Coastal tidal action flowing water grouting simulation method and system based on deep learning

Abstract

The invention discloses a coastal tidal action flowing water grouting simulation method and system based on deep learning, and relates to the technical field of flowing water grouting simulation in a coastal environment. Preprocessed tidal data, hydrogeological data and grouting parameter data are input into a pre-trained deep learning model for prediction; dynamic changes of grouting parameters under the tidal effect are predicted; and a grouting simulation model of the stratum under the tidal effect is established based on the diffusion equation, the dynamic change of the grouting parameters is substituted into the grouting simulation model for grouting simulation calculation, the diffusion behavior of grout is obtained, multi-objective optimization of the grouting parameters is conducted based on the diffusion behavior of the grout, and an optimal grouting parameter combination is obtained. A deep learning model and a physical constraint model are combined, dynamic changes of grouting parameters under the tidal condition are predicted through the deep learning model, the dynamic changes serve as input data of the physical constraint model, the grouting diffusion process under the tidal dynamic environment is accurately simulated, and an optimal grouting parameter combination is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrodynamic grouting simulation in coastal environments, and particularly to a hydrodynamic grouting simulation method and system based on deep learning for coastal tidal action. Background Art

[0002] The statements in this section merely provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] The groundwater flow pattern in coastal areas is deeply affected by the periodic changes of tides. In particular, occasional large fluctuations may occur during tidal peaks. This instability poses challenges to the grouting treatment of tunnels and underground projects. Due to the long-term action of multiple factors such as tidal scouring, fresh-salt water circulation displacement, and seawater erosion on rock masses, the engineering geological and hydrogeological conditions in coastal areas are extremely complex. Among them, sudden water inrush, as one of the main geological disasters during the development of underground spaces in coastal karst areas, seriously threatens engineering construction, the ecological environment, and people's lives and property safety. Since the ocean is the discharge boundary of the aquifer, frequent and periodic tidal actions will cause water level fluctuations in the aquifer, which in turn will change the groundwater flow velocity and even the flow direction near the boundary, thus directly affecting parameters such as the water volume and water pressure of sudden water inrush. Such fluctuations have a significant impact on engineering construction, especially underground projects.

[0004] The grouting technology is an effective means to control sudden water inrush and can effectively carry out hydrodynamic plugging and formation repair and reinforcement. However, traditional grouting methods rely on empirical formulas or numerical models and lack a real-time response mechanism for parameter selection and adjustment, making it difficult to cope with the dynamic changes in the coastal hydrodynamic environment. Existing hydrodynamic grouting research mostly focuses on constant flow velocity conditions and does not consider complex dynamic factors such as tidal cycle changes, tidal range effects, and occasional large fluctuations. In addition, few methods can intelligently optimize the selection of grouting parameters to adapt to tidal fluctuations that change over time and environment. Summary of the Invention

[0005] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a hydrodynamic grouting simulation method and system based on deep learning for coastal tidal action, which combines a deep learning model and a physical constraint model. The dynamic changes of grouting parameters under tidal conditions are predicted by the deep learning model and used as input data for the physical constraint model to accurately simulate the grouting diffusion process in a tidal dynamic environment and obtain the optimal combination of grouting parameters.

[0006] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides a hydrodynamic grouting simulation method based on deep learning for coastal tidal action, including:

[0008] Obtain tidal data, hydrogeological data, and grouting parameter data in the coastal area and perform preprocessing;

[0009] Input the preprocessed tidal data, hydrogeological data, and grouting parameter data into a pre-trained deep learning model for prediction, and predict the dynamic changes of grouting parameters under the action of tides;

[0010] Establish a grouting simulation model of the formation under the action of tides based on the diffusion equation, substitute the dynamic changes of the grouting parameters into the grouting simulation model for grouting simulation calculation to obtain the diffusion behavior of the slurry, and perform multi-objective optimization of the grouting parameters based on the diffusion behavior of the slurry to obtain the optimal grouting parameter combination.

[0011] For a further technical solution, the deep learning model sequentially includes a convolutional neural network layer, an improved long short-term memory network layer, and an improved self-attention mechanism.

[0012] For a further technical solution, the improved long short-term memory network layer enhances the modeling ability for tidal fluctuations by introducing tidal dynamic weights and periodic parameters, and its formula is expressed as:

[0013] f t =σ(W f ·[h t-1 ,x t +b f +γ f ·θ LSTM (t))

[0014] i t =σ(W i ·[h t-1 ,x t +b i +γ i ·θ LSTM (t))

[0015]

[0016]

[0017] o t =σ(W o ·[h t-1 ,x t +b o +γ o ·θ LSTM (t))

[0018] h t =o t ·tanh(C t )

[0019] Among them, f t is the forget gate, i t is the input gate, is the candidate memory cell state, C t is the memory cell state, o t is the output gate, γ f 、γ i 、γ C 、γ o are coefficients used to adjust the influence of tides on the forget, input, candidate memory, and output gates, σ is the activation function, W f 、W i 、W C 、W o are the weight matrices corresponding to the forget gate, input gate, candidate memory cell state, and output gate respectively, h t is the hidden state at the current time step t, h t-1 is the hidden state at the previous time step before the current time step t, x t is the input feature at the current time step t, b f 、b i 、b C 、b o are the bias vectors corresponding to the forget gate, input gate, candidate memory cell state, and output gate respectively, θ LSTM (t) is the tidal dynamic weight, p t is the periodic parameter.

