Roadside slope stability prediction method and device, storage medium and electronic equipment

Abstract

The application provides a road slope stability prediction method and device, a storage medium and an electronic equipment, which comprises the following steps: collecting multi-dimensional environmental parameters of a road slope in real time to form an original parameter set; analyzing the original parameter set based on a Granger causality test algorithm to determine the causal influence relationship between parameters, construct a parameter causal network, and screen out key parameters; calculating the similarity of each key parameter in the time and space dimensions through a dynamic time warping algorithm to generate a space-time correlation matrix; weighting and fusing the key parameters based on the space-time correlation matrix to obtain a fused parameter; inputting the fused parameter into a pre-trained slope stability prediction model to obtain a predicted value of slope stability; and outputting a slope instability warning information when the predicted value exceeds a dynamically updated warning threshold. In the application, the defects of inaccurate prediction and data redundancy of the current road slope stability prediction method are overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a road slope stability prediction method and device, a storage medium and an electronic equipment. BACKGROUND

[0002] The road slope stability is directly related to the road traffic safety and the safety of the surrounding area personnel and property, and timely and accurate prediction of the slope stability is of great significance for preventing geological disasters such as landslides and collapses.

[0003] The existing road slope stability prediction method mainly has the following deficiencies: in the aspect of data processing, the original monitoring data is simply used or only the conventional filtering and normalization processing is performed, the potential relationship between the multi-dimensional environmental parameters is not fully mined, and the actual influence of the parameters on the slope stability cannot be accurately reflected. At the same time, the key factors that truly affect the slope stability cannot be effectively identified, resulting in data redundancy and waste of computing resources.

[0004] In addition, the threshold setting of the existing prediction model usually adopts a fixed value or a simple adjustment method based on experience, which cannot adapt to the dynamic changes of the slope stability under different geological conditions and climate environments, and is prone to false positives or false negatives.

[0005] Therefore, the current road slope stability prediction method has the defects of inaccurate prediction and data redundancy. SUMMARY

[0006] The main purpose of the present application is to provide a road slope stability prediction method and device, a storage medium and an electronic equipment, which aims to overcome the defects of inaccurate prediction and data redundancy of the current road slope stability prediction method.

[0007] To achieve the above purpose, the present application provides a road slope stability prediction method, comprising the following steps:

[0008] Real-time collection of multi-dimensional environmental parameters of the road slope forms an original parameter set;

[0009] Based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal relationship between the parameters, a parameter causal network is constructed, and the key parameters affecting the slope stability are screened out;

[0010] The similarity of each key parameter in the time and space dimensions is calculated by the dynamic time warping algorithm to generate a spatio-temporal correlation matrix;

[0011] Based on the spatio-temporal correlation matrix, the key parameters are weighted and fused to obtain a fused parameter;

[0012] inputting the fusion parameters into a pre-trained slope stability prediction model to obtain a predicted value of the slope stability;

[0013] updating a warning threshold of the slope stability prediction model dynamically, and outputting a slope instability warning information when the predicted value exceeds the dynamically updated warning threshold.

[0014] Further, the environmental parameters include geological parameters, mechanical parameters, environmental parameters and displacement parameters.

[0015] Further, when the key parameters are fused, weights are assigned based on the correlation degree on the spatio-temporal correlation matrix, and the fusion parameters are generated through matrix point multiplication operation.

[0016] Further, based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal influence relationship between parameters, a parameter causal network is constructed, and key parameters affecting slope stability are screened out, including:

[0017] The gray correlation degree of each parameter in the original parameter set and the historical data of slope stability is calculated, and each parameter is preliminarily sorted according to the gray correlation degree, and the Granger causality test is performed on the parameters with high correlation degree first;

[0018] When performing Granger causality test, based on the variable point detection algorithm, the structural mutation points of parameters in time series are identified, and the lag order of Granger causality test is set for different mutation intervals;

[0019] Based on the causal relationship strength evaluation model, the causal relationship strength scores of each parameter pair are calculated; the parameter pairs with causal relationship strength scores higher than a threshold value are retained to construct a parameter causal network;

[0020] The Louvain algorithm is used to divide the parameter causal network into communities, calculate the cohesion coefficient of each community and the contribution degree to slope stability; the core parameters in the community with a contribution degree higher than a set value and the hub parameters connected across communities are selected as the key parameters affecting slope stability.

[0021] Further, based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal influence relationship between parameters, a parameter causal network is constructed, and key parameters affecting slope stability are screened out, including:

[0022] The original parameter set is subjected to three-dimensional convolution through a three-dimensional convolution kernel to obtain a spatio-temporal parameter set with enhanced features;

[0023] A dynamic time window division method is used to adaptively adjust the time window length according to the parameter fluctuation frequency, and the spatio-temporal parameter set is segmented to obtain a plurality of parameter subsequences;

[0024] Granger causality tests are performed on each of the parameter subsequences, and the weights of the test results for each parameter subsequence are calculated using an attention mechanism. The weighted fusion is then used to obtain a comprehensive causal relationship matrix.

[0025] Based on the recurrent neural network, the evolution of the comprehensive causal relationship matrix over time is learned, and the changing trend of the causal influence relationship of each parameter in the future period is predicted.

[0026] Based on the comprehensive causal relationship matrix and the changing trend, a parametric causal network with a time dimension is constructed. The temporal importance of each node in the parametric causal network is calculated, and parameters with temporal importance higher than a preset value are selected as key parameters affecting slope stability.

[0027] Furthermore, using a dynamic time warping algorithm, the similarity of each key parameter in time and space is calculated to generate a spatiotemporal correlation matrix, including:

[0028] A convolutional neural network is used to extract the local features of the key parameters in the spatial dimension, and then concatenate them into spatial sequence features;

[0029] Long Short-Term Memory (LSTM) networks are used to capture the long-term dependency features of the key parameters in the time dimension and concatenate them into time series features.

[0030] The time series features are mapped to quantum states. By using a dynamic time warping algorithm combined with quantum parallel computing of multiple time warping paths, the similarity of each time warping path is evaluated, and the optimal matching path is selected to obtain a time dimension similarity matrix.

[0031] Based on a pre-built spatial similarity evaluation model, the spatial sequence features are analyzed to generate a spatial dimension similarity matrix;

[0032] The spatiotemporal correlation matrix is ​​obtained by fusing the temporal similarity matrix and the spatial similarity matrix through tensor product operation.

[0033] Furthermore, the early warning threshold of the slope stability prediction model is dynamically updated, including:

[0034] Based on historical prediction data, the early warning threshold of the slope stability prediction model is dynamically updated using a sliding window algorithm.

[0035] The present invention also provides a road slope stability prediction device, comprising:

[0036] The data acquisition unit is used to collect multi-dimensional environmental parameters of the roadside slope in real time and form a raw parameter set.

[0037] The screening unit is used for analyzing the original parameter set, determining the causal influence relationship among parameters, constructing a parameter causal network and screening out key parameters affecting the slope stability based on a Granger causality test algorithm;

[0038] The association unit is used for calculating the similarity of each key parameter in the time and space dimensions by a dynamic time warping algorithm and generating a time-space association matrix;

[0039] The fusion unit is used for weighting and fusing the key parameters based on the time-space association matrix to obtain a fusion parameter;

[0040] The prediction unit is used for inputting the fusion parameter into a pre-trained slope stability prediction model to obtain a prediction value of the slope stability;

[0041] The early warning unit is used for dynamically updating an early warning threshold of the slope stability prediction model, and outputting slope instability early warning information when the prediction value exceeds the dynamically updated early warning threshold.

