Cloud computing roadbed construction slope displacement monitoring system

Abstract

The invention relates to the field of construction engineering, in particular to a roadbed construction slope displacement monitoring system based on cloud computing. A sensor network is used for collecting roadbed construction data in a roadbed, wavelet transform is used for denoising the data, a dynamic time warping algorithm in a cloud computing unit is used for aligning time sequence data in the initial roadbed construction data, and the weight of each sensor data is dynamically distributed based on an entropy weight method. A hybrid prediction model is established based on an LSTM long short-term memory network and FEA finite element analysis, a slope stress field result simulated by FEA is used as a physical constraint layer of the LSTM, an attention layer is embedded in the physical constraint layer, hyper-parameters of the initial hybrid prediction model are optimized by using a Cs-Ant improved cuckoo-ant colony combinatorial algorithm, and a slope stress field is obtained. And inputting the feature roadbed construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction. The manual analysis cost is effectively reduced, and the level of roadbed engineering full-period monitoring is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering construction, in particular to a subgrade construction slope displacement monitoring system based on cloud computing. Background Art

[0002] Traditional slope displacement monitoring methods usually rely on manual measurement, or a single total station sensor or GPS sensor for periodic data collection. The following defects usually exist in traditional measurement methods: poor real-time performance, continuous monitoring cannot be achieved by manual inspection or fixed equipment, and sudden displacements are difficult to detect in a timely manner; data processing lags, the efficiency of local storage and analysis is low, and it is impossible to handle massive data; the early warning ability is insufficient, relying on static threshold alarms, lacking a dynamic risk assessment model; high cost, when deploying a sensor network on a large scale, the hardware and maintenance costs are high. In the prior art, some patents attempt to combine wireless sensor networks or remote data transmission, but do not solve the problems of data fusion analysis, dynamic prediction, and cloud collaboration. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a subgrade construction slope displacement monitoring system based on cloud computing.

[0004] Furthermore, in order to achieve the above purpose, the technical solution of the present invention is as follows. In the above-mentioned subgrade construction slope displacement monitoring system based on cloud computing, the subgrade construction slope displacement monitoring system includes the following modules: A slope data acquisition module, which is used to collect subgrade construction data in the subgrade by using a sensor network, denoise the data by using wavelet transform, and correct the data error by using an adaptive Kalman filtering algorithm to obtain initial subgrade construction data; A cloud feature processing module, which is used to align the time series data in the initial subgrade construction data by using the DTW dynamic time warping algorithm in the cloud computing unit, and dynamically allocate the weights of each sensor data based on the entropy weight method to obtain characteristic subgrade construction data; A prediction model establishment module, which is used to establish a FEA-LSTM hybrid prediction model based on the LSTM long short-term memory network and FEA finite element analysis, use the slope stress field result simulated by FEA as the physical constraint layer of LSTM, and embed an attention layer in LSTM to obtain an initial FEA-LSTM hybrid prediction model; A model parameter optimization module, which is used to optimize the hyperparameters of the initial FEA-LSTM hybrid prediction model by using the Cs-Ant improved cuckoo-ant colony combination algorithm to obtain a target Cs-FEA-LSTM hybrid prediction model; The slope displacement prediction module is used to input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction, obtain the slope displacement trend, and perform hierarchical early warning and decision support on the slope displacement trend based on the cloud computing unit.

[0005] Further, in the above slope displacement monitoring system for subgrade construction with cloud computing, the slope data acquisition module includes the following sub-modules: The acquisition sub-module is used to collect displacement data, inclination data, humidity data, and microseismic data in the subgrade at a sampling frequency of 10HZ by using a sensor network to obtain subgrade construction data; The transmission sub-module is used to transmit the subgrade construction data to the cloud computing unit for data preprocessing; The processing sub-module is used to denoise the collected data by using wavelet transform, decompose the original signal into high-frequency noise components and low-frequency trend components, and perform soft threshold processing on the high-frequency components of the displacement signal for power frequency interference and sensor drift noise by using the threshold filtering method to obtain the first subgrade construction data; The establishment sub-module is used to establish the linear state equation and observation equation of the first subgrade construction data through the adaptive Kalman filtering algorithm, and set the sensor noise as Gaussian white noise to obtain the second subgrade construction data; The fusion sub-module is used to dynamically adjust the process noise and observation noise covariance matrix of the second subgrade construction data according to the residual sequence, fuse the multi-sensor filtering results, and generate a displacement correction value with high confidence through weighted average to obtain the initial subgrade construction data.