[0020] For a further technical solution, the improved self-attention mechanism introduces a tidal feature time weight, and its formula is expressed as:

[0021] θ attn (t) = α attn ·(sin(ωt + φ) + γ attn ·ΙΙ peak (t))

[0022] Among them, α attn is the coefficient that controls the magnitude of the attention weight, ω is the angular frequency of the tide, φ is the phase of the tide, γ attn is the coefficient that controls the influence of the tidal wave peak on the attention mechanism, ΙΙ peak (t) is the indicator function.

[0023] For a further technical solution, the grouting simulation model adopts an improved diffusion equation, which is expressed as:

[0024]

[0025] Among them, C = C(x, y, z, t) is the slurry concentration distribution, x, y, z are the spatial coordinates respectively, t is the time, D(t) is the time-varying diffusion coefficient under the action of tidal fluctuations, where $\nabla^2 C$ is the Laplacian operator of the concentration, and $S(t, C)$ is the source term.

[0026] In a further technical solution, the improved diffusion equation is discretized and solved by using the finite difference method.

[0027] In a further technical solution, the multi-objective optimization of the grouting parameters is realized by combining the genetic algorithm and the particle swarm optimization algorithm.

[0028] In a second aspect, the present invention provides a hydrodynamic grouting simulation system based on deep learning for coastal tidal action, including:

[0029] A data acquisition module, which is configured to: acquire tidal data, hydrogeological data, and grouting parameter data of the coastal area and perform preprocessing;

[0030] A network model prediction module, which is configured to: input the preprocessed tidal data, hydrogeological data, and grouting parameter data into a pre-trained deep learning model for prediction, and predict the dynamic changes of the grouting parameters under tidal action;

[0031] A physical model simulation module, which is configured to: establish a grouting simulation model of the formation under tidal action based on the diffusion equation, substitute the dynamic changes of the grouting parameters into the grouting simulation model for grouting simulation calculation to obtain the diffusion behavior of the slurry, and perform multi-objective optimization of the grouting parameters based on the diffusion behavior of the slurry to obtain the best combination of grouting parameters.

[0032] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in the hydrodynamic grouting simulation method based on deep learning for coastal tidal action described in the first aspect are implemented.

[0033] In a fourth aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the steps in the hydrodynamic grouting simulation method based on deep learning for coastal tidal action described in the first aspect are implemented.

[0034] The above one or more technical solutions have the following beneficial effects:

[0035] The present invention combines a deep learning model with a physical constraint model to simulate the grouting process under tidal conditions. First, the deep learning model predicts the dynamic changes of grouting parameters under tidal conditions, and inputs these changes into the physical constraint model. Combining with the hydrogeological characteristics of the formation, detailed grouting simulation calculations are carried out according to the diffusion equation to obtain the diffusion and penetration of the grout under different tidal conditions, as well as the original hydrogeological characteristics of the formation, and analyze the influence of hydrodynamic conditions on the grouting area. By comprehensively considering various indicators of the grouting effect (such as whether the diffusion range of the grout reaches the expectation, whether the penetration depth meets the reinforcement requirements, etc.), the long-term stability of the grouting area under different tidal conditions is evaluated. The evaluation results are provided to the user in the form of a report, providing a scientific basis for the user's decision-making in geotechnical engineering grouting construction in coastal areas.

[0036] The present invention constructs a deep learning model using a deep learning structure that combines a convolutional neural network, a long short-term memory network, and a self-attention mechanism, and improves the long short-term memory network and the self-attention mechanism considering tidal factors to predict the dynamic changes of grouting parameters under tidal action, realizing the modeling of tidal cycle fluctuations and the real-time response to occasional large fluctuations.

[0037] The present invention proposes a grouting simulation model based on the diffusion equation to describe the diffusion behavior of the grout under tidal fluctuation conditions. This model combines the time-varying diffusion coefficient of tidal action and the grouting source term, and accurately simulates the grouting diffusion process in a tidal dynamic environment by adjusting the boundary conditions in real time.

