Rock soil stability prediction method and system based on machine learning

Abstract

The invention relates to the technical field of machine learning, in particular to a rock and soil stability prediction method and system based on machine learning, and the method comprises the following steps: extracting shear stress, displacement and energy dissipation rate time series data, calculating the stress change rate of adjacent time steps, screening the stress increment sudden change period, and calculating the shear deformation gradient change rate. Mutation points in the strain direction are screened, and a key variable interval is obtained. According to the method, traditional analysis limitation is broken through by combining displacement-stress second derivative and energy dissipation rate cross validation, a hardening-softening conversion area is dynamically identified, multi-stage strain gradient differences are compared to establish a space-time correlation parameter system, and static and dynamic parameter conversion is realized by using a strain increment space gradient tracking technology; and quantitative mechanical interaction and energy dissipation collaborative analysis is carried out to describe a stress diffusion path, stress release characteristics and deformation gradient distribution are fused to construct a fracture principal direction multi-dimensional verification mechanism, and a microscopic fracture and macroscopic instability correlation model is established.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a method and system for predicting rock and soil stability based on machine learning. Background Art

[0002] The field of machine learning technology includes computational methods that train data by building mathematical models to predict and classify unknown data. The core content of this technical field is to build a generalizable model system based on historical data. Common implementation methods include supervised learning, unsupervised learning, and reinforcement learning. Machine learning is widely used in many fields such as image recognition, natural language processing, automatic control, and geological disaster prediction. It has the ability to extract correlations from complex data and perform efficient learning and reasoning. In geotechnical engineering, machine learning technology is gradually used in key links such as stability assessment, risk identification, and deformation trend prediction. Through comprehensive modeling of time series data, spatial distribution characteristics, and physical properties, it assists engineering safety analysis and judgment.

[0003] Among them, the geotechnical stability prediction method refers to the establishment of a key variable identification and evolution path derivation mechanism by collecting and processing multi-source time series data such as shear stress, displacement, and energy dissipation rate, and combining the second-order derivative analysis of the displacement-stress curve, the calculation of the energy dissipation rate per unit volume, and the strain diffusion rate accounting, etc., to identify stability characteristics and derive spatial distribution parameters. The main contents of this method include extracting mutation periods based on stress change rate, identifying key points through shear deformation gradients, forming characteristic parameter sets using multi-stage strain gradient differences, calculating strain diffusion rates in stress areas and identifying diffusion paths, and on this basis analyzing the correlation between strain increments and energy dissipation diffusion trends between regions, constructing deformation transfer paths, and correcting energy influence coefficients.

[0004] The shortcomings of existing technologies stem from the inherent limitations of static model architecture and single-dimensional analysis mode. Traditional methods rely on fixed parameter thresholds to identify key variables, and lack the ability to capture mutation periods and directional mutation points in real time through dynamic screening mechanisms, resulting in significant lags in the identification of key mechanical features. The single-dimensional criterion system is difficult to accurately analyze the complex mechanical state of the hardening-softening transition zone, which can easily lead to misjudgment of the state. Static spatial parameters cannot reflect the evolution law of strain diffusion rate, resulting in systematic deviations in the evaluation of mechanical interactions between regions. The technology for identifying the direction of fracture extension is limited to macroscopic trend judgment, lacks the ability to accurately capture microscopic gradient mutation characteristics, and is difficult to establish a correlation model between fracture mechanism and instability trend. The static setting mode of the energy influence coefficient ignores the dynamic characteristics of the actual diffusion process, resulting in insufficient accuracy in predicting the energy dissipation trend. Typical cases show that traditional methods have obvious defects in the identification of potential slip surfaces and the timeliness of early warning responses due to the lack of a multi-stage strain gradient difference analysis mechanism. These structural defects lead to the common problem of insufficient adaptability of existing technologies in dealing with complex geological conditions and dynamic evolution processes. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a rock and soil stability prediction method and system based on machine learning.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for predicting geotechnical stability based on machine learning, comprising the following steps:

[0007] S1: Extract the time series data of shear stress, displacement, and energy dissipation rate, calculate the stress change rate of adjacent time steps, screen the stress increment mutation period, calculate the shear deformation gradient change rate, screen the strain direction mutation point, and obtain the key variable interval;

[0008] S2: Based on the key variable interval, calculate the second-order derivative of the displacement-stress curve, extract the stress corresponding to the displacement mutation point, calculate the drop rate after the stress peak, screen the hardening-softening transition zone, calculate the change in the energy dissipation rate per unit volume, extract the stress / displacement / strain parameters at the time point of maximum energy loss, compare the strain gradient differences in multiple stages, and form a set of characteristic parameters;

[0009] S3: According to the characteristic parameter set, calculate the local unit strain increment, analyze the spatial gradient change trend, calculate the strain diffusion rate of the stress area, identify the stress diffusion path, and derive the spatial distribution parameters in combination with the deformation gradient change;

[0010] S4: Based on the spatial distribution parameters, calculate the correlation coefficient of strain increments between regions, screen the strong interactive influence area, analyze the energy dissipation diffusion trend, construct the deformation transfer path, call the slope sliding data, calculate the diffusion range of the high-value energy dissipation area, correct the energy influence coefficient, and generate the co-evolution parameter;

[0011] S5: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress release rate at the fracture point, analyze the deformation gradient distribution, identify the main direction of fracture extension, locate the maximum strain gradient mutation domain, and output the development trend parameters.

[0012] As a further solution of the present invention, the key variable interval includes stress mutation period, strain mutation point, shear gradient mutation point, the characteristic parameter set includes displacement mutation stress, stress drop rate, hardening-softening zone, energy loss point, the spatial distribution parameter includes strain increment, gradient change, stress diffusion, deformation gradient, and the co-evolution parameter includes strain correlation, interactive influence zone, energy diffusion, and energy coefficient.

[0013] As a further solution of the present invention, the step of obtaining the key variable interval is specifically as follows:

[0014] S101: Based on the time series data of shear stress, displacement and energy dissipation rate, the shear stress change rate of adjacent time steps is calculated to identify the sudden change period of stress increment. By calculating the mean and standard deviation of the shear stress change rate, the threshold interval exceeding the set multiple is screened to obtain the stress sudden change period;

[0015] S102: calling the stress mutation period, calculating the change rate of the shear deformation gradient, comparing the time series of the change rate, determining the key time node of the strain direction change according to the direction change characteristics of the gradient change rate, and obtaining the strain direction mutation point;

[0016] S103: Calling the stress mutation period and the strain direction mutation point, determining the interval range of key variables, calculating the coupling relationship between the shear stress increment, the shear deformation gradient change rate and the energy dissipation rate, using the formula:

[0017]

[0018] Calculate the variation characteristics of the key variable interval to obtain the key variable interval;

[0019] Among them, Δτ t1 represents the shear stress increment at the t1th time step, represents the shear deformation gradient change rate at the t1th time step, represents the rate of change of energy dissipation rate at the t1th time step, C is a small positive number to avoid the denominator approaching zero, K represents the characteristic measure of the change in the key variable interval, and T represents the total number of time steps in the key variable interval.