[0042] The application further provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of the preceding embodiments when executing the computer program.

[0043] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the method according to any one of the preceding embodiments when executed by a processor.

[0044] The application provides a road slope stability prediction method and device, a storage medium and an electronic equipment. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 FIG. 1 is a schematic diagram of a road slope stability prediction method according to an embodiment of the application;

[0046] Figure 2 FIG. 2 is a structural block diagram of a road slope stability prediction device according to an embodiment of the application;

[0047] Figure 3 FIG. 3 is a structural schematic block diagram of an electronic equipment according to an embodiment of the application.

[0048] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0049] In order to make the object, technical scheme and advantages of the application clearer, the application 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 application and not used to limit the application.

[0050] With reference to Figure 1 An embodiment of the application provides a road slope stability prediction method, which comprises the following steps:

[0051] Step S1, real-time collection of multi-dimensional environmental parameters of a road slope to form an original parameter set;

[0052] Step S2, based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal relationship between parameters, construct a parameter causal network, and screen out key parameters affecting slope stability;

[0053] Step S3, by dynamic time warping algorithm, the similarity of each key parameter in time and space dimensions is calculated to generate a spatiotemporal correlation matrix;

[0054] Step S4, based on the spatiotemporal correlation matrix, the key parameters are weighted and fused to obtain a fusion parameter;

[0055] Step S5, the fusion parameter is input into the pre-trained slope stability prediction model to obtain the prediction value of slope stability;

[0056] Step S6, dynamically update the early warning threshold of the slope stability prediction model, when the prediction value exceeds the dynamically updated early warning threshold, output the slope instability early warning information.

[0057] In this embodiment, as described in step S1 above, by deploying various sensors such as displacement sensors, humidity sensors, pressure sensors, and weather sensors on the road slope, multi-dimensional environmental parameters are collected in real time. The above environmental parameters include geomechanics parameters (such as rock-soil density, shear strength), hydrological parameters (groundwater level, rainfall), meteorological parameters (wind speed, air temperature), slope shape parameters (crack width, slope angle), etc. Multi-dimensional data collection can fully reflect the environmental state and physical characteristics of the slope, providing rich and accurate data sources for subsequent analysis. The collected data directly constitutes the original parameter set, which contains a large amount of original, unprocessed information, retains the authenticity and integrity of the data, but also has problems such as data redundancy and complex relationships between parameters, which need to be further processed in subsequent steps.

[0058] As described in the above step S2, based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal relationship between parameters, a parameter causal network is constructed, and key parameters affecting the slope stability are screened out. The core of this step is to mine the causal relationship between parameters and screen key parameters. The Granger causality test algorithm is a statistical method based on time series. Its principle is to judge whether there is a causal relationship between two variables by testing whether the historical information of one variable can improve the prediction accuracy of the other variable. The Granger causality test is performed on each parameter combination in the original parameter set to calculate the causal influence degree between parameters. According to the test results, the parameters with causal relationship are connected by edges to construct a parameter causal network. In this network, nodes represent environmental parameters, edges represent causal relationships between parameters, and the weight of edges can reflect the strength of causal influence. By analyzing the network structure and the connection of parameters, those parameters that are in a key position in the network and have a significant impact on slope stability are screened out. These key parameters play a major role in the change of slope stability, and subsequent analysis is focused on these parameters, which can effectively reduce the data processing amount and improve the analysis efficiency and prediction accuracy.

[0059] As described in the above step S3, the similarity of each key parameter in time and space dimensions is calculated by the dynamic time warping algorithm to generate a spatio-temporal correlation matrix. This step aims to mine the internal relationship of key parameters in time and space dimensions. The dynamic time warping algorithm is originally used for similarity measurement of time series data. By bending the time axis, two time series of different lengths or speeds are aligned to calculate their similarity. In this scheme, it is applied to expand the application, considering both time and space dimensions. In the time dimension, the monitoring data sequences of key parameters are arranged in chronological order, and the dynamic time warping algorithm is used to calculate the similarity of different parameter time series. In the spatial dimension, based on the spatial position relationship of slope monitoring points, the spatial distribution characteristics of parameters are taken into account in the calculation, such as adjusting the similarity measurement by spatial distance, topological structure, etc. The similarity calculation results in time and space dimensions are integrated to generate a matrix, i.e. a spatio-temporal correlation matrix. Each element in the matrix represents the similarity of two key parameters in time and space dimensions, which fully reflects the correlation characteristics of key parameters in space and time, and provides a basis for subsequent parameter fusion.

[0060] As described in step S4, the key parameters are weighted and fused based on the spatiotemporal correlation matrix to obtain a fused parameter. This step realizes the fusion of key parameters according to the spatiotemporal correlation matrix. The spatiotemporal correlation matrix reflects the mutual relationship of each key parameter in the time-space dimension. According to the similarity degree between parameters in the matrix, each key parameter is assigned a corresponding weight. Parameters with high similarity and close correlation are given a higher weight, indicating that they play an important role in reflecting the stability of the slope; otherwise, the weight is lower. By weighted summation or other methods, the key parameters are fused to obtain a fused parameter. The fused parameter integrates the spatiotemporal information of the key parameters, and can more comprehensively and comprehensively reflect the state of the slope than the original parameters, reduce information redundancy, enhance the representativeness and effectiveness of the data, and provide better input data for the subsequent prediction model.

[0061] As described in step S5, the fused parameter is input into a pre-trained slope stability prediction model to obtain a predicted value of the slope stability. This step uses the trained prediction model to process the fused parameter to obtain the prediction result. The pre-trained slope stability prediction model can be a machine learning model (such as neural network, support vector machine, etc.) or a deep learning model. In the model training stage, a large number of sample data containing historical environmental parameters and corresponding slope stability states are used, and the model parameters are continuously adjusted by optimization algorithm, so that the model can learn the mapping relationship between environmental parameters and slope stability. The fused parameter is input into the trained model, and the model analyzes and calculates the input data according to the learned rule, and outputs a quantitative value, which is the predicted value of the slope stability. The predicted value reflects the stability of the slope under the current environmental parameters, and provides an important reference for the evaluation of slope stability.

[0062] As described in step S6, the warning threshold of the slope stability prediction model is dynamically updated, and when the predicted value exceeds the dynamically updated warning threshold, the slope instability warning information is output. This step realizes the dynamic optimization and warning function of the prediction model. The environmental conditions of the slope are complex and changeable, and it is difficult for a fixed warning threshold to adapt to the evaluation needs of slope stability under different working conditions. Therefore, according to real-time monitoring data, historical slope instability conditions and geological environmental changes, etc., a dynamic algorithm is used to update the warning threshold of the prediction model. For example, the time series analysis method can be used to predict the future environmental change trend, and the critical value of slope instability under different environmental conditions in the historical data is combined to dynamically adjust the warning threshold. When the predicted value exceeds the dynamically updated warning threshold, it indicates that the slope is in a high-risk state of instability, and the system immediately outputs the slope instability warning information to remind relevant personnel to take appropriate preventive and handling measures, thereby realizing real-time and effective monitoring and warning of the stability of the road slope, and ensuring the safety of road traffic and the surrounding area.