[0038] The present invention realizes multi-objective optimization by combining the genetic algorithm (GA) and the particle swarm optimization (PSO) to optimize the parameter configuration of grouting, including the grouting volume, grouting pressure, etc., to meet the parameter selection requirements under complex tidal conditions and obtain the best grouting parameter combination under tidal conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0040] Figure 1 is a flowchart of the grouting simulation method according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0041] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0042] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0044] Embodiment 1

[0045] As Figure 1 shown, this embodiment discloses a hydrodynamic grouting simulation method based on deep learning. The method includes the following steps:

[0046] S1: Obtain tidal data, hydrogeological data, and grouting parameter data in the coastal area and perform preprocessing;

[0047] In this embodiment, for tidal data collection: By setting up multiple tidal level observation stations and tidal current monitoring points in the coastal area, using sensors and other devices to collect data such as tidal level height, tidal current speed, direction, and period in real time. The collected data is regularly transmitted and summarized and stored in the data management module. In the data preprocessing stage, missing value processing is performed on the collected tidal data, such as using the interpolation method to fill in the missing data points; outliers are identified and corrected to ensure the accuracy of the data.

[0048] For hydrogeological data collection: Obtain hydrogeological parameters such as formation lithology, permeability, and porosity in the coastal area through means such as geological exploration and borehole sampling. These data are sorted into a standard format and stored in the data management module. In the data preprocessing stage, standardization processing is performed on the hydrogeological data to make its value range meet the requirements of deep learning model training.

[0049] For grouting parameter data collection: During the grouting construction process in the coastal area, record grouting parameters such as grouting volume, grouting speed, grouting pressure, and grouting viscosity. These data are transmitted to the data management module in real time. In the data preprocessing stage, normalization processing is performed on the grouting parameter data for easy calculation and processing during model training.

[0050] In the data preprocessing stage, data standardization is specifically: Standardize all continuous variable data to meet the requirements of model input. The formula is expressed as:

[0051]

[0052] where x is the original value of the data, μ is the data mean, σ is the data standard deviation, and x normalized is the standardized value of the data.

[0053] Anomaly detection uses Isolation Forest to detect anomalies by randomly partitioning samples to build multiple trees. The specific process is as follows:

[0054] (1) Randomly sample the dataset to generate multiple subsets.

[0055] (2) Build an "Isolation Tree" for each subset. In each tree, randomly select a feature and a split value each time to partition the samples.

[0056] (3) Calculate the average path length E(h(x)) of the sample. The shorter the path length, the more anomalous the sample.

[0057] The anomaly scoring formula of Isolation Forest is:

[0058]

[0059] where E(h(x)) is the average path length of sample x in the tree, n is the size of the dataset, and c(n) is a constant for normalization. The closer s(x,n) is to 1, the more anomalous the sample. E() is the mathematical expectation, which is the value obtained by averaging the path lengths of sample x in all isolation trees, and h(x) represents the path length calculation function of sample x in a certain isolation tree.

[0060] S2: Input the preprocessed tidal data, hydrogeological data, and grouting parameter data into a pre-trained deep learning model for prediction, and predict the dynamic changes of grouting parameters under tidal action.

[0061] In this embodiment, the deep learning model uses a deep learning structure that combines a Convolutional Neural Network (CNN), a Long Short-Term Memory Network (LSTM), and a Self-Attention Mechanism (Attention) to predict the changes in parameters during the grouting process, especially the dynamic changes of grouting parameters under tidal action, and realizes the modeling of tidal cycle fluctuations and the real-time response to occasional large fluctuations. Specifically, the deep learning model sequentially includes a CNN layer, an LSTM layer, and a Self-Attention Mechanism. Set parameters such as the convolutional kernel size and stride of the CNN layer to effectively extract spatial features; set parameters such as the number of neurons and time step of the LSTM layer to adapt to the processing of time series data; set the weight parameters of the Attention mechanism to focus on important feature information. Determine the number of neurons and connection methods of the input layer, hidden layer, and output layer to build a complete deep learning model architecture.

[0062] S201: Convolutional Neural Network (CNN): It is used for the preliminary processing of input data, especially for extracting spatial features related to the grouting environment (such as underground geological structures, soil layer properties, hydrogeological features, and spatial patterns of tidal fluctuations, etc.). The convolution operation can capture local patterns of geological features, reduce interference from irrelevant information, and provide a more compact feature representation for subsequent layers. The output of the CNN layer is the extracted spatial feature maps, and these feature maps, as the basic data of time series information, are input into the LSTM layer. The output of the CNN layer provides high-quality input features for the LSTM, making subsequent time series modeling more effective. Convolution operation:

[0063]

[0064] where i and j refer to the position indices in the generated feature map (i represents the i-th row, and j represents the j-th column), Filter mn is the convolution kernel, m represents the size of the convolution kernel in the horizontal direction, n represents the size of the convolution kernel in the vertical direction, σ is the activation function, b is the bias term, and Input is the original data input into the convolutional neural network.

[0065] S202: Long Short-Term Memory Network (LSTM): The LSTM focuses on time series analysis and can learn long-term and short-term dependencies in tidal data, which is particularly suitable for dealing with the time patterns of tidal cycle fluctuations and changes in grouting parameters. It captures the dependencies before and after in the time series through memory units to cope with the dynamic changes of grouting parameters under the influence of tides. And it predicts the change trend of grouting parameters through data dependencies in the time dimension, especially for accurately modeling tidal data under long cycles and short-time fluctuations. The output of the LSTM layer is a feature sequence containing time dependencies, which includes the time series information of tidal fluctuations and changes in grouting effects, providing candidate information that needs to be concerned for the Attention mechanism.