[0006] Further, in the above slope displacement monitoring system for subgrade construction with cloud computing, the cloud feature processing module includes the following sub-modules: The data definition sub-module is used to align the time series data in the initial subgrade construction data by using the DTW dynamic time warping algorithm, and calculate the Euclidean distance matrix of multiple sensor data in the initial subgrade construction data Defined as: ; Among them, and respectively represent the time series data in the sensor, represents the th time series data of one of the sensors, represents the th time series data of another sensor, and are the lengths of the corresponding sensor time series data respectively; The distance calculation sub-module is used to solve the Euclidean distance matrix by using dynamic programming Minimum cumulative distance , and the recurrence formula is: ; Among them, the recurrence formula represents the minimum cumulative distance from the upper left corner position to position, represents the row index corresponding to the row above the current position in the minimum cumulative distance matrix ; represents the column index corresponding to the column to the left of the current position in the minimum cumulative distance matrix ; Data alignment sub-module, used for initialization conditions , representing the starting point of the minimum cumulative distance matrix whose value is equal to the element value at the position in the Euclidean distance matrix , backtracking from to to generate an alignment path , making the time sequence points in the sensor aligned according to the minimum distance to obtain aligned subgrade construction data; Data segmentation sub-module, used for dynamically allocating the weights of the data of each sensor based on the entropy weight method, normalizing the time sequence data in the aligned subgrade construction data by the range method for data standardization, and segmenting the standardized data by using a sliding window to obtain characteristic subgrade construction data.

[0007] Furthermore, in the above-mentioned subgrade construction slope displacement monitoring system of cloud computing, the prediction model establishment module includes the following sub-modules: Establishment sub-module, used for establishing an FEA-LSTM hybrid prediction model based on the LSTM long short-term memory network and FEA finite element analysis, obtaining historical subgrade construction data in the database, and training the prediction model with the data; LSTM sub-module, used for using the forget gate of the LSTM long short-term memory network in the model to control the retention ratio of historical memories, using the input gate to determine the update amount of new information, using the candidate memory unit to generate new information to be stored, and using the memory unit to update and fuse historical and current information; using the output gate to control the output of the current hidden state; Obtaining sub-module, used for obtaining the input data of FEA finite element analysis, at least including slope geometric parameters, material parameters and external loads, where the slope geometric parameters include slope height and slope gradient, the material parameters include elastic modulus, Poisson's ratio and cohesion, and the external loads at least include construction vibration intensity and rainfall infiltration pressure; Extraction sub-module, used for obtaining the internal stress tensor field of the slope by solving the equation ​ , extract the equivalent plastic strain of the node and the principal stress direction as physical constraint features: ; Among them, represents the output vector of the FEA finite element analysis simulation, including physical constraint features, represents the equivalent plastic strain, represents the th equivalent plastic strain data; in represents the principal stress direction, and the subscript , , respectively represent the principal stress directions in the , , directions; represents the Euclidean space where the vector is located, represents the dimension of the vector ; The correction sub-module is used to use the output of the FEA simulation as the input of the LSTM and correct the memory cell update.

[0008] Furthermore, in the above-mentioned subgrade construction slope displacement monitoring system of cloud computing, the prediction model establishment module further includes the following sub-modules: The embedding sub-module is used to embed the attention mechanism in the LSTM, and use the attention mechanism to calculate the weights of the data, including query vector, key vector, attention score and normalized weight calculation; The aggregation sub-module is used to generate a context vector by weighted aggregation of historical hidden states based on the attention mechanism, and output after splicing the context vector with the current hidden state: ; Among them, represents the hidden state output after being processed by the attention mechanism, represents the hyperbolic tangent activation function, represents the weight matrix, which is used to perform a linear transformation on the spliced vector; represents the hidden state of the LSTM model at the current moment , represents the context vector generated by weighted aggregation of historical hidden states through the attention mechanism, represents the bias term.

[0009] Further, in the above slope displacement monitoring system for subgrade construction in cloud computing, the model parameter optimization module includes the following units: A parameter acquisition unit, configured to acquire a hyperparameter combination of a prediction model, at least including the number of LSTM hidden layers, the LSTM learning rate, the FEA grid density, and the attention mechanism dimension, encode the hyperparameter combination into a solution vector, and define a solution space; A global exploration unit, configured to perform global exploration using CS cuckoo search, and generate a new solution by Levy flight: ; Wherein, represents the step size factor, obeys the Lévy distribution, and the step size ; represents the current optimal solution, represents the generated new solution, represents the current solution, represents element-wise multiplication, is a parameter of the Lévy distribution; A pheromone update unit, configured to update the pheromone using the ACO ant colony combination algorithm, and the ants release the pheromone concentration on the path : ; Wherein, represents the pheromone evaporation coefficient; represents the pheromone increment, represents a constant, represents the fitness value of the solution, represents at the pheromone concentration on the path at time, represents at the pheromone concentration on the path at time; A parameter adjustment unit, configured to optimize the hyperparameter combination using the Cs-Ant improved cuckoo-ant colony combination algorithm, widely search within the solution space using Levy flight to generate a candidate solution set, construct a path graph for the candidate solution set, the ants search along the path with a high pheromone concentration, and adaptively adjust the hyperparameter combination according to the number of iterations and population diversity to obtain the target Cs-FEA-LSTM hybrid prediction model.

