the learning ability and application performance of the model in complex environments. It not only emphasizes the performance improvement of the model, but also strengthens its adaptability and stability in practical scenarios, and ultimately improves the generalization ability and usability of the model. Through such a systematic and targeted solution, a solid foundation can be laid for building a more powerful machine learning model.

[0163] like Figure 7 As shown, this embodiment also provides an embodiment of a landslide surge height prediction system based on a gated cyclic unit. In this embodiment, the landslide surge height prediction system based on a gated cyclic unit is applied to the landslide surge height prediction method based on a gated cyclic unit in the above embodiment. The landslide surge height prediction system based on a gated cyclic unit includes a data preprocessing module 1, a model building module 2, and a model training module 3 that are electrically connected in sequence;

[0164] Among them, the data preprocessing module 1 is used to collect relevant data of large-scale physical model tests in the landslide surge area. The relevant data include characteristic data such as landslide volume, sliding speed, slope angle and similar material properties used to simulate real geotechnical materials, and the characteristic data is preprocessed for data standardization and normalization. The preprocessed characteristic data is used as the input of the gated recurrent unit GRU model to obtain the surge height; the model construction module 2 is used to build a gated recurrent unit GRU network using TensorFlow in the Python compilation environment; the update gate and reset gate are introduced to control the flow of information. The update gate determines how much information of the previous state should be retained in the current state, while the reset gate controls how much previous state information should be forgotten; the model training module 3 is used to train the gated recurrent unit GRU network model, and the historical landslide data is used to train the gated recurrent unit GRU network model to adjust the model parameters; the trained gated recurrent unit GRU model is deployed to the actual landslide surge prediction system for real-time monitoring.

[0165] Preferably, the data preprocessing module 1 of this embodiment collects important characteristic data related to landslides, such as landslide volume, sliding speed, slope angle and characteristics of similar materials; makes the information input to the model as detailed and diverse as possible; standardizes and normalizes the characteristic data to eliminate the dimensional influence between different features, so that the data is more suitable for deep learning model training. This step ensures the balance of the data and makes the training process more stable. Significance: Through effective data preprocessing, the sensitivity of the model to outliers or scale differences is reduced, thereby improving the learning ability and prediction accuracy of the model; ensuring that a clear and reliable data foundation is provided for subsequent model construction, which helps to improve the prediction performance and generalization ability of the model. The model construction module 2 uses TensorFlow to establish a GRU network, combined with the update gate and reset gate design, which can effectively capture the long-term and short-term dependencies in time series data; the update gate controls the retained information, and the reset gate controls the forgotten information. In this way, the model can efficiently process historical data and enhance information extraction capabilities. Significance: Through the rationally designed GRU structure, the modeling capability of complex time series data is improved to meet the needs of landslide surge height prediction; the adaptability of the model in the face of different input data is improved, so as to achieve more accurate prediction and reduce the prediction error. Model training module 3 uses historical landslide data for model training and uses the back propagation algorithm to adjust the model parameters to minimize the prediction error; it can be continuously updated and trained with new landslide data to maintain good prediction capabilities. Significance: Through the training of historical data, it is ensured that the trained GRU model can effectively reflect the real landslide surge dynamics with a certain degree of accuracy and reliability; the trained model is deployed in the real-time landslide surge prediction system, so that the monitoring data and model output can respond in time, provide decision support, and improve the early warning capability.

[0166] In summary, the landslide surge height prediction system based on the gated circulation unit in this embodiment aims to achieve real-time monitoring and prediction of complex landslide activities through effective data preprocessing, sophisticated model construction and efficient model training. Each module provides a scientific basis and strong technical support for landslide disaster prevention and mitigation, reduces disaster risks, and provides necessary data support and decision-making assistance for related fields. This comprehensive approach not only improves the accuracy and reliability of the model, but also enhances the practicality and response speed of the system, enabling it to play an important role in practical applications.

[0167] like Figure 8 As shown, this embodiment provides an embodiment of an electronic device. In this embodiment, the electronic device 4 includes a processor 41 and a memory 42 coupled to the processor 41.

[0168] The memory 42 stores program instructions for implementing the landslide surge height prediction method based on the gated cycle unit according to any of the above embodiments.