[0066] Introduce the tidal dynamic weight θ LSTM (t) represents the intensity and change of tidal fluctuations, reflecting the impact of tides on the diffusion and flow of slurry. The change intensity of tides may vary significantly at different time steps (such as high tide, low tide, or abnormal fluctuations). The intensity of tidal fluctuations changes with the tidal cycle and abnormal fluctuations. θ LSTM (t) can be modeled by the following formula:

[0067] θ LSTM (t) = α LSTM ·(sin(ωt + φ) + β LSTM ·ΙΙ abnormal (t))

[0068] where, α LSTMis the adjustment coefficient, used to control the impact of tidal fluctuations on memory; ω is the angular frequency of the tide, controlling the periodic change; φ is the initial phase of the tide, determining the starting position of the fluctuation; β LSTM is the influence coefficient of abnormal fluctuations, controlling the intensity of abnormal fluctuations on memory update; ΙΙ abnormal (t) is the indicator function, when the time t is in the abnormal tidal fluctuation, ΙΙ abnormal (t)=1, otherwise it is 0.

[0069] Furthermore, introduce the periodic parameter p t indicating the periodic change of tidal fluctuations, reflecting the impact of the peak and trough periods of tidal fluctuations on the grouting process, and adjusting the memory and update of the LSTM network by combining tidal cycle and phase information. The periodic parameter is modeled by the following formula:

[0070] p t =sin(ωt + φ)

[0071] where ω is the angular frequency of the tide, controlling the periodic change; t is the time, and φ is the initial phase of the tide, determining the starting position of the fluctuation.

[0072] Furthermore, bring the two introduced parameters, namely the tidal dynamic weight and the periodic parameter, into the LSTM to enhance the LSTM's modeling ability for tidal fluctuations. These new variables can help the LSTM dynamically adjust its memory update and forgetting process during the grouting process, especially during periods with large tidal fluctuation intensities. The new LSTM formula is as follows:

[0073] f t =σ(W f ·[h t-1 ,x t +b f +γ f ·θ LSTM (t))

[0074] i t =σ(W i ·[h t-1 ,x t +b i +γ i ·θ LSTM (t))

[0075]

[0076]

[0077] o t =σ(W o ·[h t-1 ,x t +bo +γ o ·θ LSTM (t))

[0078] h t =o t ·tanh(C t )

[0079] Among them, f t is the forget gate, i t is the input gate, is the candidate memory cell state, C t is the memory cell state, o t is the output gate, γ f 、γ i 、γ C 、γ o are coefficients used to adjust the influence of tides on the forget, input, candidate memory, and output gates. σ is the activation function, W f 、W i 、W C 、W o are the weight matrices corresponding to the forget gate, input gate, candidate memory cell state, and output gate respectively. h t is the hidden state at the current time step t, h t-1 is the hidden state at the previous time step before the current time step t, x t is the input feature at the current time step t, b f 、b i 、b C 、b o are the bias vectors corresponding to the forget gate, input gate, candidate memory cell state, and output gate respectively. θ LSTM (t) is the tidal dynamic weight, p t is the periodic parameter.

[0080] By introducing the tidal dynamic weight and the periodic parameter, the improved LSTM model can dynamically adjust the memory and update of input features at different stages of tidal fluctuations. Especially during tidal peaks, troughs, and abnormal fluctuation periods, the model can pay more attention and adjust the prediction of grouting parameters according to the tidal intensity. At high tide, θ LSTM (t) will increase, and the input and state updates of the model will be more concentrated to cope with possible strong tidal changes; during low tide or abnormal fluctuations, the periodic parameter p t also plays a regulatory role to ensure that the model can capture the influence of these critical moments and avoid error accumulation.

[0081] S203: Attention Mechanism: The Attention mechanism is used to weight and focus on specific time steps or features in the time series to capture key data segments highly relevant to the grouting effect, especially the abnormally large fluctuations or key features in tidal fluctuations. By assigning a weight to each time step of the LSTM output, it focuses on the time points that have a greater impact on the grouting effect (such as abnormal fluctuations or tidal peak points), thereby improving the model's performance in large fluctuation scenarios. This weighted attention helps the model be more accurate in dealing with abnormal tidal fluctuations. After being processed by the Attention mechanism, the model generates a weighted feature representation, which is more focused on the features of key time steps, making the model more targeted and accurate when predicting the grouting effect.