[0020] As a further solution of the present invention, the step of acquiring the characteristic parameter set is specifically as follows:

[0021] S201: Based on the key variable interval, calculate the second-order derivative of the displacement-stress curve, identify the mutation point of the second-order derivative, and extract the stress value corresponding to the mutation point to obtain a set of stress mutation points;

[0022] S202: calling the stress mutation point set, calculating the drop rate after the stress peak, and screening the hardening-softening transition zone, calculating the change of the energy dissipation rate per unit volume in the transition zone, and obtaining the energy dissipation rate change set;

[0023] S203: calling the energy dissipation rate change set, extracting the maximum energy loss time point, calculating the stress, displacement, and strain parameters at the time point, analyzing the strain gradient difference in the differentiation stage, and using the formula:

[0024]

[0025] The strain gradient variation is calculated, and the strain gradient in the differentiation stage is compared to obtain a set of characteristic parameters;

[0026] Among them, ΔG represents the change of strain gradient, Δσ s1 represents the stress increment in stage s1, Δε s1 represents the strain increment in stage s1, E s1 Represents the energy dissipation rate per unit volume in stage s1, W s1 represents the displacement of the s1th stage, and S represents the total number of stages.

[0027] As a further solution of the present invention, the step of acquiring the spatial distribution parameters is specifically as follows:

[0028] S301: Based on the characteristic parameter set, the local unit strain increment is calculated, and the local deformation parameter is obtained according to the difference between the initial coordinates and the deformed coordinates of the multiple units. At the same time, the strain increment data is called to calculate the displacement vector between adjacent units. According to the relationship between the strain increment of the multiple units and the strain change of the adjacent units, the spatial gradient is calculated to obtain the local strain gradient change value;

[0029] S302: Call the local strain gradient change value to calculate the strain diffusion rate in the stress area using the formula:

[0030]

[0031] Calculate the diffusion effect coefficient of the unit and obtain the strain diffusion rate distribution;

[0032] Among them, D eff represents the diffusion effect coefficient, represents the local strain gradient change value of the u1th unit, U represents the number of calculation units, γ u1 represents the stress non-uniformity parameter of the u1th unit, represents the maximum local strain gradient change value among all elements, α represents the overall strain distribution adjustment coefficient, and β represents the maximum gradient offset correction parameter;

[0033] S303: calling the strain diffusion rate distribution, identifying the stress diffusion path, calculating the main diffusion direction according to the strain gradient direction of multiple units, comparing the strain diffusion rate in combination with the deformation gradient change, classifying the diffusion area according to the strain gradient change value, and deriving the spatial distribution parameters.

[0034] As a further solution of the present invention, the step of obtaining the co-evolution parameter is specifically as follows:

[0035] S401: Based on the spatial distribution parameters, calculate the strain increment correlation coefficient between regions, screen the regions where the strain increment correlation coefficient is greater than the set threshold, call the strain increment data matrix at the same time, extract the index number of the region that meets the conditions, extract the strain increment correlation coefficient corresponding to the region according to the index number, construct the interaction matrix between regions, and obtain the index set of the strong interaction influence area;

[0036] S402: calling the strong interaction influence zone index set, analyzing the diffusion trend of energy dissipation, calculating the energy dissipation increment of multiple regions, and dividing the connection relationship between regions according to the interaction influence intensity based on the interaction influence matrix between regions, extracting the region index on the deformation path, calculating the energy dissipation diffusion amount of multiple regions on the path, and obtaining the energy diffusion value of the deformation transfer path;

[0037] S403: calling the energy diffusion value of the deformation transfer path, combining with the slope sliding data, calculating the diffusion range of the high-value energy dissipation area, and correcting the energy influence coefficient, using the formula:

[0038]

[0039] Calculate and obtain the co-evolution parameters;

[0040] Among them, E syn represents the coevolution parameter, ΔE r1 represents the energy dissipation increment of region r1, A r1 represents the spatial area of ​​region r1, R r1 represents the dissipation rate of region r1 during the energy dissipation process, S r1represents the sliding stability coefficient of region r1, C r1 represents the energy influence coefficient of region r1, M r1 represents the quality parameter of region r1, D r1 represents the propagation distance of region r1 on the deformation transfer path, and R represents the total number of regions.

[0041] As a further embodiment of the present invention, the method further comprises:

[0042] S5: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress release rate at the fracture point, analyze the deformation gradient distribution, identify the main direction of fracture extension, locate the maximum strain gradient mutation domain, and output the development trend parameters;

[0043] The development trend parameters include stress change, stress release rate, expansion direction, and gradient mutation domain.

[0044] As a further solution of the present invention, the step of obtaining the development trend parameter is specifically as follows:

[0045] S501: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress distribution data of the differentiated positions in the region, summarize the stress gradient, and calculate the stress release rate of multiple positions according to the stress field distribution characteristics to obtain the stress release rate of the fracture point;

[0046] S502: calling the stress release rate of the fracture point, calculating the stress release gradient change along the fracture area, summarizing the deformation rate change of the differentiated area, and analyzing the gradient distribution characteristics, using the formula:

[0047]

[0048] Calculate the deformation gradient distribution and identify the deformation gradient distribution characteristics within the region;

[0049] Among them, G s represents the deformation gradient distribution characteristics, Δσ i represents the stress change at the i-th point along the x direction, Δx i represents the distance of the i-th point along the x direction, Δσ j represents the stress change at the jth point along the y direction, Δy j represents the distance of the jth point along the y direction, v k represents the deformation rate at the kth position, p is the number of calculation points for the deformation rate, and X and Y represent the number of calculation points along the x and y directions, respectively;

[0050] S503: calling the deformation gradient distribution feature, screening the deformation gradient mutation area, calculating the change trend of the stress release rate in the differentiated direction, identifying the main direction of fracture extension, locating the maximum strain gradient mutation area, and outputting the development trend parameters.