[0010] Further, in the above slope displacement monitoring system for subgrade construction in cloud computing, the slope displacement prediction module includes the following units: A displacement prediction unit, which is used to input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction, output the slope displacement prediction values and their confidence intervals within the next 6 to 24 hours, and generate a displacement rate index and an acceleration index at the same time; A grading unit, which is used to divide four-level early warnings according to the displacement prediction values, rates and environmental factors in the cloud computing unit, including at least a normal level, a caution level, a warning level and an emergency level.

[0011] Furthermore, in the method for realizing the above-mentioned slope displacement monitoring system for subgrade construction in cloud computing, the method includes the following steps: Collect subgrade construction data in the subgrade by using a sensor network, denoise the data by using wavelet transform, and correct the data error by using an adaptive Kalman filtering algorithm to obtain initial subgrade construction data; Use the DTW dynamic time warping algorithm in the cloud computing unit to align the time series data in the initial subgrade construction data, and dynamically allocate the weights of each sensor data based on the entropy weight method to obtain characteristic subgrade construction data; Establish an FEA-LSTM hybrid prediction model based on the LSTM long short-term memory network and FEA finite element analysis. Use the slope stress field results simulated by FEA as the physical constraint layer of LSTM, and embed an attention layer in LSTM to obtain an initial FEA-LSTM hybrid prediction model; Use the Cs-Ant improved cuckoo-ant colony combination algorithm to optimize the hyperparameters of the initial FEA-LSTM hybrid prediction model to obtain the target Cs-FEA-LSTM hybrid prediction model; Input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction to obtain the slope displacement trend, and perform hierarchical early warning and decision support on the slope displacement trend based on the cloud computing unit.

[0012] Furthermore, in the above-mentioned slope displacement monitoring method for subgrade construction in cloud computing, the method includes the following steps: Embed an attention mechanism in LSTM, and use the attention mechanism to calculate the weights of the data, including query vector, key vector, attention score and normalized weight calculation; Generate a context vector by weighted aggregation of historical hidden states based on the attention mechanism, and splice the context vector with the current hidden state and then output: ; Among them, represents the hidden state output after being processed by the attention mechanism, represents the hyperbolic tangent activation function, denotes a weight matrix for performing a linear transformation on the concatenated vector; represents the LSTM model at the current moment of the hidden state, represents the context vector generated by weighted aggregation of historical hidden states through the attention mechanism, denotes the bias term.

[0013] Furthermore, in the above method for monitoring slope displacement during subgrade construction in cloud computing, the obtaining of voice data in the real-time environment through the microphone array includes: Inputting the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction, outputting the slope displacement prediction values and their confidence intervals within the next 6 to 24 hours, and simultaneously generating a displacement rate index and an acceleration index; Dividing four-level early warnings according to the displacement prediction values, rates, and environmental factors in the cloud computing unit, including at least a normal level, a caution level, a warning level, and an emergency level.

[0014] Its beneficial effects are as follows: By obtaining multi-dimensional construction data in real time through the sensor network, combining wavelet transform denoising and adaptive Kalman filtering to correct errors, effectively eliminating the interference of environmental noise and equipment errors, ensuring the high reliability of the original data, and laying a precise data foundation for subsequent analysis. It can adaptively focus on key construction parameters (such as stress, displacement, etc.), significantly improving the characterization ability of feature data and avoiding information distortion caused by traditional fixed weights. Secondly, in the FEA-LSTM, the hybrid prediction model combines the physical mechanism of finite element analysis (FEA) and the data-driven advantages of long short-term memory network (LSTM). By embedding a physical constraint layer of the slope stress field, engineering mechanics laws are integrated into the deep learning process, which not only retains the ability of LSTM for prediction to capture time series features but also avoids the "black box" defect of a pure data model, improving the physical interpretability of the prediction results. The introduction of the attention layer further strengthens the model's attention to key spatio-temporal features, significantly improving the prediction accuracy under complex working conditions. The Cs-Ant prediction improvement algorithm realizes intelligent hyperparameter optimization. Compared with a single optimization algorithm, its combined strategy has both global search ability and local fine-tuning advantages, greatly shortening the model training period and improving the generalization performance. The cloud computing-based hierarchical early warning and decision support system can process a large amount of monitoring data in real time and output a visualized displacement trend, and combine with the engineering safety threshold to achieve multi-level risk early warning, providing an efficient and intelligent decision-making basis for construction safety control, effectively reducing the manual analysis cost, and improving the automation and scientific level of the whole-cycle monitoring of subgrade engineering. Description of the Drawings