[0169] The processor 41 is used to execute the program instructions stored in the memory 42 to perform landslide surge height prediction based on the gated cycle unit.

[0170] The processor 41 may also be referred to as a CPU (Central Processing Unit). The processor 41 may be an integrated circuit chip having signal processing capabilities. The processor 41 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0171] Further, Fig. 9 The schematic diagram of the structure of the storage medium of an embodiment of the present application is that the storage medium 5 of the embodiment of the present application stores program instructions 51 that can implement all the above methods, wherein the program instructions 51 can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0172] like Fig.10 As shown, the physical model test data is compared with the predicted values ​​of the gated recurrent unit method; Fig.11 This is the K-fold cross validation result diagram (K=5). From the attached figure, we can see that the physical model test data results and the predicted values ​​have a high degree of overlap, proving that the accuracy of the model prediction results is very high;

[0173] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0174] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units. The above is only an implementation mode of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, is also included in the patent protection scope of the present invention.

[0175] The specific implementation methods of the invention are described in detail above, but they are only examples, and the invention is not limited to the specific implementation methods described above. For those skilled in the art, any equivalent modification or substitution of the invention is also within the scope of the invention, therefore, the equalization, modification, improvement, etc. made without departing from the spirit and principle of the invention should be included in the scope of the invention.

Claims

1. A landslide surge height prediction method based on gated cyclic units, characterized in that: The landslide surge height prediction method based on gated circulation units comprises: Collect relevant data from large-scale physical model tests in the landslide surge area, and perform data standardization and normalization preprocessing on the characteristic data. The preprocessed characteristic data is used as the input of the gated recurrent unit (GRU) model to obtain the surge height. Use Zhang Liuxun in the Python compilation environment to build a gated recurrent unit GRU network; introduce update gates and reset gates to control the flow of information. The update gate determines how much information from the previous state should be retained in the current state, while the reset gate controls how much previous state information should be forgotten; Train the gated recurrent unit (GRU) network model and use historical landslide data to train the gated recurrent unit (GRU) network model to adjust the model parameters; deploy the trained gated recurrent unit (GRU) model to the actual landslide surge prediction system for real-time monitoring.

2. The method for predicting landslide surge height based on gated cyclic units according to claim 1, characterized in that: The relevant data include landslide volume, sliding speed, slope angle and characteristic data used to simulate the similar material properties of real geotechnical materials.

3. The method for predicting landslide surge height based on gated cyclic units according to claim 1, characterized in that: Build a gated recurrent unit GRU network, including: Design the input layer, GRU layer, fully connected layer and output layer of the gated recurrent unit GRU network; The mean square error is used as the loss function of the regression task, and the learning rate is dynamically adjusted according to the gradient changes; When new data is input, the weights are updated through the back-propagation algorithm, and the functions of the reset gate and update gate are adjusted through feedback when different environmental conditions change.

4. The method for predicting landslide surge height based on gated cyclic units according to claim 3, characterized in that: The input layer is used to input feature vectors. The GRU layer captures long-term and short-term dependencies in time series data through a gate mechanism. The fully connected layer maps the output of the GRU layer to a new feature space and combines the time series information extracted from the GRU layer to generate a more advanced feature representation. The output layer maps the output of the fully connected layer to a specific prediction result, the height of the landslide surge.