[0051] A geotechnical stability prediction system based on machine learning, wherein the geotechnical stability prediction system based on machine learning is used to execute the above-mentioned geotechnical stability prediction method based on machine learning, and the system comprises:

[0052] The time series feature extraction module monitors the shear stress time series data and the displacement time series data, calculates the stress difference between adjacent time steps, selects the time period when the stress difference exceeds the stress change threshold, extracts the displacement data of the corresponding time period, calculates the shear deformation gradient change rate, identifies the time point of strain direction mutation, intercepts the energy dissipation rate time series interval according to the time point of strain direction mutation, and generates the key variable interval;

[0053] The key parameter generation module calculates the second-order derivative value of the displacement-stress curve based on the key variable interval, extracts the displacement mutation point where the second-order derivative value exceeds the curvature threshold, calculates the drop rate difference after the displacement mutation point corresponds to the stress peak, selects the section where the rate difference exceeds the softening conversion threshold, superimposes the unit volume energy dissipation rate fluctuation value of the corresponding section, extracts the stress, displacement, and strain parameters at the peak time point of the energy dissipation rate fluctuation, compares the strain gradient difference before and after the peak time point, and generates a characteristic parameter set;

[0054] The spatial gradient analysis module extracts the local unit strain increment time series data according to the characteristic parameter set, calculates the difference of adjacent unit strain increments, analyzes the difference gradient distribution direction, calculates the strain increment propagation rate in the stress concentration area, locates the main strain diffusion path by combining the strain increment propagation direction and the rate difference, and generates spatial distribution parameters;

[0055] The co-evolution modeling module calls the inter-regional strain increment time series data within the spatial distribution parameters, calculates the synchronous volatility of the inter-regional strain increment, screens the adjacent areas whose synchronous volatility exceeds the interactive influence threshold, extracts the energy dissipation rate time series data of the screened areas, calculates the propagation direction and rate of the energy dissipation rate, and adjusts the boundary range of the high-value energy dissipation area in combination with the coordinates of the sliding starting point in the historical data of slope sliding to generate co-evolution parameters;

[0056] The fracture trend prediction module extracts the time series data of stress drop rate in the fracture zone based on the co-evolution parameters, calculates the difference of stress release in adjacent time steps, selects the time points when the difference exceeds the fracture triggering threshold, locates the deformation gradient distribution direction at the corresponding time points, calculates the strain gradient mutation value of the unit at the outer edge of the fracture zone, extracts the unit coordinates whose mutation value exceeds the extension threshold, determines the main direction of fracture extension according to the difference of the angle between the unit coordinate distribution direction and the energy dissipation rate propagation direction, and generates the fracture extension vector.

[0057] Compared with the prior art, the advantages and positive effects of the present invention are:

[0058] In the present invention, a refined stability prediction system is constructed through dynamic screening mechanism and multi-dimensional feature analysis. Based on the collaborative identification mechanism of stress increment mutation period and strain direction mutation point, the extraction accuracy of key variable interval is enhanced, and the critical characteristics of the evolution of stress state inside rock and soil are effectively captured. The cross-validation method of the second-order derivative of displacement-stress curve and the change of energy dissipation rate breaks through the limitations of traditional single-dimensional parameter analysis, combines the dynamic identification of hardening-softening transition zone and the comparison of multi-stage strain gradient differences, and forms a characteristic parameter system with spatiotemporal correlation. The spatial gradient evolution tracking technology of local strain increment converts static parameter analysis into dynamic diffusion process monitoring, and realizes the accurate characterization of stress diffusion path through the collaborative analysis of inter-regional mechanical interaction quantitative model and energy dissipation trend. The main direction identification technology of fracture extension integrates the multi-dimensional verification mechanism of stress release characteristics and deformation gradient distribution, establishes the correlation model of microscopic fracture mechanism and macroscopic instability trend, and significantly improves the spatiotemporal continuity of prediction results. The full-process data coupling mechanism enables the model to have adaptive feature extraction and dynamic feedback capabilities, and achieves substantial breakthroughs in early warning sensitivity and spatial parameter derivation accuracy compared with traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with 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 intended to limit the present invention.

[0061] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0062] Embodiment 1

[0063] See also Figure 1The present invention provides a technical solution: a method for predicting geotechnical stability based on machine learning, comprising the following steps:

[0064] S1: Extract the time series data of shear stress, displacement, and energy dissipation rate, calculate the stress change rate of adjacent time steps, screen the stress increment mutation period, calculate the shear deformation gradient change rate, screen the strain direction mutation point, and obtain the key variable interval;

[0065] S2: Based on the key variable interval, calculate the second-order derivative of the displacement-stress curve, extract the stress corresponding to the displacement mutation point, calculate the drop rate after the stress peak, screen the hardening-softening transition zone, calculate the change in the energy dissipation rate per unit volume, extract the stress / displacement / strain parameters at the time point of maximum energy loss, compare the strain gradient differences in multiple stages, and form a set of characteristic parameters;

[0066] S3: Based on the characteristic parameter set, calculate the local unit strain increment, analyze the spatial gradient change trend, calculate the strain diffusion rate in the stress area, identify the stress diffusion path, and derive the spatial distribution parameters in combination with the deformation gradient change;

[0067] S4: Based on the spatial distribution parameters, calculate the correlation coefficient of strain increment between regions, screen the strong interactive influence area, analyze the energy dissipation diffusion trend, construct the deformation transfer path, call the slope sliding data, calculate the diffusion range of the high-value energy dissipation area, correct the energy influence coefficient, and generate the co-evolution parameters;

[0068] S5: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress release rate at the fracture point, analyze the deformation gradient distribution, identify the main direction of fracture extension, locate the maximum strain gradient mutation domain, and output the development trend parameters.

[0069] The key variable intervals include stress mutation periods, strain mutation points, and shear gradient mutation points. The characteristic parameter sets include displacement mutation stress, stress drop rate, hardening-softening zone, and energy loss point. The spatial distribution parameters include strain increment, gradient change, stress diffusion, and deformation gradient. The co-evolution parameters include strain correlation, interactive influence zone, energy diffusion, and energy coefficient. The development trend parameters include stress change, stress release rate, expansion direction, and gradient mutation domain.

[0070] The specific steps for obtaining the key variable interval are:

[0071] S101: Based on the time series data of shear stress, displacement and energy dissipation rate, the shear stress change rate of adjacent time steps is calculated to identify the sudden change period of stress increment. By calculating the mean and standard deviation of the shear stress change rate, the threshold interval exceeding the set multiple is screened to obtain the stress sudden change period;

[0072] First, the shear stress value τ at each time step needs to bet Perform statistics and calculate the shear stress change rate Δτ of adjacent time steps t , the rate of change is given by Δτ t =τ t -τ t-1 The calculated shear stress change rate time series {Δτ t}, and use the mean μ Δτ and standard deviation σ Δτ Statistical analysis was performed, and the mean was calculated as follows: The formula for calculating the standard deviation is Where N is the total number of time steps, then set the mutation threshold δ of the shear stress change rate τ , generally take k times the standard deviation, that is, δ τ = k·σ Δτ , usually k can be 3 or 5 to ensure the significance of the mutation period, and then screen for those that satisfy |Δτ t |>δ τ The time step t constitutes the stress mutation period. For example, in some experimental data, μ Δτ =0.5MPa,σ Δτ =0.2MPa, if k=3, then δ τ =0.6MPa, so in the time series data, all the conditions satisfying |Δτ t The time step of |>0.6MPa is marked as the stress mutation period, and finally multiple discrete time intervals [t start ,t end ], these intervals represent the periods when shear stress changes suddenly.