[0082] Similarly, the periodicity and abnormality of tidal fluctuations are further introduced, and a tidal feature time weight θ attn (t) is added, which is used to adjust the attention allocation at specific tidal periods. The tidal feature time weight θ attn (t) is dynamically generated according to the intensity and periodic changes of tidal fluctuations. The specific formula is as follows:

[0083] θ attn (t) = α attn ·(sin(ωt + φ) + γ attn ·ΙΙ peak (t))

[0084] Among them, α attn is the coefficient that controls the magnitude of the attention weight; ω is the angular frequency of the tide, which controls the periodicity of the tide; φ is the phase of the tide, which determines the starting moment of the tidal fluctuation; ΙΙ peak (t) is the indicator function. When the time t is at the peak of the tide, ΙΙ peak (t) = 1, otherwise it is 0; γ attn is the coefficient that controls the influence of the tidal wave peak on the attention mechanism.

[0085] Through this design, the model can adaptively adjust the attention weight of each time step according to the periodicity and abnormal fluctuation characteristics of the tide. When the tidal fluctuation reaches the peak or an anomaly occurs, the weight will increase accordingly, strengthening the model's attention to these key periods.

[0086] Furthermore, after introducing the tidal feature time weight, the improved attention mechanism formula is as follows:

[0087]

[0088] Among them, e t is the attention score at the current time step t; h t-1 is the hidden state of the previous time step before the current time step t; xt is the input feature at the current time step (such as tidal level, flow velocity, etc.); θ attn (t) is the tidal feature time weight, reflecting the influence degree of tidal fluctuation on the grouting process; T represents vector transpose, w a is the weight of the attention score; W h 、W x 、W θ are the parameter matrices to be learned.

[0089] Here, by introducing θ attn (t), the model can dynamically adjust the attention allocation at each time step according to the characteristics of tidal fluctuations. Especially during the peak period or abnormal fluctuations of tides, it can strengthen the attention to important moments.

[0090] Furthermore, according to the score e t , calculate the attention weight α t , which is expressed as:

[0091]

[0092] where e t is the attention score at the current time step t, and α t represents the attention weight of the model at time step t. e l is the attention score for all time steps l during calculation, and l is an index representing the set of all time steps.

[0093] Due to the introduction of the tidal feature time weight θ attn (t), the model can automatically allocate higher attention during the period when tidal fluctuations are most significant (such as the high tide period of tides).

[0094] Furthermore, by weighting the output features, generate the weighted feature representation c as the input for the grouting process simulation:

[0095]

[0096] where h t is the output of the LSTM layer, representing the hidden state at time step t; α t is the attention weight at time step t.

[0097] In this way, the model will pay more attention to the time steps when tidal fluctuations are most significant (such as tidal wave peaks, wave troughs, or sudden abnormal fluctuations), thereby improving the prediction accuracy and stability.

[0098] The pre - processed historical data (including tidal data, hydro - geological data, and grouting parameter data) are divided into different data sets according to a certain ratio (such as 80% training set, 10% validation set, and 10% test set). The training set data are used to train the constructed deep - learning model. Training parameters such as an appropriate learning rate (such as 0.001) and the number of training epochs (such as 100 epochs) are set. The stochastic gradient descent algorithm is used as the training algorithm, and the weight parameters of the model are adjusted by minimizing the mean - square error loss function. During the training process, every certain number of epochs (such as 10 epochs), the performance of the model is monitored using the validation set data. When the accuracy on the validation set no longer improves or signs of over - fitting appear, hyperparameters such as the learning rate or the number of neurons in the hidden layer are adjusted for optimization. After multiple rounds of training and optimization, the test set data are used to evaluate the finally optimized model to ensure that the model has good generalization ability.

[0099] When it is necessary to predict the dynamic changes of grouting parameters, the pre - processed historical data (including tidal data, hydro - geological data, and grouting parameter data) and the set initial values of grouting parameters (determined according to the experience of previous similar coastal grouting projects) are prepared as the input data of the model. These data are input into the trained deep - learning model. The model predicts the changes of grouting parameters over time under this tidal condition according to the rules it has learned, including whether the grouting volume needs to be adjusted in a timely manner, how the grouting speed should be changed, etc., to adapt to the periodic changes of tides and possible abnormal conditions (such as large tidal fluctuations), and provides grouting optimization suggestions (such as adjusting grouting rate, pressure, etc.) under abnormal tidal conditions.

[0100] S3: Establish a grouting simulation model of the formation under the action of tides based on the diffusion equation, substitute the dynamic changes of the grouting parameters into the grouting simulation model for grouting simulation calculation, and obtain the diffusion behavior of the slurry.

[0101] In this embodiment, under the action of tides, the grouting process is affected by various dynamic factors, including tidal water - level changes, flow - velocity fluctuations, and slurry diffusion characteristics, etc. Therefore, the present invention innovatively proposes a grouting simulation model based on the diffusion equation to describe the diffusion behavior of the slurry under tidal fluctuation conditions, specifically including: slurry diffusion range, slurry penetration depth, slurry flow path, slurry flow - velocity changes, etc. This model combines the time - varying diffusion coefficient of tidal action and the grouting source term, and accurately simulates the grouting diffusion process in the tidal dynamic environment by adjusting the boundary conditions in real - time.