[0063] In the embodiment, by analyzing the original parameter set, the causal relationship between parameters is determined, a parameter causal network is constructed, key parameters affecting the slope stability are screened out, the similarity of each key parameter in the time and space dimensions is calculated, a space-time correlation matrix is generated, and the warning threshold of the slope stability prediction model is dynamically updated; the potential relationship between multi-dimensional environmental parameters is fully tapped, and the defects of inaccurate prediction and data redundancy of current road slope stability prediction methods are overcome.

[0064] In an embodiment, the environmental parameters include geological parameters, mechanical parameters, environmental parameters, and displacement parameters.

[0065] In the embodiment, the geological parameters are key data reflecting the geological structure and composition characteristics of the road slope. They cover information such as rock-soil type, rock layer thickness, geological structure (such as the location and characteristics of faults and folds), etc. Rock-soil type determines the basic physical properties of the slope, such as the significant differences in shear strength and permeability of different rock-soils such as clay, sand, and rock, which directly affect the stability of the slope. Rock layer thickness and geological structure reveal the geological history and potential weak links of the slope, and the rock mass near the fault is poor in integrity and easy to become a sliding surface, and the shape and trend of the fold also change the stress distribution. Accurate acquisition of these geological parameters can provide a basic geological background for subsequent analysis of the stability of the slope and help to judge the congenital stability conditions of the slope.

[0066] Mechanical parameters are used to describe the mechanical behavior and bearing capacity of the rock-soil mass of the slope. They mainly include the density, elastic modulus, Poisson's ratio, cohesion, and internal friction angle of the rock-soil mass. The density of the rock-soil mass affects its own gravity and in turn affects the size of the sliding force; the elastic modulus and Poisson's ratio reflect the deformation characteristics of the material under stress; the cohesion and internal friction angle are key indicators of the shear strength of the rock-soil, which determine the ability of the slope to resist shear failure. By monitoring and analyzing these mechanical parameters, a mechanical model of the slope can be established to simulate its stress state under different loads and environmental conditions and to predict whether the slope will deform or become unstable, which is an important basis for quantitative analysis of stability.

[0067] The environmental parameters focus on external environmental factors, mainly including meteorological conditions (such as rainfall, rainfall intensity, air temperature, wind speed, sunshine duration, etc.) and hydrological conditions (groundwater level, surface runoff, pore water pressure, etc.). Rainfall and rainfall intensity directly affect the water content of rock-soil mass, causing rock-soil mass to saturate and soften, reducing shear strength and increasing sliding force, which is a common inducement for slope instability. Changes in air temperature can cause rock-soil mass to expand and contract, and over a long period of time, cracks may form, damaging the slope structure. Rising groundwater levels increase pore water pressure, reduce effective stress, and reduce slope stability; the erosive action of surface runoff can erode the slope surface and change the slope form. Real-time changes in these environmental parameters will continuously change the stress and physical state of the slope, so continuous monitoring and inclusion in the analysis can more accurately reflect the dynamic stability of the slope in the actual environment.

[0068] The displacement parameter is the most intuitive manifestation of slope stability changes. The displacement sensor monitors the displacement of the slope surface or internal in real time, including horizontal displacement, vertical displacement, displacement rate, etc. Small displacement changes can be a precursor to slope instability, and a continuous displacement growth trend is a direct indication of declining slope stability. Through monitoring and analysis of displacement parameters, signs of slope deformation can be captured in a timely manner, and the causes and development trends of deformation can be judged in combination with other parameters. For example, when the displacement rate suddenly increases, accompanied by rising groundwater levels and increasing rainfall, the slope can be quickly assessed as being in a high-risk state, providing key information for timely warning. The dynamic changes in displacement parameters are an important basis for judging slope stability and play an intuitive and key indicating role in the entire prediction process.

[0069] In an embodiment, when the key parameters are weighted and fused, weights are assigned based on the correlation degree on the spatio-temporal correlation matrix, and the fused parameters are generated through matrix point multiplication operation.

[0070] In this embodiment, the value of each element in the spatio-temporal correlation matrix represents the similarity or correlation tightness of the corresponding two key parameters in the time and space dimensions. The higher the correlation degree, the stronger the synergistic effect of the two parameters in reflecting the slope stability state, and the more significant the influence on the slope stability change, so a higher weight should be assigned; conversely, parameters with low correlation degree have less synergistic effect on slope stability, and the weight is correspondingly reduced. This weight allocation method based on correlation degree can fully exploit the potential relationship between parameters, making the weight setting more consistent with the actual physical mechanism and data characteristics of the slope.

[0071] In this embodiment, firstly, the spatio-temporal correlation matrix is processed by row or column (usually by column, each column corresponds to a key parameter), and the elements of each column are normalized to ensure that the sum of the weights is 1, thereby obtaining a weight vector corresponding to each key parameter. The weight vector reflects the influence of the correlation degree of other parameters with the key parameter in the spatio-temporal dimension on its importance. Then, the matrix composed of various key parameters (each key parameter as a column of the matrix, arranged in time sequence or spatial position) is multiplied by the above weight vector. When performing matrix multiplication, each element of the weight vector is multiplied by the corresponding element of the key parameter matrix column, realizing the weighting of the data of each key parameter at different time or spatial position. Finally, the weighted column data is added to obtain the final fusion parameter. This process integrates the information of each key parameter in the spatio-temporal dimension according to its close correlation degree through mathematical operation.

[0072] The above-mentioned weighted fusion method based on the spatio-temporal correlation matrix can effectively highlight the role of parameters that have a greater impact on the stability of the slope and are closely related to each other, while reducing the interference of parameters with low correlation and small impact. The generated fusion parameter not only contains the original information of each key parameter, but also incorporates the spatio-temporal correlation characteristics between parameters. Compared with simple parameter superposition or average processing, it can more comprehensively and accurately reflect the actual state of the slope. As the input of the subsequent slope stability prediction model, the fusion parameter can significantly improve the sensitivity and prediction accuracy of the model to the change of the slope stability, making the prediction result more in line with the actual situation and providing a more reliable basis for slope instability warning.

[0073] In an embodiment, based on the Granger causality test algorithm, the causal influence relationship between the parameters in the original parameter set is determined, a parameter causal network is constructed, and key parameters affecting the stability of the slope are selected, including:

[0074] The gray correlation degree of each parameter in the original parameter set with the historical data of the slope stability is calculated, and each parameter is preliminarily sorted according to the gray correlation degree, and Granger causality test is performed on the parameter pairs with high correlation degree first;

[0075] When performing Granger causality test, based on the change point detection algorithm, the structural mutation points of the parameters in the time sequence are identified, and the lag order of Granger causality test is set for different mutation intervals;

[0076] Based on the causal relationship strength evaluation model, the causal relationship strength scores of each parameter pair are calculated; the parameter pairs with a causal relationship strength score higher than a threshold value are retained to construct a parameter causal network;

[0077] The Louvain algorithm is used to divide the parameter causal network into communities, and the cohesion coefficient of each community and the contribution degree to the slope stability are calculated; the core parameters in the community with a contribution degree higher than a set value and the key parameters connected across communities are selected as the key parameters affecting the slope stability.