[0073] S102: calling the stress mutation period, calculating the change rate of the shear deformation gradient, comparing the time series of the change rate, determining the key time node of the strain direction change according to the direction change characteristics of the gradient change rate, and obtaining the strain direction mutation point;

[0074] The specific calculation method is: Where Δu is the local shear displacement, h is the shear layer thickness, and the shear deformation gradient change rate of adjacent time steps is calculated And construct the time series Subsequent analysis The direction of the change is to judge its sign change. If in the adjacent time step The sign of changes, that is The time step t is recorded as the mutation point of the strain direction. For example, assuming that the shear deformation gradient γ of a certain experimental data is t The values ​​of time t are 0.05, 0.08, 0.12, and 0.07 (unit: mm), corresponding to The calculated values ​​are 0.03, 0.04, and -0.05 respectively, at time step t = 4 From positive to negative, it meets the mutation judgment condition, so this moment is recorded as the mutation point of strain direction.

[0075] S103: Call the stress mutation period and strain direction mutation point, determine the interval range of key variables, calculate the coupling relationship between the shear stress increment, shear deformation gradient change rate and energy dissipation rate, and use the formula:

[0076]

[0077] Calculate the variation characteristics of the key variable interval to obtain the key variable interval;

[0078] Among them, Δτ t1 represents the shear stress increment at the t1th time step, represents the shear deformation gradient change rate at the t1th time step, represents the rate of change of energy dissipation rate at the t1th time step, C is a small positive number to avoid the denominator approaching zero, K represents the characteristic measure of the change in the key variable interval, and T represents the total number of time steps in the key variable interval.

[0079] formula:

[0080]

[0081] For calculation, assume that the experimental data is as shown in the following table:

[0082] Table 1 Key variable time series data table

[0083]

[0084]

[0085] calculate have to;

[0086] 0.8×0.02+1.2×0.03+0.6×0.02+1.5×0.04

[0087] =0.016+0.036+0.012+0.06=0.124;

[0088] calculate have to

[0089]

[0090] Assume C = 0.1, then

[0091]

[0092] The results show that the coupling relationship between the shear stress increment and the shear deformation gradient change rate has significant characteristics under the influence of the change in energy dissipation rate, and finally the change characteristic metric K = 0.0329 is obtained in the key variable interval.

[0093] The specific steps for obtaining the feature parameter set are:

[0094] S201: Based on the key variable interval, the second-order derivative of the displacement-stress curve is calculated, the mutation point of the second-order derivative is identified, and the stress value corresponding to the mutation point is extracted to obtain a set of stress mutation points;

[0095] First, the second-order derivative of the displacement-stress curve is calculated based on the key variable interval. Specifically, assuming that the displacement-stress data set is {(ε i ,σ i )}, where ε i is the strain of the ith data point, σ i is the corresponding stress value, then the calculation formula of the second-order derivative is as follows:

[0096]

[0097] Traverse all data points, calculate the second-order derivative, and set the change threshold δ. The point is determined to be a mutation point. The selection of the threshold δ can be based on 2-3 times the global second-order derivative mean. For example, if the global mean μ is calculated to be 5.6, then the threshold δ is 11.2. For the data in the following table:

[0098]

[0099] Among them, the second-order derivative reaches 15.1 at ε = 0.003, exceeding the threshold of 11.2, so this value is included in the set of stress mutation points. This result shows that at ε = 0.003, a stress mutation occurs inside the material, and this point is used to define the subsequent softening area.

[0100] S202: calling a set of stress mutation points, calculating the drop rate after the stress peak, and screening a hardening-softening transition zone, calculating the change in energy dissipation rate per unit volume in the transition zone, and obtaining a set of energy dissipation rate changes;

[0101] Calculate the rate of decrease after the stress peak:

[0102]

[0103] Assuming the peak stress is 250MPa, the minimum stress after the mutation point is 180MPa, the peak value corresponds to a strain of 0.004, and the minimum stress corresponds to a strain of 0.006, then:

[0104]

[0105] The drop rate threshold γ is set to 30MPa / mm, and all areas with a drop rate greater than γ are defined as the hardening-softening transition zone. In this area, the energy dissipation rate per unit volume is calculated:

[0106]

[0107] Assuming that the calculated value is E = 120 J / m3, the maximum energy dissipation rate is E max =150J / m3, minimum value E min =100J / m3, then:

[0108]

[0109] This result shows that in the transition zone, the material enters a softening stage after the stress peak, and the energy loss rate reaches 50%, which is used to judge the fracture trend of the material.

[0110] S203: Call the energy dissipation rate change set, extract the maximum energy loss time point, calculate the stress, displacement, and strain parameters at the time point, analyze the strain gradient difference in the differentiation stage, and use the formula:

[0111]

[0112] The strain gradient variation is calculated, and the strain gradient in the differentiation stage is compared to obtain a set of characteristic parameters;

[0113] Among them, ΔG represents the change of strain gradient, Δσ s1 represents the stress increment in stage s1, Δε s1 represents the strain increment in stage s1, E s1 Represents the energy dissipation rate per unit volume in stage s1, W s1 represents the displacement of the s1th stage, and S represents the total number of stages.

[0114] Calculate the strain gradient change:

[0115]

[0116] Among them, Table 2 lists the parameters required for calculation:

[0117] Table 2 Strain gradient calculation parameters

[0118] Stage 1 <![CDATA[Δσ s1 (MPa)]]> <![CDATA[Δε s1 ]]> <h2 style=";text-align:left;direction:ltr"><![CDATA[E <h2 style=";text-align:left;direction:ltr"> s1 <h2 style=";text-align:left;direction:ltr"> (J / m3)]]><h2 style=";text-align:left;direction:ltr"> <![CDATA[W s1 (mm)]]> 1 50 0.002 120 0.5 2 40 0.003 110 0.6 3 30 0.004 100 0.7

[0119] Calculate the strain gradient change:

[0120]

[0121] Final result:

[0122] ΔG = 60.8;

[0123] The results show that the strain gradient change is high, indicating that the internal structural heterogeneity of the material has increased and entered a severe deformation stage, which can be used to determine the critical area of ​​failure. The final characteristic parameter set is used to analyze the plastic-brittle transition and serves as an important basis for material fracture prediction.