[0102] S301: Physical constraint model based on the diffusion equation (grouting simulation model)

[0103] In traditional diffusion models, the diffusion process of slurry usually depends on the diffusion coefficient of the substance. However, under the action of tides, the grouting process is affected by factors such as tidal water level and flow velocity changes, resulting in dynamic changes in the diffusion coefficient and flow behavior. To more accurately simulate this process, the present invention adopts the following improved diffusion equation:

[0104]

[0105] where C = C(x, y, z, t) is the slurry concentration distribution, x, y, and z are the spatial coordinates respectively, t is the time; D(t) is the time-varying diffusion coefficient under the action of tidal fluctuations, representing the speed of flow and slurry diffusion caused by tides; is the Laplace operator of the concentration, representing the spatial change of the concentration; S(t, C) is the source term, representing the grouting source or other disturbance sources caused by tides.

[0106] Furthermore, tidal fluctuations will dynamically change the diffusion coefficient. The mathematical description of the time-varying diffusion coefficient D(t) is as follows:

[0107] D(t) = D 0 ·(1 + α·sin(ωt + φ))

[0108] where D 0 is the reference diffusion coefficient, the diffusion value without tidal fluctuations; α is the tidal fluctuation intensity coefficient; ω is the tidal angular frequency, related to the tidal period; φ is the phase of the tidal fluctuation, used to reflect the initial state of the tide.

[0109] This formula reflects the periodic modulation of tidal fluctuations on the slurry diffusion speed, enabling the diffusion model to flexibly adapt to the tidal dynamic environment.

[0110] Furthermore, the source term S(t, C) is used to describe the characteristics of the slurry injection process, including the grouting rate, attenuation behavior, etc. Its expression is:

[0111] S(t, C) = γ·C in (t) - β·C

[0112] where γ is the grouting rate coefficient; C in (t) is the grouting source concentration, representing the slurry concentration at the injection point; β is the slurry attenuation coefficient, describing the concentration reduction of the slurry due to mixing, sedimentation, etc. during the diffusion process; C = C(x, y, z, t) is the slurry concentration distribution, x, y, and z are the spatial coordinates respectively, t is the time.

[0113] The source term model combined with the tidal fluctuation conditions can accurately reflect the kinetic characteristics of slurry injection and diffusion.

[0114] Furthermore, in a tidal fluctuation environment, the diffusion equation needs to incorporate boundary conditions and initial conditions. Boundary conditions: provided by the tidal prediction results generated by the deep learning model (dynamic changes in grouting parameters), expressed as: Where represents the predicted change in grouting parameters. Initial condition: The concentration distribution at the initial moment of grouting is set by the grouting plan C = C(x, y, z, 0).

[0115] S302: Since the improved diffusion equation is a partial differential equation (PDE), it needs to be solved numerically. In the present invention, the finite difference method (FDM) is used to discretize and solve the diffusion equation.

[0116] (1) Spatial and temporal discretization

[0117] Let the spatial grid points be x i , y j , z k , and the time step be Δt. Then the discretized form of the diffusion equation is:

[0118]

[0119] Where is the concentration at the j-th time step and the i-th spatial point; is the Laplace operator in discretized form, expressed according to the finite difference method as:

[0120] (2) Time stepping

[0121] Explicit or implicit methods are used for time stepping.

[0122] Explicit method:

[0123] Implicit method (more suitable for larger time steps):

[0124]

[0125] (3) Real-time adjustment

[0126] After each time step calculation, the diffusion coefficient D(t) and the source term S(t, C) are dynamically updated in combination with the dynamic changes in the grouting parameters generated by the deep learning model to ensure that the simulation results can reflect the tidal dynamic changes.

[0127] S303: Combine the deep learning model and the physical constraint model for mutual optimization, including the deep learning model outputting the boundary conditions and diffusion coefficients of the physical constraint model, and the physical constraint model feeding back the training data of the deep learning model. Specifically, the deep learning model, i.e., the CNN+LSTM+Attention model, generates the dynamic changes of the grouting parameters for real-time updating of the boundary conditions and diffusion coefficients of the diffusion equation; the physical constraint model, i.e., the solution result of the diffusion equation, can be used for the training of the deep learning model to further optimize the prediction of the influence of tidal fluctuations on the grouting process.

[0128] S304: Achieve multi-objective optimization by combining the genetic algorithm (GA) and particle swarm optimization (PSO) to optimize the parameter configuration of grouting, including grouting volume, grouting pressure, etc., to meet the parameter selection requirements under complex tidal conditions. GA is used to generate the initial population, and the diversity of solutions is expanded through selection, crossover, and mutation operations, while PSO performs local search and accelerates convergence on the initial solutions provided by GA to find the optimal combination of grouting parameters.