[0078] In the present embodiment, first, the parameters in the original parameter set are preliminarily screened through grey correlation degree analysis, the range of Granger causality test is narrowed, and the analysis efficiency is improved. Grey correlation degree analysis is a multi-factor statistical analysis method, which calculates the similarity of the geometric shapes of data sequence curves to determine the degree of correlation between factors. In the present embodiment, each parameter in the original parameter set is regarded as a factor, and the historical data of the slope stability is regarded as a reference sequence, and the grey correlation degrees between the parameter sequences and the reference sequence are calculated. The greater the correlation degree value, the more similar the parameter is to the trend of the slope stability, and the greater the potential impact of the parameter on the slope stability. After the parameters are sorted in descending order of the grey correlation degrees, the parameters with high correlation degrees are preferentially selected for Granger causality test. This is because the parameters with high correlation degrees are more likely to have a causal relationship with the slope stability, and preferentially testing them can focus on key factors and reduce unnecessary calculation, while ensuring the accuracy of the analysis and significantly improving the analysis efficiency.

[0079] When performing Granger causality test, the Granger causality test is optimized in view of the dynamic change characteristics of time series data. The change point detection algorithm can identify points where the data characteristics in the time series change significantly, i.e. structural mutation points. In the study of slope stability, due to the influence of geological activities, meteorological conditions and other factors, the time series of parameters may have multiple mutation points, and the change law and mutual relationship of data in different intervals may be different. After determining these mutation points based on the change point detection algorithm, the time series is divided into multiple mutation intervals. For each interval, the appropriate Granger causality test lag order is set according to the fluctuation characteristics and change law of the data. The lag order reflects the influence of the historical information of one variable on the current value of another variable, and different mutation intervals require different lag orders to accurately capture the causal relationship. In this way, the Granger causality test can better adapt to the dynamic changes of the data, improve the accuracy of the causal relationship judgment, and avoid misjudgment or omission caused by fixed lag order.

[0080] Further, a causal relationship strength evaluation model is constructed to quantify the causal relationship strength between each parameter pair, and then significant causal relationships are screened to construct a parameter causal network. The causal relationship strength evaluation model comprehensively considers the statistics (such as F statistics and P value) obtained by Granger causality test, the fluctuation amplitude of the parameters themselves, the sample size and other factors, and scores the causal relationship strength of each parameter pair through a specific calculation formula. The F statistics and P value reflect the significance level of the causal relationship, the fluctuation amplitude reflects the influence degree of parameter change on other parameters, and the sample size affects the reliability of the result. The higher the score, the stronger the causal relationship between the parameter pairs. A reasonable threshold is set, and parameter pairs with scores higher than the threshold are retained, which constitute the edges of the parameter causal network. With each parameter as a node and the retained parameter pairs as edges, a parameter causal network is constructed. The network directly shows the causal relationship between the parameters, and by removing the causal relationships with weak strength, the network is more concise and clear, highlighting the key causal relationships.

[0081] Finally, the Louvain algorithm is used to analyze the parameter causal network in depth and screen out key parameters that truly affect the slope stability. The Louvain algorithm is an efficient graph community division algorithm that can divide the parameter causal network into multiple closely connected communities, and the parameters in each community have strong internal correlation. The cohesion coefficient of each community is calculated, which measures the tightness of the connection between the nodes in the community. At the same time, combined with the historical data of the slope stability and the network structure, the contribution of each community to the slope stability is evaluated. The higher the contribution, the greater the influence of the parameters in the community on the slope stability. Core parameters are selected from the communities with a contribution higher than a set value. These core parameters play a key role in the community and have a close causal relationship with other parameters. In addition, hub parameters that connect across communities are selected. These parameters are key nodes for information transmission between different communities and are crucial for the connectivity of the entire network and the slope stability. In this way, the local role and global influence of the parameters in the network are comprehensively considered, and the selected key parameters can more accurately reflect the core factors affecting the slope stability.

[0082] In an embodiment, based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal influence relationship between the parameters, a parameter causal network is constructed, and key parameters affecting the slope stability are screened, including:

[0083] The original parameter set is subjected to three-dimensional convolution through a three-dimensional convolution kernel to obtain a feature-enhanced spatiotemporal parameter set;

[0084] A dynamic time window division method is used to adaptively adjust the time window length according to the parameter fluctuation frequency, and the spatiotemporal parameter set is segmented to obtain multiple parameter subsequences;

[0085] performing Granger causality test on each of the parameter subsequences respectively, and calculating weights of test results of the parameter subsequences by using an attention mechanism, and obtaining a comprehensive causality matrix by weighted fusion;

[0086] learning evolution rules of the comprehensive causality matrix over time based on a recurrent neural network, and predicting a change trend of causality influence relationship of each parameter in a future period;

[0087] constructing a parameter causality network with a time dimension based on the comprehensive causality matrix and the change trend, calculating time sequence importance of each node in the parameter causality network, and selecting parameters with time sequence importance higher than a preset value as key parameters affecting slope stability.

[0088] In the embodiment, first, a three-dimensional convolution operation is used to extract and enhance features of the original parameter set. A three-dimensional convolution kernel can process data in time, space and parameter channel dimensions simultaneously, and can better capture local features of parameters in time and space dimensions and correlation features between different parameters compared with a traditional two-dimensional convolution. When processing the original parameter set, the parameters are arranged in three dimensions according to time sequence, spatial position and parameter type, and the three-dimensional convolution kernel slides on the data body to extract feature information of parameters in a local region in time and space by convolution operation. For example, in a slope monitoring scenario, the coordinated change features between different types of parameters (such as displacement, humidity and pressure parameters) in a local spatial region in a time period can be effectively extracted. After three-dimensional convolution, the time and space parameter set with enhanced features retains not only the basic information of the original parameters, but also strengthens the time and space feature expression of the parameters, providing a more representative data basis for subsequent causality analysis, and helping to more accurately mine the causality between parameters.

[0089] Further, in view of the dynamic change characteristics of the parameter time series, a dynamic time window division strategy is adopted. Different parameters have different fluctuation frequencies in different time periods. For example, meteorological parameters such as rainfall have high fluctuation frequencies in the rainy season, while geological structure related parameters have low fluctuation frequencies under normal conditions. The traditional fixed time window division method cannot adapt to such changes, which may lead to insufficient data feature extraction or information redundancy. The dynamic time window division method adjusts the time window length in real time and adaptively by analyzing the fluctuation frequency of the parameters. In specific implementation, the fluctuation statistics (such as standard deviation, variance) of the parameters within a certain time range can be calculated using a sliding window. When the fluctuation statistics exceed the set threshold, the time window length is shortened to capture rapidly changing features. When the fluctuation is stable, the time window length is appropriately lengthened to obtain more comprehensive trend information. In this way, the enhanced spatiotemporal parameter set is divided into multiple parameter subsequences, and the parameter data in each subsequence has similar fluctuation characteristics, which enables more accurate analysis of the causal relationship between parameters under specific fluctuation conditions and improves the accuracy of causal relationship judgment.

[0090] Then, Granger causality test is performed on the parameter subsequences, and attention mechanism is introduced to optimize result fusion. Granger causality test is independently performed on each parameter subsequence to analyze the causal influence relationship between parameters under specific fluctuation conditions, which can capture the dynamic changes of the causal relationship between parameters more accurately than the test on the entire time series. However, due to differences in data characteristics, fluctuation conditions, etc., the importance of the test results of different parameter subsequences to the overall causal relationship judgment is also different. The attention mechanism learns the features of different parameter subsequences and assigns appropriate weights to the test results of each subsequence. For example, for a subsequence containing slope instability precursor data, its test results are more critical to the causal relationship of slope stability and are assigned a higher weight. While the subsequence containing normal fluctuation data has a relatively low weight. According to the calculated weights, the Granger causality test results of each parameter subsequence are weighted and fused to integrate them into a comprehensive causal relationship matrix. This matrix considers the causal relationship between parameters under different fluctuation conditions and more comprehensively reflects the causal relationship characteristics between parameters, providing more accurate information for subsequent analysis.