[0124] The specific steps for obtaining spatial distribution parameters are as follows:

[0125] S301: Based on the characteristic parameter set, the local unit strain increment is calculated, and the local deformation parameter is obtained according to the difference between the initial coordinates and the deformed coordinates of the multiple units. At the same time, the strain increment data is called to calculate the displacement vector between adjacent units. According to the relationship between the strain increment of the multiple units and the strain change of the adjacent units, the spatial gradient is calculated to obtain the local strain gradient change value;

[0126] First, determine the initial coordinates of the local unit (x i0 ,y i0 ,z i0 ) and record the coordinates after deformation (x i1 ,y i1 ,z i1 ), compare the two sets of coordinate data, and calculate the deformation of each unit one by one This value reflects the actual deformation degree of the unit. For example, the initial coordinates of a unit are (5.0, 2.0, 1.0), and the coordinates after deformation are (5.2, 2.1, 1.1). The deformation variable is calculated as follows:

[0127]

[0128] Then according to the strain increment definition Assume that the initial length of the unit is l 0 =5.0, then

[0129]

[0130] The result shows that the unit deformed by 4.898% relative to its initial length during the deformation process. This value will be used to construct the strain change relationship between multiple units when calculating the strain gradient later.

[0131] Next, call the strain increment data, analyze the displacement changes between adjacent units, and calculate the displacement vector in

[0132] dx=x i1 -x i0 ,dy=y i1 -y i0 ,dz=z i1-z i0 ;

[0133] Furthermore, based on the strain increment relationship of multiple units, the spatial gradient is calculated:

[0134]

[0135] Among them, ε i+1 and ε i Represent the strain increment of adjacent units, Δx is the distance between two units, assuming that the distance between adjacent units is 0.5, then

[0136]

[0137] The result shows that there is a strain gradient change value of 0.00604 between adjacent units, which will be used in the subsequent steps to calculate the strain diffusion rate to measure the strain diffusion characteristics of the local area.

[0138] S302: Call the local strain gradient change value to calculate the strain diffusion rate in the stress area using the formula:

[0139]

[0140] Calculate the diffusion effect coefficient of the unit and obtain the strain diffusion rate distribution;

[0141] Among them, D eff represents the diffusion effect coefficient, represents the local strain gradient change value of the u1th unit, U represents the number of calculation units, γ u1 represents the stress non-uniformity parameter of the u1th unit, represents the maximum local strain gradient change value among all elements, α represents the overall strain distribution adjustment coefficient, and β represents the maximum gradient offset correction parameter;

[0142] First determine the diffusion effect coefficient D eff :

[0143]

[0144] Where: U = 5 (select 5 units for calculation); γ u1 is the stress non-uniformity parameter of the unit, let γ u1 In order

[0145] [0.1,0.12,0.09,0.11,0.13];

[0146] calculate

[0147]

[0148] average value:

[0149]

[0150] Assume that the maximum strain gradient change value Parameters α = 1.5, β = 0.002;

[0151]

[0152] D eff =0.00541×0.00548=0.0000297;

[0153] The result shows that the diffusion effect coefficient of local strain is 0.0000297, which will be used to calculate the diffusion path in the next step to analyze the propagation direction and intensity of the strain.

[0154] S303: Call the strain diffusion rate distribution, identify the stress diffusion path, calculate the main diffusion direction according to the strain gradient direction of multiple units, compare the strain diffusion rate in combination with the deformation gradient change, classify the diffusion area according to the strain gradient change value, and derive the spatial distribution parameters.

[0155] First, calculate the strain gradient direction of multiple units and define the gradient direction vector:

[0156]

[0157] Assuming that the strain gradients of a unit in three directions are 0.006, 0.005, and 0.004 respectively, then

[0158]

[0159] After normalization:

[0160]

[0161] The results show that the main diffusion direction of the unit tends to be in the x-axis direction and is affected by the y and z directions. This result can be used to guide the classification of diffusion areas and path derivation.

[0162] Finally, according to the diffusion area division, the spatial distribution parameters are derived to provide support for the subsequent strain field analysis.

[0163] The specific steps for obtaining the co-evolution parameters are as follows:

[0164] S401: Based on the spatial distribution parameters, the strain increment correlation coefficient between regions is calculated, and the regions where the strain increment correlation coefficient is greater than the set threshold are screened. At the same time, the strain increment data matrix is ​​called to extract the index number of the region that meets the conditions, and the strain increment correlation coefficient corresponding to the region is extracted according to the index number, and the interaction matrix between regions is constructed to obtain the index set of the strong interaction influence area;

[0165] The correlation coefficient is calculated by comparing the strain increment difference between two regions. For adjacent regions i and j, the correlation coefficient is calculated as follows:

[0166]

[0167] Among them, Δε i and Δε j Represent the strain increment of region i and region j respectively, max(Δε) represents the maximum strain increment in the study area, and this formula is used to normalize the strain increment correlation coefficient between different regions. Next, set the threshold T c (usually in the range of 0.7-0.9) to screen out areas with correlation coefficients greater than the threshold. These areas have a strong strain transfer relationship during geological deformation to ensure that the calculated area has a high coupling. The index number of the screened area is stored in the array R c = {r 1 ,r 2 ,...,r n}, and according to R c Extract the strain increment correlation coefficient of the corresponding area and construct the interaction influence matrix M between regions ij , each element M of this matrix ij represents the strain increment correlation coefficient between regions i and j. Finally, the region index set R with the largest interaction influence is selected. s ,These regions constitute the index set of strong interaction influence regions.

[0168] Table 3 shows an example data set used to calculate the strain increment correlation coefficient and region screening:

[0169] Table 3 Calculation example of strain increment correlation coefficient

[0170]

[0171] As shown in Table 3, regions A1, A2, and A4 meet the screening conditions (C ij >0.7), and thus was included in the strong interaction influence region index set R s .

[0172] S402: calling the strong interaction influence zone index set, analyzing the diffusion trend of energy dissipation, calculating the energy dissipation increment of multiple regions, and dividing the connection relationship between regions according to the interaction influence intensity based on the interaction influence matrix between regions, extracting the region index on the deformation path, calculating the energy dissipation diffusion amount of multiple regions on the path, and obtaining the energy diffusion value of the deformation transfer path;

[0173] According to the obtained strong interaction influence area index set R s , further analyze the energy dissipation diffusion trend of these areas, calculate the energy dissipation increment of each area, and the calculation method of energy dissipation increment is:

[0174]

[0175] in, and They represent the energy dissipation value of region r at two adjacent moments, based on the interaction matrix M between regions. ij , determine the inter-regional connectivity according to the intensity of the interaction. The specific method is: if M ij Greater than the set energy interaction threshold T E (usually 0.6-0.8), it is considered that regions i and j have an energy interaction relationship, and they are connected into a deformation path to construct the energy dissipation transfer path P.