[0129] (1) Objective function

[0130] F(X) = η 1 · Grouting effect - η 2 · Cost + η 3 · Fluctuation adaptability

[0131] Among them, X is the parameter to be optimized, including parameters such as grouting volume, pressure, and flow rate; the grouting effect represents the optimization goal of the grouting effect, such as penetration depth or flow rate; the cost represents the cost control goal, including grouting material and time costs; the fluctuation adaptability is a measure of the robustness of the grouting parameters under different tidal fluctuation conditions. η 1 、η 2 、η 3 respectively represent the weight coefficients used to balance the relationships among the grouting effect, cost, and fluctuation adaptability.

[0132] (2) Optimization algorithm

[0133] Genetic algorithm (GA): Generate multiple initial candidate solutions (different combinations of grouting parameters), and calculate the fitness based on the simulation results of the physical equations. Expand the search range through selection, crossover, and mutation operations to find better solutions, generate the next generation of candidate solutions, and the final population is passed to PSO for further optimization.

[0134] Offspring = α · Parent 1 + (1 - α) · Parent 2

[0135] Among them, α is the crossover coefficient, usually set to 0.5.

[0136] Particle Swarm Optimization (PSO): Particle Swarm Optimization (PSO) simulates the collaborative search process of a group. Each solution is regarded as a "particle" and is updated based on the individual's historical best position and the global best position. PSO is used for local search on the population provided by GA to accelerate convergence to the optimal solution. The formula is expressed as:

[0137]

[0138] where ω is the inertia weight, controlling the particle's dependence on the previous velocity; c 1 、c 2 are learning factors, adjusting the influence of individual cognition and group cognition; r 1 、r 2 are uniformly distributed random numbers, used to increase search diversity; represents the velocity of the i-th particle at time step t + 1, i is the index of the particle, used to distinguish different particle individuals, t represents the time step, p i represents the individual historical best position of particle i, represents the velocity of the i-th particle at time step t, x represents the particle, g represents the global best position of the group, that is, the optimal solution position found by the entire particle swarm during the search process.

[0139] S305: Real-time monitor the parameters of the grouting process (such as grouting pressure, slurry flow rate, etc.) through sensors, and adjust the grouting strategy in combination with the output of numerical simulation, that is, the optimal grouting parameter combination obtained after numerical simulation optimization.

[0140] The main simulation process is as follows:

[0141] (1) Real-time data collection: Collect real-time data from sensors and input it into the deep learning model.

[0142] (2) Feedback adjustment: Compare the optimal parameters predicted by the deep learning model with the real-time data feedback by the sensors. If there is a significant deviation, adjust the parameters.

[0143] (3) Update and display: Apply the adjusted parameters to the grouting equipment and display the simulation results in the form of charts in real time.

[0144] The present invention uses a deep learning model to predict the dynamic changes of grouting parameters. Based on historical data and the laws learned by the model, it estimates the possible changes in grouting parameters over time under tidal action. It mainly aims to reflect how grouting parameters such as grouting volume, grouting speed, and grouting pressure change over time during tidal fluctuations, helping to understand the dynamic characteristics of the grouting process. However, this is not necessarily the optimal parameter combination. The best grouting parameter combination obtained by using genetic algorithm (GA) and particle swarm optimization (PSO) for multi-objective optimization in S304 is obtained by comprehensively considering various factors such as grouting effect (such as whether the slurry diffusion range reaches the expectation, whether the penetration depth meets the reinforcement requirements, etc.), cost (grouting material and time cost), and fluctuation adaptability (the robustness of grouting parameters under different tidal fluctuation conditions). This stage is to further adjust and screen the parameters through an optimization algorithm based on the prediction of the deep learning model.

[0145] Embodiment 2

[0146] This embodiment discloses a hydrodynamic grouting simulation system based on deep learning for coastal tidal action, including:

[0147] A data acquisition module, which is configured to: acquire tidal data, hydrogeological data, and grouting parameter data of the coastal area and perform preprocessing;

[0148] A network model prediction module, which is configured to: input the preprocessed tidal data, hydrogeological data, and grouting parameter data into a pre-trained deep learning model for prediction, and predict the dynamic changes of grouting parameters under tidal action;

[0149] A physical model simulation module, which is configured to: establish a grouting simulation model of the formation under tidal action based on the diffusion equation, substitute the dynamic changes of the grouting parameters into the grouting simulation model for grouting simulation calculation to obtain the diffusion behavior of the slurry, and perform multi-objective optimization of the grouting parameters based on the diffusion behavior of the slurry to obtain the best grouting parameter combination.

[0150] Embodiment 3

[0151] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps of the method in Embodiment 1.

[0152] Embodiment 4

[0153] The purpose of this embodiment is to provide a computer-readable storage medium. A computer-readable storage medium stores a computer program, and when the program is executed by a processor, it executes the steps of the method in Embodiment 1.