[0091] Further, the time dynamic characteristics of the integrated causal relationship matrix are modeled and predicted by using a recurrent neural network (RNN). The RNN has a memory function and can process data with time series characteristics. Through the cyclic connection of the hidden layer, the RNN can learn the dependence relationship of the data in the time dimension. The integrated causal relationship matrix is input into the RNN in time sequence, and the network learns the evolution law of the causal influence relationship of each parameter with time by continuously updating the hidden layer state. For example, during the learning process, the RNN can capture the trend that the causal relationship between the rainfall parameter and the slope displacement parameter gradually strengthens as the rainfall continues. Based on the learned evolution law, the RNN can predict the trend of the change of the causal influence relationship of each parameter in the future time period, and obtain the causal relationship prediction matrix at different time points in the future. This prediction capability enables the system not only to analyze the causal relationship between the current parameters, but also to predict the change of the future causal relationship in advance, thereby providing strong support for the forward-looking analysis and early warning of the slope stability.

[0092] Finally, key parameters are selected by constructing a parameter causal network and analyzing node importance. The parameter causal network with a time dimension is constructed by taking parameters as nodes and causal relationships as edges, combining the integrated causal relationship matrix and the predicted trend of the change of the causal relationship. The network not only shows the current causal relationship between the parameters, but also reflects the change of the causal relationship with time, and can more truly reflect the dynamic correlation between the parameters in the slope system. In the parameter causal network, the time sequence importance of each node (i.e., parameter) is calculated, which comprehensively considers the causal relationship strength of the parameter at different time points, the influence degree on other parameters, and the key degree in the evolution process of the causal relationship, etc. For example, in the process of slope instability, the time sequence importance of those parameters whose causal relationship strength continuously increases and which have an important influence on other key parameters will be higher. A preset value is set, and parameters with a time sequence importance higher than the preset value are selected as key parameters affecting the slope stability. These key parameters play a leading role in the process of slope stability change, and by focusing on these parameters, the slope stability can be more accurately analyzed and predicted, thereby providing a core basis for slope monitoring and early warning.

[0093] In an embodiment, the similarity of each key parameter in the time and space dimensions is calculated by a dynamic time warping algorithm to generate a spatiotemporal correlation matrix, including:

[0094] Local features of the key parameters in the spatial dimension are extracted by using a convolutional neural network, and are spliced into spatial sequence features;

[0095] Long-term dependence features of the key parameters in the time dimension are captured by using a long short-term memory network, and are spliced into time sequence features;

[0096] The time sequence features are mapped to quantum states, a dynamic time warping algorithm is used to combine quantum parallel computing of multiple time warping paths, similarity of each time warping path is evaluated, and an optimal matching path is selected to obtain a time dimension similarity matrix;

[0097] Based on the pre-constructed spatial similarity evaluation model, the spatial sequence features are analyzed to generate a spatial dimension similarity matrix.

[0098] The time dimension similarity matrix and the spatial dimension similarity matrix are fused through tensor product operation to obtain the spatiotemporal correlation matrix.

[0099] In this embodiment, first, in the prediction of road slope stability, the key parameters are collected from different sensors, and the original data of the key parameters are different in dimension, numerical range, and sampling frequency. For example, the unit of displacement parameter may be millimeter, while the unit of pressure parameter is Pascal, and the numerical magnitude of the two is greatly different. At the same time, the sampling frequencies of different sensors may also be inconsistent. The spatiotemporal dimension standardization processing unifies the time sequences of the key parameters to the same sampling frequency and time reference through linear transformation or interpolation method in the time dimension, and ensures the consistency of the time scale. In the spatial dimension, the parameter data of different position monitoring points are converted to a unified spatial coordinate system according to the geographic coordinate system, and the spatial coordinate values are normalized to eliminate the spatial scale difference. After standardization, the standardized spatiotemporal parameter set is formed, so that the key parameters have comparability in the spatiotemporal dimension, and the interference of dimension and scale on the similarity calculation result is avoided.

[0100] Further, the feature information of the key parameters in the spatial dimension is mined by means of the powerful local feature extraction capability of the convolutional neural network (CNN). The convolutional layer of the CNN performs convolution operation on the data by sliding the convolution kernel, and can automatically extract the local features of the data. When processing the spatial data of the key parameters, the spatial dimension data (such as parameter values of different monitoring points) in the standardized spatiotemporal parameter set is organized into a two-dimensional or three-dimensional data structure as the input of the CNN. The convolution kernel slides on the data structure to extract the local correlation features between the parameters of adjacent monitoring points, such as the cooperative variation features of displacement, humidity and other parameters in a certain region of the slope. After a plurality of convolution, pooling and other operations, the data dimension is further compressed and the feature expression is further strengthened, and finally the extracted local features are spliced in the order of spatial position to form the spatial sequence features. Such spatial sequence features not only retain the original information of the parameters of each monitoring point, but also highlight the local correlation characteristics of the parameters in the space, providing more representative feature data for subsequent calculation of spatial dimension similarity.

[0101] Further, the long-term dependence problem in time series data is solved by using a long short-term memory network (LSTM), and the dynamic change characteristics of the key parameters in the time dimension are obtained. LSTM is a special recurrent neural network that can effectively process and remember information in long time series through a gating mechanism (input gate, forget gate, and output gate), avoiding the problem of gradient vanishing or explosion. The time series data in the standardized spatiotemporal parameter set are input into the LSTM, and when processing the data at each time step, the LSTM selectively retains or forgets historical information according to the gating mechanism, thereby capturing the long-term dependence relationship of the parameters in the time dimension. For example, when analyzing the change of slope displacement over time, LSTM can remember the change trend of the displacement in the early stage, and combine the current data to judge whether the displacement change has a long-term trend such as acceleration or deceleration. With the advancement of time steps, the LSTM outputs the hidden state corresponding to each time step, and these hidden states are spliced in chronological order to form a time series feature. This time series feature contains the long-term evolution law and dynamic change information of the key parameters in the time dimension, providing a key feature representation for subsequent time dimension similarity calculation.

[0102] Then, quantum computing is combined with the dynamic time warping algorithm to improve the efficiency and accuracy of time dimension similarity calculation. First, the time series feature output by the LSTM is mapped to a quantum state, and the superposition and entanglement characteristics of the quantum state are used to realize parallel computing of multiple time warping paths. The traditional dynamic time warping algorithm needs to traverse all possible time alignment paths when calculating the similarity of two time series, and the computational complexity is high. Based on quantum parallel computing, multiple time warping paths can be evaluated simultaneously, greatly improving the computational efficiency. During the calculation process, the quantum state corresponding to each time warping path is calculated by designing appropriate quantum gate operations and measurement methods, and the similarity measure value of the two time series under each path is obtained. Then, the optimal matching path with the highest similarity is selected according to these similarity measure values, and the similarity value corresponding to this path is taken as the similarity score of the two time series in the time dimension. The above calculation is performed on all pairs of time series of key parameters to obtain a time dimension similarity matrix, and each element in the matrix represents the similarity of the corresponding two key parameter time series, providing a quantitative basis for comprehensive analysis of the correlation between parameters in the time dimension.