[0015] Various other advantages and benefits will become clear to those of ordinary skill in the art by reading the following detailed description of the preferred embodiments. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0016] Figure 1 FIG. 1 is a schematic diagram of the first embodiment of a slope displacement monitoring system for subgrade construction in cloud computing according to an embodiment of the present invention; Figure 2 FIG. 2 is a schematic diagram of the second embodiment of a slope displacement monitoring system for subgrade construction in cloud computing according to an embodiment of the present invention; Figure 3 FIG. 3 is a schematic diagram of the third embodiment of a slope displacement monitoring system for subgrade construction in cloud computing according to an embodiment of the present invention. Detailed Embodiments

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will be specifically described below with reference to the accompanying drawings. As Figure 1 shown, a slope displacement monitoring system for subgrade construction in cloud computing includes the following modules: 101. Slope data acquisition module, which is used to collect subgrade construction data in the subgrade by using a sensor network, denoise the data by using wavelet transform, and correct the data error through an adaptive Kalman filtering algorithm to obtain initial subgrade construction data; Specifically, this embodiment further includes an acquisition sub-module, which is used to collect displacement data, inclination data, humidity data, and microseismic data in the subgrade by using a sensor network at a sampling frequency of 10HZ to obtain subgrade construction data; A transmission sub-module, which is used to transmit the subgrade construction data to the cloud computing unit for data preprocessing; A processing sub-module, which is used to denoise the collected data by using wavelet transform, decompose the original signal into high-frequency noise components and low-frequency trend components, and perform soft threshold processing on the high-frequency components of the displacement signal for power frequency interference and sensor drift noise by using the threshold filtering method to obtain the first roadbed construction data; An establishment sub-module, which is used to establish a linear state equation and an observation equation of the first roadbed construction data by using the adaptive Kalman filtering algorithm, set the sensor noise as Gaussian white noise, and obtain the second roadbed construction data; A fusion sub-module, which is used to dynamically adjust the process noise and observation noise covariance matrix of the second roadbed construction data according to the residual sequence, fuse the multi-sensor filtering results, and generate a displacement correction value with high confidence through weighted averaging to obtain the initial roadbed construction data.

[0020] 102. A cloud feature processing module, which is used to align the time series data in the initial roadbed construction data by using the DTW dynamic time warping algorithm in the cloud computing unit, and dynamically allocate the weights of the data of each sensor based on the entropy weight method to obtain the feature roadbed construction data; Specifically, this embodiment further includes a data definition sub-module, which is used to align the time series data in the initial roadbed construction data by using the DTW dynamic time warping algorithm, and calculate the Euclidean distance matrix of the data of multiple sensors in the initial roadbed construction data Defined as: ; Among them, and respectively represent the time series data in the sensors, represents the th time series data of one of the sensors, represents the th time series data of another sensor, and are the lengths of the corresponding sensor time series data respectively; A distance calculation sub-module, which is used to solve the minimum cumulative distance of the Euclidean distance matrix by using dynamic programming , and the recurrence formula is: ; Among them, the recurrence formula represents the minimum cumulative distance from the upper left corner position of the matrix to the position, represents the row index corresponding to the row above the current position in the minimum cumulative distance matrix; represents the column index corresponding to the column to the left of the current position in the minimum cumulative distance matrix; The data alignment sub-module is used for initialization conditions , representing the starting point of the minimum cumulative distance matrix whose value is equal to the element value at the position in the Euclidean distance matrix in . Backtracking from in reverse to to generate an alignment path , making the time series points in the sensor aligned according to the minimum distance to obtain the aligned subgrade construction data; The data segmentation sub-module is used to dynamically allocate the weights of the data of each sensor based on the entropy weight method, normalize the time series data in the aligned subgrade construction data using the range method, and segment the normalized data using a sliding window to obtain the characteristic subgrade construction data.