[0176] The regional index on the deformation path is extracted, and the energy dissipation diffusion amount of multiple regions in the path is calculated. The energy diffusion amount is calculated as follows:

[0177]

[0178] Among them, ΔE r is the energy dissipation increment of region r, d r is the propagation distance of region r on the path. This formula is used to measure the energy diffusion intensity on the deformation transfer path. After calculation, the energy diffusion value of the deformation transfer path is obtained.

[0179] S403: Call the energy diffusion value of the deformation transfer path, combine it with the slope sliding data, calculate the diffusion range of the high-value energy dissipation area, and correct the energy influence coefficient. The formula is:

[0180]

[0181] Calculate and obtain the co-evolution parameters;

[0182] Among them, E syn represents the coevolution parameter, ΔE r1 represents the energy dissipation increment of region r1, A r1 represents the spatial area of ​​region r1, R r1represents the dissipation rate of region r1 during the energy dissipation process, S r1 represents the sliding stability coefficient of region r1, C r1 represents the energy influence coefficient of region r1, M r1 represents the quality parameter of region r1, D r1 represents the propagation distance of region r1 on the deformation transfer path, and R represents the total number of regions.

[0183] Set the sliding stability coefficient S r The calculation formula is:

[0184]

[0185] Among them, F r is the anti-slip force of area r, μ r is the slip moment of region r, and at the same time, the energy influence coefficient C is set r The calculation method is:

[0186]

[0187] Among them, E r is the energy dissipation value of region r, ∑E is the total energy dissipation value of all regions, and this coefficient is used to measure the energy proportion of each region. Substitute the calculated parameters into the co-evolution parameter formula:

[0188]

[0189] Set up an actual example, assuming that the data of a certain area is as follows: ΔE r1 =5000J, A r1 =200m 2 , R r1 =0.85, S r1 =1.2, C r1 =0.15, M r1 =300kg, D r1 =50m

[0190] The calculation is as follows:

[0191]

[0192] |29.41-0.833|=28.577;

[0193]

[0194] E syn =28.577×0.15×2.449=10.497;

[0195] The final co-evolution parameter calculation result is E syn=10.497. This result shows that the coevolution intensity in this area is high and interacts significantly with the energy dissipation of the slope deformation process.

[0196] The specific steps for obtaining development trend parameters are as follows:

[0197] S501: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress distribution data of the differentiated positions in the region, summarize the stress gradient, and calculate the stress release rate of multiple positions according to the stress field distribution characteristics to obtain the stress release rate of the fracture point;

[0198] It is necessary to measure the stress distribution at different locations in the area first. The measurement can use strain sensors arranged at multiple points to obtain initial stress data. For example, 50 monitoring points at different depths and different geological conditions along the fault zone are selected. The stress measurement value of each monitoring point can be expressed as σ i These measurements can be obtained by using strain gauges combined with borehole pressure measurement. After obtaining the initial data, the stress difference Δσ between adjacent monitoring points is calculated. i =σ i+1 -σ i , and divided by the corresponding distance Δx i To obtain the stress gradient, By traversing all the measuring points, the stress gradient field of the entire fracture area can be formed. Then, according to the changing trend of stress distribution, the stress release rate in the area is calculated. The specific method is to monitor the stress release rate at different time points t 1 and t 2 The stress value and the rate of change of stress with time are calculated For a certain monitoring point, t 1 =0 hours, σ=5MPa, t 2 = 24 hours, σ = 4.8MPa, then the stress release rate In this way, the stress release rate of all monitoring points is calculated, and its spatial variation law is summarized, thereby obtaining the stress release rate of the fracture point.

[0199] S502: Call the stress release rate at the fracture point, calculate the stress release gradient change along the fracture zone, summarize the deformation rate change in the differentiated area, and analyze the gradient distribution characteristics, using the formula:

[0200]

[0201] Calculate the deformation gradient distribution and identify the deformation gradient distribution characteristics within the region;

[0202] Among them, G s represents the deformation gradient distribution characteristics, Δσ i represents the stress change at the i-th point along the x direction, Δx irepresents the distance of the i-th point along the x direction, Δσ j represents the stress change at the jth point along the y direction, Δy j represents the distance of the jth point along the y direction, v k represents the deformation rate at the kth position, p is the number of calculation points for the deformation rate, and X and Y represent the number of calculation points along the x and y directions, respectively;

[0203] For example, 50 monitoring points are selected, and the deformation rate v of each point is k It can be obtained through GPS measurement or surface deformation monitoring, and its average value can be calculated Assume the deformation rate data is as follows:

[0204] Table 4 Deformation rate table of monitoring points

[0205] Monitoring point number Deformation rate (mm / d) A1 2.1 A2 2.5 A3 1.8 A4 3.0 A5 2.3

[0206] As shown in Table 4, assuming a total of 50 monitoring points, the average deformation rate is calculated For example, if the total deformation rate is 120 mm / d, then the average deformation rate is Combined with the stress gradient calculation formula:

[0207]

[0208] Calculate the deformation gradient distribution, for example, along the X direction at 5 points The calculated values ​​are

[0209] 0.3, 0.25, 0.4, 0.35, 0.28 MPa / m, 5 points along the Y direction The calculated values ​​are 0.2, 0.3, 0.33, 0.27, 0.29 MPa / m respectively, then:

[0210]

[0211] This value is used to characterize the deformation gradient characteristics within the region.

[0212] S503: Call the deformation gradient distribution characteristics, screen the deformation gradient mutation area, calculate the change trend of the stress release rate in the differentiated direction, identify the main direction of fracture extension, locate the maximum strain gradient mutation area, and output the development trend parameters.

[0213] Based on the deformation gradient feature G calculated above s , further screen the deformation gradient mutation area, the screening criterion can be set as ΔG s The area with a value > 0.5 was taken as the mutation area, and the G of all monitoring points was compared. sValues, extract high value areas and calculate the change trend of stress release rate in different directions. For example, for monitoring points A1, A2, and A3, the stress release rates are -0.008, -0.007, and -0.0065 MPa / h, respectively. Then calculate the change rate in their directions:

[0214]

[0215] In this way, the gradient changes of all regions are traversed to identify the mutation area of ​​deformation gradient. If the gradient change rate of the adjacent area of ​​a point or If it is greater than a preset threshold, for example, more than 0.003MPa / km, the area can be considered as a stress mutation zone. Combined with the directionality of fracture extension, the maximum strain gradient mutation area is located and the corresponding development trend parameters are output.