[0103] Further, the spatial sequence characteristics of the key parameters are analyzed in depth by relying on a pre-constructed spatial similarity evaluation model. The spatial similarity evaluation model can be constructed based on machine learning algorithms (such as support vector machines, random forests, etc.) or deep learning algorithms (such as graph neural networks), and its training data comes from historical monitoring data of the slope and related geological and geographical information. The input of the model is the spatial sequence characteristics extracted and spliced by the CNN, and the output is the similarity scores of the key parameters in the spatial dimension. During the training process, the model learns the mapping relationship between the parameter characteristics and the similarity at different spatial positions, such as considering the geographical distance of the monitoring points, the similarity of the geological conditions, and the difference in parameter values, etc., to establish the rules or model parameters of spatial similarity evaluation. When new spatial sequence characteristics are input, the model calculates the spatial similarity scores between the key parameters according to the learned knowledge, organizes these scores into a matrix form according to the parameter correspondence relationship, and generates a spatial dimension similarity matrix. This matrix reflects the correlation degree of the key parameters in the spatial dimension, providing an important reference for analyzing the mutual relationship between the parameters at different positions of the slope.

[0104] Finally, the time and spatial dimension similarity information is organically integrated through tensor product operation. Tensor product is a mathematical operation that can combine the information of two matrices to generate a new tensor (in the case of two-dimensional matrix, the fusion result is still a matrix). The time dimension similarity matrix describes the similarity relationship of the key parameters in the time sequence, the spatial dimension similarity matrix describes the similarity relationship of the parameters in the spatial position, and the spatio-temporal correlation matrix needs to reflect the comprehensive correlation characteristics of the parameters in the time and space dimensions. The time dimension similarity matrix and the spatial dimension similarity matrix are used as the operands of the tensor product operation, and the tensor product operation is performed. During the operation, the matrix elements are calculated according to the corresponding relationship, so that the fused matrix elements contain both time dimension similarity information and spatial dimension similarity information. The final spatio-temporal correlation matrix fully describes the mutual correlation relationship of the key parameters in the time and space dimensions, providing core data support for subsequent parameter weighted fusion, slope stability prediction, etc. based on spatio-temporal correlation information, and helping to more accurately grasp the spatio-temporal rules of slope stability change.

[0105] In an embodiment, the warning threshold of the slope stability prediction model is dynamically updated, comprising:

[0106] Based on the historical prediction data, the warning threshold of the slope stability prediction model is dynamically updated by using a sliding window algorithm.

[0107] In this embodiment, by introducing a sliding window algorithm, the dynamic adjustment of the early warning threshold of the slope stability prediction model is realized based on historical prediction data to adapt to the complex changes of the slope environment. The core of the sliding window algorithm is to divide the historical prediction data according to a certain time sequence to form a continuous and overlapping window data set. In this scheme, the window size can be set according to actual needs, for example, a window containing the last N prediction data is selected. As time goes on, the window slides backward at a fixed step, and each time it slides, the early warning threshold is updated based on the data in the current window, so that the model can continuously adapt to new data features and changes in the slope state.

[0108] In specific operation, first, historical prediction data is collected and organized, which includes past predictions of slope stability and corresponding actual slope state information (such as whether instability occurs, instability degree, etc.). The data in the window is analyzed in units of sliding windows, and key statistical features such as mean, standard deviation, maximum value, and minimum value of the prediction values are extracted. These statistical features reflect the distribution and fluctuation of the slope stability prediction values in this time period. For example, if the standard deviation of the prediction values in the window is large, it indicates that the slope stability fluctuates violently in this period, and the original early warning threshold may no longer be applicable.

[0109] Based on the statistical characteristics of the window data, combined with the actual working conditions of the slope and historical instability cases, the early warning threshold is updated using appropriate calculation methods. One feasible way is to set the early warning threshold as the mean value plus k times the standard deviation (k is a coefficient determined according to risk preference and historical experience) according to the mean value and standard deviation of the prediction values. When the window slides to a new time period, with the update of data, the statistical characteristics of the data in the window will also change, thereby causing the calculated early warning threshold to be adjusted accordingly. For example, during the rainy season, the fluctuation of slope stability increases due to the influence of rainfall. Through the sliding window algorithm, the early warning threshold calculated using more prediction data under the influence of rainfall in this period will be lowered to capture the slope instability risk more sensitively. Dynamic updating of the early warning threshold using the sliding window algorithm has significant advantages. On the one hand, compared to a fixed threshold, this method can reflect the dynamic change trend of the slope stability prediction value in real time, avoiding false positives or false negatives caused by environmental changes. For example, in areas with complex geological conditions or frequent climate changes, a fixed threshold is difficult to adapt to changing conditions, while a dynamically updated threshold can be adjusted flexibly according to actual data. On the other hand, by segmenting and analyzing historical prediction data, the interference caused by short-term data fluctuations is effectively filtered, making the adjustment of the early warning threshold more scientific and reliable, thereby providing a strong guarantee for accurate assessment and timely warning of slope stability, better serving road safety maintenance and disaster prevention work.

[0110] In an embodiment, after outputting the slope instability early warning information, it includes:

[0111] compile the slope instability early warning information into character stream information;

[0112] group the character stream information according to preset rules to obtain a plurality of character combinations; each character combination includes one or two characters;

[0113] obtain a preset graph structure template, and sequentially add each character combination to each node of the graph structure template to generate a character combination graph structure; each node has a corresponding serial number;

[0114] sequentially detect whether the character combination in each node of the character combination graph structure is repeated with the previous node, if repeated, remove the repetition, and take the node after removing the repetition as a target node;

[0115] sequentially connect each target node to generate a curve; based on the position relationship between the curve and each node, reorder the character combinations on each node to obtain a character combination sequence;

[0116] obtain an encoding table; the encoding table includes one-to-one mapping of original data columns and encoded data columns;

[0117] sequentially replace the character combinations in the character combination sequence into the encoded data columns of the encoding table until all the character combinations are replaced in the encoding table or the encoded data columns are completely replaced to obtain a replaced encoding table;

[0118] encode the original parameter set based on the replaced encoding table to obtain encoded data and send the encoded data to a management terminal.

[0119] In this embodiment, the slope instability early warning information includes various types of data, such as slope displacement values, geological parameter changes, weather conditions, etc. These different formats of data are uniformly converted into character stream form for subsequent processing and analysis. A specific encoding method (such as UTF-8 encoding) is used to convert various data elements in the early warning information into character sequences to form continuous character stream information. For example, the displacement value "10.5" is converted into characters "1" "0" "5", and the geological parameter "sandstone" is converted into corresponding character combinations. The character representations of all data are integrated to obtain complete character stream information.

[0120] Further, grouping the character stream information and dividing it into smaller character combinations helps to process and analyze the data more meticulously, while limiting the number of characters in each combination (1 or 2 characters) can make the grouping more standardized and orderly. According to the preset grouping rules, 1 or 2 characters are sequentially selected as a character combination from the starting position of the character stream. For example, the character stream is "ABCD", and according to the rules, the character combinations "A", "BC", and "D" can be obtained. Reasonably dividing the character stream information provides basic data units for subsequent generation of graph structure and further processing, making the data structure clearer and facilitating various analyses and operations.