[0021] 103. The prediction model establishment module is used to establish an FEA-LSTM hybrid prediction model based on the LSTM long short-term memory network and FEA finite element analysis. The slope stress field results simulated by FEA are used as the physical constraint layer of LSTM, and an attention layer is embedded in LSTM to obtain the initial FEA-LSTM hybrid prediction model; Specifically, this embodiment further includes an establishment sub-module for establishing an FEA-LSTM hybrid prediction model based on the LSTM long short-term memory network and FEA finite element analysis, obtaining the historical subgrade construction data in the database, and training the prediction model using the data; The LSTM sub-module is used to use the forget gate of the LSTM long short-term memory network in the model to control the retention ratio of historical memories, use the input gate to determine the update amount of new information, use the candidate memory unit to generate new information to be stored, and use the memory unit to update and fuse historical and current information; use the output gate to control the output of the current hidden state; The acquisition sub-module is used to acquire the input data of FEA finite element analysis, including at least slope geometric parameters, material parameters, and external loads. The slope geometric parameters include slope height and slope gradient, the material parameters include elastic modulus, Poisson's ratio, and cohesion, and the external loads include at least construction vibration intensity and rainfall infiltration pressure; The extraction sub-module is used to obtain the internal stress tensor field of the slope by solving the equation , and extract the equivalent plastic strain of the nodes and the principal stress direction as physical constraint features: ; ; Among them, represents the output vector simulated by FEA finite element analysis, including physical constraint features, represents the equivalent plastic strain, Indicates the equivalent plastic strain data; in indicates the principal stress direction, and the subscript , , respectively indicate the principal stress directions in , , directions; represents the Euclidean space where the vector is located, represents the dimension of the vector ; The correction sub-module is used to take the output of the FEA simulation as the input of the LSTM and correct the memory cell update.

[0022] The embedding sub-module is used to embed the attention mechanism in the LSTM, calculate the weights of the data using the attention mechanism, including query vector, key vector, attention score and normalized weight calculation; The aggregation sub-module is used to generate a context vector by weighted aggregation of historical hidden states based on the attention mechanism, and output after concatenating the context vector with the current hidden state: ; Among them, represents the hidden state output after being processed by the attention mechanism, represents the hyperbolic tangent activation function, represents the weight matrix, which is used to perform a linear transformation on the concatenated vector; represents the hidden state of the LSTM model at the current time , represents the context vector generated by weighted aggregation of historical hidden states through the attention mechanism, represents the bias term.

[0023] 104. The model parameter optimization module is used to optimize the hyperparameters of the initial FEA-LSTM hybrid prediction model by using the Cs-Ant improved cuckoo-ant colony combination algorithm to obtain the target Cs-FEA-LSTM hybrid prediction model; Specifically, this embodiment further includes a parameter acquisition unit, which is used to obtain the hyperparameter combination of the prediction model, at least including the number of LSTM hidden layers, the LSTM learning rate, the FEA grid density and the attention mechanism dimension, encode the hyperparameter combination as a solution vector, and define the solution space; The global exploration unit is used to perform global exploration using CS cuckoo search, and Levy flight generates a new solution: ; Among them, represents the step size factor, obeys the Lévy distribution, and the step size ; represents the current optimal solution, represents the newly generated solution, represents the current solution, represents element-wise multiplication, is the parameter of the Lévy distribution; The pheromone update unit is used to update the pheromone using the ACO ant colony combination algorithm. Ants release pheromone concentration on the path ; : ; Among them, represents the pheromone evaporation coefficient; represents the pheromone increment, represents a constant, represents the fitness value of the solution, represents at time, the pheromone concentration on the path ; represents at time, the pheromone concentration on the path ; The parameter adjustment unit is used to optimize the hyperparameter combination using the Cs-Ant improved cuckoo-ant colony combination algorithm, widely search the solution space using Levy flight to generate a candidate solution set, construct a path graph for the candidate solution set, ants search along the path with high pheromone concentration, and adaptively adjust the hyperparameter combination according to the number of iterations and population diversity to obtain the target Cs-FEA-LSTM hybrid prediction model.

[0024] 105. The slope displacement prediction module is used to input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction to obtain the slope displacement trend, and perform hierarchical early warning and decision support on the slope displacement trend based on the cloud computing unit.

[0025] Specifically, this embodiment further includes a displacement prediction unit, which is used to input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction, output the slope displacement prediction value and its confidence interval within the next 6 to 24 hours, and generate a displacement rate index and an acceleration index at the same time; The grade division unit is used to divide four-level early warnings in the cloud computing unit according to the displacement prediction value, rate and environmental factors, including at least the normal level, the attention level, the warning level and the emergency level.