5. The method for predicting landslide surge height based on gated cyclic units according to claim 3, characterized in that: in, The input layer sets the input feature vector x t for: x t =[x t,1 ,x t,2 ,…,x t,n ] Where n represents the number of features, including landslide related parameters, landslide volume, rate, slope; environmental information, rainfall, soil moisture, temperature, and data of historical landslide events; The input data needs to be organized as a 3D tensor: In the formula, m is the number of samples, T is the time step, and n is the number of features in each time step; The reset gate r of the GRU layer t It is an important gating mechanism used to control how much historical information should be forgotten when calculating a new hidden state. Its calculation formula is: r t =σ(W r ·[h t-1 ,x t ]+b r ) Where W r represents the weight matrix of the reset gate, h t-1 is the hidden state of the previous time step, σ(·) is the sigmoid activation function with an output range between [0,1], and b r is the bias term for the reset gate; Update gate z t It is used to control the degree to which the previous hidden state retains the current state. The calculation formula is: z t =σ(W z ·[h t-1 ,x t ]+b z ) Where W z is the weight matrix of the update gate, b z is the bias term of the update gate; Candidate hidden states It is generated when deciding to update the current hidden state, and its calculation formula is: Where ⊙ represents element-by-element multiplication, W is the weight matrix of the candidate state, and b is the bias term of the candidate state; Current hidden state h t The update formula is: Multiple GRU layers are stacked, and the GRU units in the subsequent layers not only obtain information from the previous layer, but also combine the hidden states of the previous layers in their calculations, forming a residual connection mechanism, namely: In the formula, Represents the current hidden state of the lth layer, which is calculated based on the hidden state of the previous layer and the hidden states of the first two layers. Represents the current hidden state of the l-1th layer, providing information from the previous layer. Represents the current hidden state of layer l-2, providing information from earlier layers; The fully connected layer converts the output h of the GRU layer into t Input the fully connected layer and perform nonlinear mapping through the ReLU activation function: h fc =ReLU(W fc h t +b fc ) In the formula, h fc represents the output of the fully connected layer, that is, the activation value after the output from the GRU layer to the fully connected layer, W fc represents the weight matrix of the fully connected layer, which determines the contribution of the input features to the output features, b fc The bias term of the fully connected layer is used to adjust the output; the ReLU activation function maps negative values ​​to zero, that is, ReLU(x) = max(0,x), introducing nonlinear characteristics; The output layer maps the features to the predicted values ​​through full connection and uses a linear activation function: y pred =W out ·h fc +b out In the formula, y pred is the final predicted value of surge height; W out The weight matrix of the output layer is used to map the features of the fully connected layer to the output values, b out Bias term for the output layer.

6. The method for predicting landslide surge height based on gated cyclic units according to claim 3, characterized in that: Loss function: Where, L is the mean square error calculated by the loss function, which measures the difference between the surge height predicted by the model and the true value, N is the total number of samples used to calculate the average loss, The true surge height of the i-th sample, The model predicts the swell height for the i-th sample.

7. The method for predicting landslide surge height based on gated cyclic units according to claim 1, characterized in that: The reset gate and update gate adopt a dynamic coupling strategy, including: Calculate the interaction between the reset gate and the update gate. When calculating the reset gate at the current moment, consider the combined information of the previous hidden state and the current input. Introduce the update gate at the previous moment, and refer to both the current input and the previous hidden state to determine the amount of information retained at the current moment. In the process of generating candidate hidden states, historical information is integrated. Through comprehensive analysis of historical information and new inputs, the candidate states reflect the characteristics of the current moment. The hidden state update process and its dynamic coupling, the update gate is responsible for determining how much information from the previous hidden state should be retained in the current hidden state.

8. The method for predicting landslide surge height based on gated cyclic units according to claim 7, characterized in that: The generation process of candidate hidden states involves the interaction of reset gates and update gates, and the weight of the reset gate affects the contribution of the previous hidden state to the candidate hidden state.

9. The method for predicting landslide surge height based on gated cyclic units according to claim 7, characterized in that: Through the dynamic coupling mechanism, the reset gate and the update gate jointly determine the flow and storage of information; when updating the hidden state at the current moment, the reset gate determines the extent to which the previous state needs to be forgotten.

10. A landslide surge height prediction system based on gated cyclical unit, which is applied to the landslide surge height prediction method based on gated cyclical unit as claimed in any one of claims 1 to 9, characterized in that: The landslide surge height prediction system based on gated circulation unit includes: The data preprocessing module is used to collect relevant data of large-scale physical model tests in the landslide surge area, and perform data standardization and normalization preprocessing on the characteristic data. The preprocessed characteristic data is used as the input of the gated recurrent unit GRU model to obtain the surge height; Model building module, used to build a gated recurrent unit (GRU) network using tensorflow in Python compilation environment; introduce update gate and reset gate to control the flow of information. The update gate determines how much information of the previous state should be retained in the current state, while the reset gate controls how much information of the previous state should be forgotten. The model training module is used to train the gated recurrent unit (GRU) network model. The historical landslide data is used to train the gated recurrent unit (GRU) network model to adjust the model parameters. The trained gated recurrent unit (GRU) model is deployed to the actual landslide surge prediction system for real-time monitoring.

Images