[0121] Further, by using a preset graph structure template, the character combinations are associated with the nodes of the graph to construct a character combination graph structure. The position and order of the character combinations are determined by the node sequence number, which helps to analyze and process the character combinations from the perspective of the graph. First, a pre-set graph structure template is obtained, which defines the number of nodes, connection relationships, and other information. Then, each character combination is added to the corresponding node of the graph structure template according to the order of the character combinations. If the number of nodes is greater than the number of character combinations, the remaining nodes remain empty or are filled according to specific rules. Each node is assigned a unique sequence number to identify its position in the graph structure. Presenting the character combinations in the form of a graph structure facilitates the analysis of the associations and order between the character combinations, providing an intuitive structural framework for subsequent deduplication and reordering operations, while the node sequence number provides a clear index for data processing.

[0122] In the character combination graph structure, detecting and removing duplicate character combinations can reduce data redundancy, making the character combination sequence more concise and accurate, and highlighting the uniqueness and effectiveness of the data. Starting from the first node of the graph structure, the character combination of the current node is compared with the character combinations of all previous nodes (determined according to the node sequence number). If a duplicate is found (i.e., the character combination is exactly the same), the current node is deleted, and the non-duplicate node is retained. The node that is removed is used as the target node. For example, the character combinations of the first three nodes in the graph structure are "A", "BC", and "A", respectively. The character combination "A" of the third node is detected as a duplicate of the first node, and the third node is removed and used as the target node.

[0123] In this embodiment, duplicate character combinations are removed to optimize the character combination graph structure, making subsequent processing based on more refined data, improving the efficiency and accuracy of data processing, and avoiding errors or redundant analysis caused by duplicate data.

[0124] Further, by generating a curve through the connection of the target nodes, and reordering the character combinations using the positional relationship of the curve and the nodes, the potential order relationship between the character combinations can be mined, making the character combination sequence more in line with certain logic or rules, and providing a more reasonable order for subsequent data encoding. Specifically, the de-duplicated target nodes are connected in order according to their positions in the graph structure, forming a curve. According to the direction of the curve and the positions of the nodes on the curve, the order of the character combinations is re-determined. For example, if the curve passes through the target nodes from left to right in order, the order of the character combinations on the nodes is adjusted according to the order of the nodes on the curve, obtaining a new character combination sequence. The position relationship of the nodes can be determined by calculating the coordinates or distances of the nodes on the curve, and then the order is determined. The character combinations are rearranged to make the character combination sequence more unique and unique, providing a more suitable data order for the encoding operation based on the encoding table.

[0125] The encoding table, as the basis for data encoding, converts character combinations into a specific encoding form through the one-to-one mapping relationship between the original data column and the encoding data column, realizes the encryption or standardized representation of data, and facilitates data transmission and storage. A pre-prepared encoding table is obtained, which includes two columns, one column is the original data column, which is used to store the original character combinations or other related data; the other column is the encoding data column, which stores the encoding values corresponding to the original data column. For example, the encoding table specifies that the original data "12" corresponds to the encoding data "AB", which clearly defines the mapping relationship between the original data and the encoding data. Specific rules and basis are provided for data encoding, so that character combinations can be converted according to a specific mapping relationship, effectively encoding data, ensuring the security and standardization of data in the transmission process, and facilitating the decoding and analysis of data by the management terminal.

[0126] Finally, the character combinations in the character combination sequence are matched with the encoding data column in the encoding table, and the corresponding contents in the encoding data column are replaced with the character combinations, generating a unique replacement encoding table, preparing for subsequent data transmission. Starting from the first character combination in the character combination sequence, the first encoding data in the encoding data column is replaced with the character combination. This process is repeated to process each character combination in the character combination sequence in turn, until all character combinations are replaced in the encoding table, or the encoding data column is completely replaced, obtaining a replacement encoding table.

[0127] Further, the original parameter set is encoded using the replacement encoding table, and the original parameters are converted into encoding data corresponding to the replacement encoding table, realizing the standardization and encryption of data, ensuring the security and effectiveness of data in the transmission process, and facilitating the reception and analysis of the management terminal.

[0128] According to the mapping relationship between the character combination in the replacement encoding table and the original data, each parameter in the original parameter set is encoded. For example, the original parameter set contains parameters "12" and "4", according to the mapping rule of the replacement encoding table, "12" is encoded into the corresponding encoding value, and "4" is encoded into the corresponding value, and the encoded data is obtained. Then, through a specific communication protocol and channel, the encoded data is sent to the management terminal, ensuring the accuracy and reliability of data transmission. The original parameter set is encoded and sent, realizing the safe transmission of data from the sending end to the management terminal. The management terminal can decode the received encoded data according to the encoding table, restore the original parameter set, and then further analyze and decide, so that the original parameter set has stronger security in the sending process through the improved encoding table, avoiding data leakage. At the same time, the generation of the above-mentioned replacement encoding table is closely related to the slope instability early warning information, which enhances the reusability and traceability.

[0129] Reference Figure 2 In another embodiment of the present application, a road slope stability prediction device is also provided, comprising:

[0130] The acquisition unit is configured to acquire real-time multi-dimensional environmental parameters of the road slope to form an original parameter set;

[0131] The screening unit is configured to analyze the original parameter set based on a Granger causality test algorithm, determine the causal influence relationship between parameters, construct a parameter causal network, and screen out key parameters affecting slope stability;

[0132] The correlation unit is configured to calculate the similarity of each key parameter in the time and space dimensions by a dynamic time warping algorithm, and generate a space-time correlation matrix;

[0133] The fusion unit is configured to weight and fuse the key parameters based on the space-time correlation matrix to obtain a fused parameter;

[0134] The prediction unit is configured to input the fused parameter into a pre-trained slope stability prediction model to obtain a predicted value of the slope stability;

[0135] The early warning unit is configured to dynamically update the early warning threshold of the slope stability prediction model, and output a slope instability early warning information when the predicted value exceeds the dynamically updated early warning threshold.

[0136] In the present embodiment, the specific implementation of each unit in the above device embodiment is described in the above method embodiment, which will not be repeated here.

[0137] Reference Figure 3 In the present embodiment, an electronic device can be a server, and its internal structure can be as follows Figure 3The electronic device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store the corresponding data in the embodiment. The network interface of the electronic device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement the above method.

[0138] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied.

[0139] The computer readable storage medium in the embodiment of the present application can be a volatile readable storage medium or a non-volatile readable storage medium.

[0140] In summary, the road slope stability prediction method, device, storage medium and electronic device provided in the embodiment of the present application include: collecting multi-dimensional environmental parameters of a road slope in real time to form an original parameter set; analyzing the original parameter set based on a Granger causality test algorithm, determining the causal influence relationship between parameters, constructing a parameter causal network, and screening out key parameters affecting slope stability; calculating the similarity of each key parameter in the time and space dimensions through a dynamic time warping algorithm to generate a spatio-temporal correlation matrix; weighting and fusing the key parameters based on the spatio-temporal correlation matrix to obtain a fused parameter; inputting the fused parameter into a pre-trained slope stability prediction model to obtain a predicted value of slope stability; and dynamically updating the warning threshold of the slope stability prediction model, and outputting slope instability warning information when the predicted value exceeds the dynamically updated warning threshold. In the present application, the original parameter set is analyzed to determine the causal influence relationship between parameters, a parameter causal network is constructed, key parameters affecting slope stability are screened out, the similarity of each key parameter in the time and space dimensions is calculated to generate a spatio-temporal correlation matrix, and the warning threshold of the slope stability prediction model is dynamically updated. The potential relationship between multi-dimensional environmental parameters is fully tapped, and the defects of inaccurate prediction and data redundancy of current road slope stability prediction methods are overcome.