[0026] Normal (green): Displacement rate ≤ threshold 1, predicted trend is stable, no external risk factors; Caution for prediction (blue): Displacement rate ∈ (threshold 1, threshold 2 for prediction], or predicted displacement increase ≥ 5% in the next 3 hours; Warning for prediction (yellow): Displacement rate ∈ (threshold 2, threshold 3 for prediction], or rainfall exceeds the warning value and predicted displacement increase ≥ 10%; Emergency for prediction (red): Displacement rate > threshold 3, or microseismic signal continues to increase, predicted slip risk is extremely high.

[0027] Its beneficial effects are as follows. By obtaining multi-dimensional construction data in real time through a sensor network, combining wavelet transform denoising and adaptive Kalman filtering to correct errors, it effectively eliminates the interference of environmental noise and equipment errors, ensures the high reliability of the original data, and lays a precise data foundation for subsequent analysis. It can adaptively focus on key construction parameters (such as stress, displacement, etc.), significantly improving the characterization ability of feature data and avoiding information distortion caused by traditional fixed weights. Secondly, the FEA-LSTM hybrid prediction model in prediction fuses the physical mechanism of finite element analysis FEA and the data-driven advantages of long short-term memory network LSTM. By embedding a physical constraint layer of the slope stress field, it integrates engineering mechanics laws into the deep learning process, not only retaining the ability of LSTM in prediction to capture time series features, but also avoiding the "black box" defect of pure data models, and improving the physical interpretability of prediction results. The introduction of the attention layer further strengthens the model's attention to key spatio-temporal features, significantly improving the prediction accuracy under complex working conditions. The Cs-Ant improved algorithm in prediction realizes intelligent hyperparameter optimization. Compared with a single optimization algorithm, its combined strategy has both global search ability and local fine-tuning advantages, greatly shortening the model training cycle and improving the generalization performance. The cloud computing-based hierarchical early warning and decision support system can process massive monitoring data in real time and output a visualized displacement trend, realizing multi-level risk early warning in combination with engineering safety thresholds, providing an efficient and intelligent decision-making basis for construction safety control, effectively reducing the cost of manual analysis, and improving the automation and scientific level of the whole-cycle monitoring of subgrade engineering.

[0028] In this embodiment, please refer to Figure 2 , the second embodiment of a subgrade construction slope displacement monitoring system based on cloud computing in the embodiment of the present invention. The cloud feature processing module includes the following sub-modules: The data definition sub-module is used to align the time series data in the initial subgrade construction data by using the DTW dynamic time warping algorithm, and calculate the Euclidean distance matrix of multiple sensor data in the initial subgrade construction data Defined as: ; Among them, and respectively represent the time series data in the sensor, Represents the th time series data of one of the sensors, represents the th time series data of another sensor, and are the lengths of the time series data of the corresponding sensors respectively; Distance calculation sub-module, used to solve the minimum cumulative distance of the Euclidean distance matrix using dynamic programming , and the recurrence formula is: ; Among them, the recurrence formula represents the minimum cumulative distance from the upper left corner position of the matrix to position, represents the row index corresponding to the row above the current position in the minimum cumulative distance matrix; represents the column index corresponding to the column to the left of the current position in the minimum cumulative distance matrix; Data alignment sub-module, used to initialize the condition , which represents the starting point of the minimum cumulative distance matrix , and its value is equal to the element value at the position in the Euclidean distance matrix . Backtracking from to , generating an alignment path , making the time series points in the sensor aligned according to the minimum distance to obtain the aligned subgrade construction data; Data segmentation sub-module, used to dynamically allocate the weights of the data of each sensor based on the entropy weight method, normalize the time series data in the aligned subgrade construction data using the range method, and segment the normalized data using a sliding window to obtain the characteristic subgrade construction data.

[0029] Its beneficial effect is that it can adaptively focus on key construction parameters (such as stress, displacement, etc.), significantly improve the characterization ability of feature data, and avoid information distortion caused by traditional fixed weights.

[0030] In this embodiment, please refer to Figure 3 , the third embodiment of a subgrade construction slope displacement monitoring system based on cloud computing in the embodiments of the present invention. The model parameter optimization module includes the following sub-units: Parameter acquisition unit, used to obtain the hyperparameter combination of the prediction model, at least including the number of LSTM hidden layers, LSTM learning rate, FEA grid density, and attention mechanism dimension, encode the hyperparameter combination into a solution vector, and define the solution space; Global exploration unit, which is used to perform global exploration using CS cuckoo search and generate new solutions using Levy flight: ; Among them, represents the step size factor, obeys the Lévy distribution, and the step size ; represents the current optimal solution, represents the generated new solution, represents the current solution, represents element-wise multiplication, is the parameter of the Lévy distribution; Pheromone update unit, which is used to update the pheromone using the ACO ant colony combination algorithm. Ants release the pheromone concentration on the path : ; Among them, represents the pheromone evaporation coefficient; represents the pheromone increment, represents a constant, represents the fitness value of the solution, represents at the pheromone concentration on the path at time, represents at the pheromone concentration on the path at time; Parameter adjustment unit, which is used to optimize the hyperparameter combination using the Cs-Ant improved cuckoo-ant colony combination algorithm, widely search in the solution space using Levy flight to generate a candidate solution set, construct a path graph for the candidate solution set, ants search along the path with high pheromone concentration, and adaptively adjust the hyperparameter combination according to the number of iterations and population diversity to obtain the target Cs-FEA-LSTM hybrid prediction model.