[0216] A geotechnical stability prediction system based on machine learning, the geotechnical stability prediction system based on machine learning is used to execute the above-mentioned geotechnical stability prediction method based on machine learning, and the system includes:

[0217] The time series feature extraction module monitors the shear stress time series data and the displacement time series data, calculates the stress difference between adjacent time steps, selects the time period when the stress difference exceeds the stress change threshold, extracts the displacement data of the corresponding time period, calculates the shear deformation gradient change rate, identifies the time point of strain direction mutation, intercepts the energy dissipation rate time series interval according to the time point of strain direction mutation, and generates the key variable interval;

[0218] The key parameter generation module calculates the second-order derivative value of the displacement-stress curve based on the key variable interval, extracts the displacement mutation point where the second-order derivative value exceeds the curvature threshold, calculates the difference in the rate of decrease after the stress peak corresponding to the displacement mutation point, selects the section where the rate difference exceeds the softening conversion threshold, superimposes the unit volume energy dissipation rate fluctuation value of the corresponding section, extracts the stress, displacement, and strain parameters at the peak time of the energy dissipation rate fluctuation, compares the strain gradient difference before and after the peak time point, and generates a characteristic parameter set;

[0219] The spatial gradient analysis module extracts the local unit strain increment time series data according to the characteristic parameter set, calculates the difference of adjacent unit strain increments, analyzes the difference gradient distribution direction, calculates the strain increment propagation rate in the stress concentration area, locates the main strain diffusion path by combining the strain increment propagation direction and rate difference, and generates spatial distribution parameters;

[0220] The co-evolution modeling module calls the time series data of strain increments between regions within the spatial distribution parameters, calculates the synchronous volatility of strain increments between regions, screens the adjacent regions whose synchronous volatility exceeds the threshold of interaction influence, extracts the time series data of energy dissipation rate in the screened regions, calculates the propagation direction and rate of energy dissipation rate, and adjusts the boundary range of the high-value energy dissipation area in combination with the coordinates of the sliding starting point in the historical data of slope sliding to generate co-evolution parameters;

[0221] The fracture trend prediction module extracts the time series data of stress drop rate in the fracture zone based on the co-evolution parameters, calculates the difference of stress release in adjacent time steps, selects the time points when the difference exceeds the fracture triggering threshold, locates the deformation gradient distribution direction at the corresponding time points, calculates the strain gradient mutation value of the outer edge unit of the fracture zone, extracts the unit coordinates whose mutation value exceeds the extension threshold, determines the main direction of fracture extension according to the angle difference between the unit coordinate distribution direction and the energy dissipation rate propagation direction, and generates the fracture extension vector.

[0222] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting geotechnical stability based on machine learning, characterized in that: The following steps are involved: S1: Extract the time series data of shear stress, displacement, and energy dissipation rate, calculate the stress change rate of adjacent time steps, screen the stress increment mutation period, calculate the shear deformation gradient change rate, screen the strain direction mutation point, and obtain the key variable interval; S2: Based on the key variable interval, calculate the second-order derivative of the displacement-stress curve, extract the stress corresponding to the displacement mutation point, calculate the drop rate after the stress peak, screen the hardening-softening transition zone, calculate the change in the energy dissipation rate per unit volume, extract the stress / displacement / strain parameters at the time point of maximum energy loss, compare the strain gradient differences in multiple stages, and form a set of characteristic parameters; S3: According to the characteristic parameter set, calculate the local unit strain increment, analyze the spatial gradient change trend, calculate the strain diffusion rate of the stress area, identify the stress diffusion path, and derive the spatial distribution parameters in combination with the deformation gradient change; S4: Based on the spatial distribution parameters, calculate the correlation coefficient of strain increments between regions, screen the strong interactive influence areas, analyze the energy dissipation diffusion trend, construct the deformation transfer path, call the slope slip data, calculate the diffusion range of the high-value energy dissipation area, correct the energy influence coefficient, and generate the co-evolution parameters.

2. The method for predicting geotechnical stability based on machine learning according to claim 1, characterized in that: The key variable interval includes stress mutation period, strain mutation point, and shear gradient mutation point. The characteristic parameter set includes displacement mutation stress, stress drop rate, hardening-softening zone, and energy loss point. The spatial distribution parameters include strain increment, gradient change, stress diffusion, and deformation gradient. The co-evolution parameters include strain correlation, interactive influence zone, energy diffusion, and energy coefficient.

3. The method for predicting geotechnical stability based on machine learning according to claim 2, characterized in that: The steps for obtaining the key variable interval are specifically as follows: S101: Based on the time series data of shear stress, displacement and energy dissipation rate, the shear stress change rate of adjacent time steps is calculated to identify the sudden change period of stress increment. By calculating the mean and standard deviation of the shear stress change rate, the threshold interval exceeding the set multiple is screened to obtain the stress sudden change period; S102: calling the stress mutation period, calculating the change rate of the shear deformation gradient, comparing the time series of the change rate, determining the key time node of the strain direction change according to the direction change characteristics of the gradient change rate, and obtaining the strain direction mutation point; S103: Calling the stress mutation period and the strain direction mutation point, determining the interval range of key variables, calculating the coupling relationship between the shear stress increment, the shear deformation gradient change rate and the energy dissipation rate, using the formula: Calculate the variation characteristics of the key variable interval to obtain the key variable interval; Among them, Δτ t1 represents the shear stress increment at the t1th time step, represents the shear deformation gradient change rate at the t1th time step, represents the rate of change of energy dissipation rate at the t1th time step, C is a small positive number to avoid the denominator approaching zero, K represents the characteristic measure of the change in the key variable interval, and T represents the total number of time steps in the key variable interval.

4. The method for predicting geotechnical stability based on machine learning according to claim 3, characterized in that: The steps for obtaining the characteristic parameter set are specifically as follows: S201: Based on the key variable interval, calculate the second-order derivative of the displacement-stress curve, identify the mutation point of the second-order derivative, and extract the stress value corresponding to the mutation point to obtain a set of stress mutation points; S202: calling the stress mutation point set, calculating the drop rate after the stress peak, and screening the hardening-softening transition zone, calculating the change of the energy dissipation rate per unit volume in the transition zone, and obtaining the energy dissipation rate change set; S203: calling the energy dissipation rate change set, extracting the maximum energy loss time point, calculating the stress, displacement, and strain parameters at the time point, analyzing the strain gradient difference in the differentiation stage, and using the formula: The strain gradient variation is calculated, and the strain gradient in the differentiation stage is compared to obtain a set of characteristic parameters; Among them, ΔG represents the change of strain gradient, Δσ s1 represents the stress increment in stage s1, Δε s1 represents the strain increment in stage s1, E s1 Represents the energy dissipation rate per unit volume in stage s1, W s1 represents the displacement of the s1th stage, and S represents the total number of stages.