[0141] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0142] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, device, article or method that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, device, article or method. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article or method that includes the element.

[0143] The above description is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, based on the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A method for predicting the stability of road slopes, characterized in that, Includes the following steps: Real-time collection of multi-dimensional environmental parameters of roadside slopes to form an original parameter set; Based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal relationship between each parameter, construct a parameter causal network, and screen out the key parameters affecting slope stability. The process involves calculating the similarity of key parameters in both time and space dimensions using a dynamic time warping algorithm to generate a spatiotemporal correlation matrix. This includes: extracting local features of the key parameters in the spatial dimension using a convolutional neural network and concatenating them into spatial sequence features; capturing long-term dependency features of the key parameters in the time dimension using a long short-term memory network and concatenating them into time series features; mapping the time series features to quantum states; evaluating the similarity of each time warping path using the dynamic time warping algorithm combined with quantum parallel computation to select the optimal matching path and obtain a time dimension similarity matrix; analyzing the spatial sequence features based on a pre-built spatial similarity evaluation model to generate a spatial dimension similarity matrix; and fusing the time dimension similarity matrix and the spatial dimension similarity matrix using tensor product operations to obtain the spatiotemporal correlation matrix. Based on the spatiotemporal correlation matrix, the key parameters are weighted and fused to obtain fused parameters; when weighting and fusing the key parameters, weights are assigned based on the degree of correlation on the spatiotemporal correlation matrix, and fused parameters are generated through matrix dot product operation; The fusion parameters are input into the pre-trained slope stability prediction model to obtain the predicted value of slope stability. The warning threshold of the slope stability prediction model is dynamically updated. When the predicted value exceeds the dynamically updated warning threshold, slope instability warning information is output.

2. The road slope stability prediction method according to claim 1, characterized in that, The environmental parameters include geological parameters, mechanical parameters, environmental parameters, and displacement parameters.

3. The road slope stability prediction method according to claim 1, characterized in that, Based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal relationships between parameters, a parameter causal network is constructed, and key parameters affecting slope stability are screened out, including: Calculate the grey relational degree between each parameter in the original parameter set and the historical data of slope stability. Sort the parameters according to the grey relational degree and prioritize Granger causality test for parameter pairs with high relational degree. When performing Granger causality tests, the structural abrupt change points of parameters in the time series are identified based on the change point detection algorithm, and the lag order of Granger causality tests is set for different abrupt change intervals. Based on the causal relationship strength assessment model, the causal relationship strength score between each parameter pair is calculated; parameter pairs with causal relationship strength scores higher than the threshold are retained to construct a parameter causal network. The Louvain algorithm was used to divide the parameter causal network into communities, and the cohesion coefficient and contribution to slope stability of each community were calculated. Core parameters in communities with contributions higher than a set value, as well as pivotal parameters connecting across communities, were selected as key parameters affecting slope stability.

4. The road slope stability prediction method according to claim 1, characterized in that, Based on the Granger causality test algorithm, the original parameter set is analyzed to determine the causal relationships between parameters, a parameter causal network is constructed, and key parameters affecting slope stability are screened out, including: The original parameter set is subjected to three-dimensional convolution using a three-dimensional convolution kernel to obtain a spatiotemporal parameter set with enhanced features. A dynamic time window segmentation method is adopted, and the time window length is adaptively adjusted according to the parameter fluctuation frequency to segment the spatiotemporal parameter set and obtain multiple parameter subsequences. Granger causality tests are performed on each of the parameter subsequences, and the weights of the test results for each parameter subsequence are calculated using an attention mechanism. The weighted fusion is then used to obtain a comprehensive causal relationship matrix. Based on the recurrent neural network, the evolution of the comprehensive causal relationship matrix over time is learned, and the changing trend of the causal influence relationship of each parameter in the future period is predicted. Based on the comprehensive causal relationship matrix and the changing trend, a parametric causal network with a time dimension is constructed. The temporal importance of each node in the parametric causal network is calculated, and parameters with temporal importance higher than a preset value are selected as key parameters affecting slope stability.

5. The road slope stability prediction method according to claim 1, characterized in that, Dynamically updating the early warning threshold of the slope stability prediction model includes: Based on historical prediction data, the early warning threshold of the slope stability prediction model is dynamically updated using a sliding window algorithm.

6. A road slope stability prediction device, characterized in that, include: The data acquisition unit is used to collect multi-dimensional environmental parameters of the roadside slope in real time and form a raw parameter set. The screening unit is used to analyze the original parameter set based on the Granger causality test algorithm, determine the causal relationship between each parameter, construct a parameter causal network, and screen out the key parameters that affect slope stability. The association unit is used to calculate the similarity of each key parameter in time and space dimensions using a dynamic time warping algorithm to generate a spatiotemporal association matrix. This includes: extracting local features of the key parameters in the spatial dimension using a convolutional neural network and concatenating them into spatial sequence features; capturing long short-term memory features of the key parameters in the time dimension using a long short-term memory network and concatenating them into time series features; mapping the time series features to quantum states; evaluating the similarity of each time warping path using a dynamic time warping algorithm combined with quantum parallel computation to select the optimal matching path and obtain a time dimension similarity matrix; analyzing the spatial sequence features based on a pre-built spatial similarity evaluation model to generate a spatial dimension similarity matrix; and fusing the time dimension similarity matrix and the spatial dimension similarity matrix through tensor product operations to obtain the spatiotemporal association matrix. The fusion unit is used to perform weighted fusion of the key parameters based on the spatiotemporal correlation matrix to obtain fusion parameters; when performing weighted fusion of the key parameters, weights are assigned based on the degree of correlation on the spatiotemporal correlation matrix, and fusion parameters are generated through matrix dot product operation; The prediction unit is used to input the fused parameters into the pre-trained slope stability prediction model to obtain the predicted value of slope stability. The early warning unit is used to dynamically update the early warning threshold of the slope stability prediction model. When the predicted value exceeds the dynamically updated early warning threshold, it outputs slope instability early warning information. The slope instability early warning information is compiled into a character stream; the character stream information is grouped according to a preset rule to obtain multiple character combinations; each character combination includes one or two characters; Obtain a preset graph structure template, and add each character combination sequentially to each node of the graph structure template to generate a character combination graph structure; where each node has a corresponding sequence number; Sequentially check whether the character combination in each node of the character combination graph structure is repeated with the previous node. If it is repeated, deduplication is performed and the deduplicated node is taken as the target node. Connect each target node sequentially to generate a curve; based on the positional relationship between the curve and each node, reorder the character combinations on each node to obtain a character combination sequence; Obtain the encoding table; the encoding table includes original data columns and encoded data columns that are mapped one-to-one; The character combinations in the character combination sequence are sequentially replaced into the encoded data column of the encoding table until all character combinations are replaced into the encoding table, or the entire encoded data column is replaced, to obtain the replacement encoding table; The original parameter set is encoded based on the replacement encoding table to obtain encoded data, which is then sent to the management terminal.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images