[0031] Its beneficial effect is that the prediction improvement algorithm in Cs-Ant realizes intelligent optimization of hyperparameters. Compared with a single optimization algorithm, its combination strategy has both global search ability and local fine-tuning advantages, greatly shortening the model training cycle and improving the generalization performance.

[0032] ​The basic principles, main features and advantages of the present invention have been shown and described above. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A subgrade construction slope displacement monitoring system for cloud computing, characterized in that, The subgrade construction slope displacement monitoring system includes the following modules: The slope data acquisition module is used to collect subgrade construction data in the subgrade by using a sensor network, denoise the data by using wavelet transform, and correct the data error through an adaptive Kalman filtering algorithm to obtain initial subgrade construction data; The cloud feature processing module is used to align the time series data in the initial subgrade construction data by using the DTW (Dynamic Time Warping) algorithm in the cloud computing unit, and dynamically allocate the weights of each sensor data based on the entropy weight method to obtain characteristic subgrade construction data; The prediction model establishment module is used to establish an FEA-LSTM hybrid prediction model based on the LSTM (Long Short-Term Memory) network and FEA (Finite Element Analysis). The slope stress field results simulated by FEA are used as the physical constraint layer of LSTM, and an attention layer is embedded in LSTM to obtain an initial FEA-LSTM hybrid prediction model; The model parameter optimization module is used to optimize the hyperparameters of the initial FEA-LSTM hybrid prediction model by using the Cs-Ant improved cuckoo-ant colony combination algorithm to obtain a target Cs-FEA-LSTM hybrid prediction model; The slope displacement prediction module is used to input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction to obtain the slope displacement trend, and perform hierarchical early warning and decision support on the slope displacement trend based on the cloud computing unit.

2. The slope displacement monitoring system for subgrade construction in cloud computing according to claim 1, characterized in that, The slope data acquisition module includes the following sub-modules: The acquisition sub-module is used to collect displacement data, inclination data, humidity data, and microseismic data in the subgrade by using a sensor network at a sampling frequency of 10HZ to obtain subgrade construction data; The transmission sub-module is used to transmit the subgrade construction data to the cloud computing unit for data preprocessing; The processing sub-module is used to denoise the collected data by using wavelet transform to obtain the first subgrade construction data; The establishment sub-module is used to establish a linear state equation and an observation equation of the first subgrade construction data through an adaptive Kalman filtering algorithm, and set the sensor noise as Gaussian white noise to obtain the second subgrade construction data; The fusion sub-module is used to dynamically adjust the process noise and observation noise covariance matrix of the second subgrade construction data according to the residual sequence to obtain the initial subgrade construction data.

3. The slope displacement monitoring system for subgrade construction in cloud computing according to claim 1, wherein The cloud feature processing module includes the following sub-modules: The data definition sub-module is used to align the time series data in the initial subgrade construction data by using the DTW algorithm and calculate the Euclidean distance in multiple sensor data in the initial subgrade construction data; A distance calculation sub-module, which is used to solve the minimum cumulative distance of the Euclidean distance matrix by using dynamic programming ; A data alignment sub-module for initializing conditions , representing the starting point of the minimum cumulative distance matrix , whose value is equal to the element value at the position in the Euclidean distance matrix . Backtracking from in reverse to to generate an alignment path , making the timing points in the sensor aligned according to the minimum distance to obtain the aligned subgrade construction data; The data segmentation sub-module is used to dynamically allocate the weights of each sensor data based on the entropy weight method, normalize the time series data in the aligned subgrade construction data by using the range method, and segment the normalized data by using a sliding window to obtain characteristic subgrade construction data.