5. The method for predicting geotechnical stability based on machine learning according to claim 4, characterized in that: The steps for obtaining the spatial distribution parameters are specifically as follows: S301: Based on the characteristic parameter set, the local unit strain increment is calculated, and the local deformation parameter is obtained according to the difference between the initial coordinates and the deformed coordinates of the multiple units. At the same time, the strain increment data is called to calculate the displacement vector between adjacent units. According to the relationship between the strain increment of the multiple units and the strain change of the adjacent units, the spatial gradient is calculated to obtain the local strain gradient change value; S302: Call the local strain gradient change value to calculate the strain diffusion rate in the stress area using the formula: Calculate the diffusion effect coefficient of the unit and obtain the strain diffusion rate distribution; Among them, D eff represents the diffusion effect coefficient, represents the local strain gradient change value of the u1th unit, U represents the number of calculation units, γ u1 represents the stress non-uniformity parameter of the u1th unit, represents the maximum local strain gradient change value among all elements, α represents the overall strain distribution adjustment coefficient, and β represents the maximum gradient offset correction parameter; S303: calling the strain diffusion rate distribution, identifying the stress diffusion path, calculating the main diffusion direction according to the strain gradient direction of multiple units, comparing the strain diffusion rate in combination with the deformation gradient change, classifying the diffusion area according to the strain gradient change value, and deriving the spatial distribution parameters.

6. The method for predicting geotechnical stability based on machine learning according to claim 5, characterized in that: The steps for obtaining the co-evolution parameters are specifically as follows: S401: Based on the spatial distribution parameters, calculate the strain increment correlation coefficient between regions, screen the regions where the strain increment correlation coefficient is greater than the set threshold, call the strain increment data matrix at the same time, extract the index number of the region that meets the conditions, extract the strain increment correlation coefficient corresponding to the region according to the index number, construct the interaction matrix between regions, and obtain the index set of the strong interaction influence area; S402: calling the strong interaction influence zone index set, analyzing the diffusion trend of energy dissipation, calculating the energy dissipation increment of multiple regions, and dividing the connection relationship between regions according to the interaction influence intensity based on the interaction influence matrix between regions, extracting the region index on the deformation path, calculating the energy dissipation diffusion amount of multiple regions on the path, and obtaining the energy diffusion value of the deformation transfer path; S403: calling the energy diffusion value of the deformation transfer path, combining with the slope sliding data, calculating the diffusion range of the high-value energy dissipation area, and correcting the energy influence coefficient, using the formula: Calculate and obtain the co-evolution parameters; Among them, E syn represents the coevolution parameter, ΔE r1 represents the energy dissipation increment of region r1, A r1 represents the spatial area of ​​region r1, R r1 represents the dissipation rate of region r1 during the energy dissipation process, S r1 represents the sliding stability coefficient of region r1, C r1 represents the energy influence coefficient of region r1, M r1 represents the quality parameter of region r1, D r1 represents the propagation distance of region r1 on the deformation transfer path, and R represents the total number of regions.

7. The method for predicting geotechnical stability based on machine learning according to claim 6, characterized in that: The method further comprises: S5: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress release rate at the fracture point, analyze the deformation gradient distribution, identify the main direction of fracture extension, locate the maximum strain gradient mutation domain, and output the development trend parameters; The development trend parameters include stress change, stress release rate, expansion direction, and gradient mutation domain.

8. The method for predicting geotechnical stability based on machine learning according to claim 7, characterized in that: The steps for obtaining the development trend parameters are specifically as follows: S501: Based on the co-evolution parameters, calculate the stress change in the fracture zone, extract the stress distribution data of the differentiated positions in the region, summarize the stress gradient, and calculate the stress release rate of multiple positions according to the stress field distribution characteristics to obtain the stress release rate of the fracture point; S502: calling the stress release rate of the fracture point, calculating the stress release gradient change along the fracture area, summarizing the deformation rate change of the differentiated area, and analyzing the gradient distribution characteristics, using the formula: Calculate the deformation gradient distribution and identify the deformation gradient distribution characteristics within the region; Among them, G s represents the deformation gradient distribution characteristics, Δσ i represents the stress change at the i-th point along the x direction, Δx i represents the distance of the i-th point along the x direction, Δσ j represents the stress change at the jth point along the y direction, Δy j represents the distance of the jth point along the y direction, v k represents the deformation rate at the kth position, p is the number of calculation points for the deformation rate, and X and Y represent the number of calculation points along the x and y directions, respectively; S503: calling the deformation gradient distribution feature, screening the deformation gradient mutation area, calculating the change trend of the stress release rate in the differentiated direction, identifying the main direction of fracture extension, locating the maximum strain gradient mutation area, and outputting the development trend parameters.

9. A geotechnical stability prediction system based on machine learning, characterized in that: According to the method for predicting geotechnical stability based on machine learning according to any one of claims 1 to 8, the system comprises: The time series feature extraction module monitors the shear stress time series data and the displacement time series data, calculates the stress difference between adjacent time steps, selects the time period when the stress difference exceeds the stress change threshold, extracts the displacement data of the corresponding time period, calculates the shear deformation gradient change rate, identifies the time point of strain direction mutation, intercepts the energy dissipation rate time series interval according to the time point of strain direction mutation, and generates the key variable interval; The key parameter generation module calculates the second-order derivative value of the displacement-stress curve based on the key variable interval, extracts the displacement mutation point where the second-order derivative value exceeds the curvature threshold, calculates the drop rate difference after the displacement mutation point corresponds to the stress peak, selects the section where the rate difference exceeds the softening conversion threshold, superimposes the unit volume energy dissipation rate fluctuation value of the corresponding section, extracts the stress, displacement, and strain parameters at the peak time point of the energy dissipation rate fluctuation, compares the strain gradient difference before and after the peak time point, and generates a characteristic parameter set; The spatial gradient analysis module extracts the local unit strain increment time series data according to the characteristic parameter set, calculates the difference of adjacent unit strain increments, analyzes the difference gradient distribution direction, calculates the strain increment propagation rate in the stress concentration area, locates the main strain diffusion path by combining the strain increment propagation direction and the rate difference, and generates spatial distribution parameters; The co-evolution modeling module calls the inter-regional strain increment time series data within the spatial distribution parameters, calculates the synchronous volatility of the inter-regional strain increment, screens the adjacent areas whose synchronous volatility exceeds the interactive influence threshold, extracts the energy dissipation rate time series data of the screened areas, calculates the propagation direction and rate of the energy dissipation rate, and adjusts the boundary range of the high-value energy dissipation area in combination with the coordinates of the sliding starting point in the historical data of slope sliding to generate co-evolution parameters; The fracture trend prediction module extracts the time series data of stress drop rate in the fracture zone based on the co-evolution parameters, calculates the difference of stress release in adjacent time steps, selects the time points when the difference exceeds the fracture triggering threshold, locates the deformation gradient distribution direction at the corresponding time points, calculates the strain gradient mutation value of the unit at the outer edge of the fracture zone, extracts the unit coordinates whose mutation value exceeds the extension threshold, determines the main direction of fracture extension according to the difference of the angle between the unit coordinate distribution direction and the energy dissipation rate propagation direction, and generates the fracture extension vector.

Images