[0154] In the devices of the above-mentioned Embodiments 3 and 4, the steps involved correspond to those of Method Embodiment 1. For the specific implementation manners, reference may be made to the relevant description part of Embodiment 1. The term "computer-readable storage medium" should be understood to include a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to execute any method in the present invention.

[0155] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0156] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0157] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A method for simulating dynamic water grouting under coastal tidal action based on deep learning, characterized in that: include: Obtain tidal data, hydrogeological data and grouting parameter data in coastal areas and perform preprocessing; The pre-processed tidal data, hydrogeological data and grouting parameter data are input into the pre-trained deep learning model for prediction, and the dynamic changes of grouting parameters under the action of tides are predicted; A grouting simulation model of the stratum under tidal action is established based on the diffusion equation, and the dynamic changes of the grouting parameters are substituted into the grouting simulation model for grouting simulation calculation to obtain the diffusion behavior of the slurry. Based on the diffusion behavior of the slurry, multi-objective optimization of the grouting parameters is performed to obtain the optimal grouting parameter combination.

2. The method for simulating dynamic water grouting based on coastal tidal action based on deep learning as claimed in claim 1, characterized in that: The deep learning model includes a convolutional neural network layer, an improved long short-term memory network layer and an improved self-attention mechanism in sequence.

3. The method for simulating dynamic water grouting based on coastal tidal action based on deep learning as claimed in claim 2, characterized in that: The improved long short-term memory network layer enhances the modeling capability of tidal fluctuations by introducing tidal dynamic weights and periodic parameters, and its formula is expressed as follows: f t =σ(W f ·[h t-1 ,x t ]+b f +g f ·i LSTM (t)) I t =σ(W i ·[h t-1 ,x t ]+b i +g i ·i LSTM (t)) the t =σ(W o ·[h t-1 ,x t ]+b o +g o ·i LSTM (t)) h t =o t ·tanh(C t ) Among them, f t For the forget gate, i t is the input gate, is the candidate memory cell state, C t is the memory unit state, o t is the output gate, γ f , γ i , γ C , γ o is the coefficient used to adjust the effect of tide on forgetting, input, candidate memory and output gates, σ is the activation function, W f , W i , W C , W o They are the weight matrices corresponding to the forget gate, input gate, candidate memory unit state and output gate, respectively. t is the hidden state at the current time step t, h t-1 is the hidden state of the previous time step t, x t is the input feature of the current time step t, b f , b i , b C , b o are the bias vectors corresponding to the forget gate, input gate, candidate memory cell state, and output gate, respectively, θ LSTM (t) is the tidal dynamic weight, p t is a periodic parameter.

4. The method for simulating dynamic water grouting based on coastal tidal action based on deep learning as claimed in claim 2, characterized in that: The improved self-attention mechanism introduces the tidal feature time weight, which is expressed as follows: i attn (t)=a attn ·(sin(ωt+φ)+γ attn ·II peak (t)) Among them, α attn is the coefficient that controls the attention weight, ω is the angular frequency of the tide, φ is the phase of the tide, and γ attn is the coefficient that controls the effect of the tidal peak on the attention mechanism, ΙΙ peak (t) is the indicator function.

5. The method for simulating dynamic water grouting based on coastal tidal action based on deep learning as claimed in claim 1, characterized in that: The grouting simulation model adopts an improved diffusion equation, which is expressed as: Where C = C (x, y, z, t) is the slurry concentration distribution, x, y, z are spatial coordinates, t is time, D (t) is the time-varying diffusion coefficient under the action of tidal fluctuations, ▽ 2 C is the Laplace operator of concentration, and S(t,C) is the source term.

6. The method for simulating dynamic water grouting based on coastal tidal action based on deep learning as claimed in claim 5, characterized in that: The improved diffusion equation is discretized and solved using the finite difference method.

7. The method for simulating dynamic water grouting based on coastal tidal action based on deep learning as claimed in claim 1, characterized in that: Multi-objective optimization of grouting parameters is achieved by combining genetic algorithm and particle swarm optimization algorithm.

8. A dynamic water grouting simulation system for coastal tidal action based on deep learning, characterized in that: include: A data acquisition module is configured to: acquire tidal data, hydrogeological data and grouting parameter data of the coastal area and perform preprocessing; The network model prediction module is configured to: input the pre-processed tidal data, hydrogeological data and grouting parameter data into a pre-trained deep learning model for prediction, and predict the dynamic changes of grouting parameters under the action of tides; The physical model simulation module is configured as follows: a grouting simulation model of the stratum under tidal action is established based on the diffusion equation, the dynamic changes of the grouting parameters are substituted into the grouting simulation model to perform grouting simulation calculations to obtain the diffusion behavior of the slurry, and multi-objective optimization of the grouting parameters is performed based on the diffusion behavior of the slurry to obtain the best grouting parameter combination.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for simulating dynamic water grouting based on deep learning of coastal tidal action are implemented as described in any one of claims 1-7.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for simulating dynamic water grouting based on deep learning of coastal tidal action are implemented as described in any one of claims 1-7.

Images