4. The slope displacement monitoring system for subgrade construction in cloud computing according to claim 1, characterized in that, The prediction model establishment module includes the following sub-modules: A sub-module is established to build an FEA-LSTM hybrid prediction model based on the LSTM long short-term memory network and FEA finite element analysis, obtain historical subgrade construction data in the database, and use the data to train the prediction model; The LSTM sub-module is used to use the forget gate of the LSTM long short-term memory network in the model to determine the retention ratio of historical memories, use the input gate to determine the update amount of new information, use the candidate memory unit to generate new information to be stored, and use the memory unit to update and fuse historical and current information; use the output gate to control the output of the current hidden state; The acquisition sub-module is used to acquire the input data of FEA finite element analysis, including at least slope geometric parameters, material parameters and external loads. The slope geometric parameters include slope height and slope gradient. The material parameters include elastic modulus, Poisson's ratio and cohesion. The external loads include at least construction vibration intensity and rainfall infiltration pressure; The extraction sub-module is used to obtain the internal stress tensor field of the slope by solving equations; The correction sub-module is used to use the output of the FEA simulation as the input of the LSTM and correct the memory unit update.

5. The slope displacement monitoring system for subgrade construction in cloud computing according to claim 1, characterized in that, The prediction model building module further includes the following sub-modules: The embedding sub-module is used to embed the attention mechanism in the LSTM and calculate the weights of the data using the attention mechanism, including query vector, key vector, attention score and normalized weight calculation; The aggregation sub-module is used to generate a context vector by weighted aggregation of historical hidden states based on the attention mechanism, and splice and output the context vector with the current hidden state.

6. The slope displacement monitoring system for subgrade construction in cloud computing according to claim 1, wherein, The model parameter optimization module includes the following units: The parameter acquisition unit is used to obtain the hyperparameter combination of the prediction model, including at least the number of LSTM hidden layers, the LSTM learning rate, the FEA mesh density and the attention mechanism dimension, encode the hyperparameter combination as a solution vector, and define the solution space; The global exploration unit is used to use CS cuckoo search for global exploration and generate new solutions by Levy flight; The pheromone update unit is used to update the pheromone using the ACO ant colony combination algorithm, and the ants release pheromone concentration on the path; The parameter adjustment unit is used to optimize the hyperparameter combination using the Cs-Ant improved cuckoo-ant colony combination algorithm to obtain the target Cs-FEA-LSTM hybrid prediction model.

7. The slope displacement monitoring system for subgrade construction in cloud computing according to claim 1, characterized in that, The slope displacement prediction module includes the following units: The displacement prediction unit is used to input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction, output the slope displacement prediction value and its confidence interval within the next 6 to 24 hours, and generate a displacement rate index and an acceleration index at the same time; The grade division unit is used to divide four-level early warnings according to the displacement prediction value, rate and environmental factors in the cloud computing unit, including at least normal level, attention level, warning level and emergency level.

8. A method for implementing a slope displacement monitoring system for subgrade construction in cloud computing as described in claim 1, characterized in that, The method includes the following steps: Use the sensor network to collect subgrade construction data in the subgrade, denoise the data using wavelet transform, and correct the data error through the adaptive Kalman filter algorithm to obtain the initial subgrade construction data; Align the time series data in the initial subgrade construction data using the DTW (Dynamic Time Warping) algorithm in the cloud computing unit, and dynamically allocate the weights of each sensor data based on the entropy weight method to obtain the characteristic subgrade construction data; Establish an FEA-LSTM hybrid prediction model based on the LSTM (Long Short-Term Memory) network and FEA (Finite Element Analysis). Use the slope stress field results simulated by FEA as the physical constraint layer of LSTM, and embed an attention layer in LSTM to obtain the initial FEA-LSTM hybrid prediction model; Use the Cs-Ant improved cuckoo-ant colony combination algorithm to optimize the hyperparameters of the initial FEA-LSTM hybrid prediction model to obtain the target Cs-FEA-LSTM hybrid prediction model; Input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction to obtain the slope displacement trend, and perform hierarchical early warning and decision support on the slope displacement trend based on the cloud computing unit.

9. A method for implementing a slope displacement monitoring system for subgrade construction in cloud computing as described in claim 1, characterized in that, The method includes the following steps: Embed an attention mechanism in LSTM, and use the attention mechanism to calculate the weights of the data, including query vector, key vector, attention score, and normalized weight calculation; Generate a context vector by weighted aggregation of historical hidden states based on the attention mechanism, and splice the context vector with the current hidden state and then output.

10. A method for implementing a subgrade construction slope displacement monitoring system for cloud computing as described in claim 1, characterized in that, The method includes the following steps: Input the characteristic subgrade construction data into the target Cs-FEA-LSTM hybrid prediction model for prediction, output the slope displacement prediction values and their confidence intervals within 6 to 24 hours in the future, and simultaneously generate displacement rate indicators and acceleration indicators; Divide into four levels of early warning in the cloud computing unit according to the displacement prediction values, rates, and environmental factors, including at least normal level, attention level, warning level, and emergency level.

Images