Geological disaster susceptibility evaluation method based on machine learning model

Abstract

The application discloses a geological disaster susceptibility evaluation method based on a machine learning model and particularly relates to the field of geological disaster risk assessment, which is used to solve the problem that the existing geological disaster susceptibility evaluation method cannot effectively reflect the continuous disturbance effect of engineering activities and is disconnected with the actual engineering promotion stage. The application obtains historical engineering activity information and historical geological disaster information, calculates the spatial kernel density of different engineering activity types under the scale of a geological environment unit, constructs a geological sensitivity coefficient matrix in combination with the historical geological disaster occurrence frequency, corrects the engineering activity intensity to obtain an engineering activity influence intensity data sequence, further extracts a triggering feature index reflecting the disturbance accumulation and change characteristics of engineering activities, evaluates the geological disaster triggering potential by using a machine learning model and outputs a risk partition result, so that the dynamic evaluation of geological disaster susceptibility facing the engineering activity scene is realized.

Description

Technical Field

[0001] This invention relates to the field of geological disaster risk assessment technology, and more specifically, to a geological disaster susceptibility assessment method based on machine learning models. Background Technology

[0002] Current geological hazard susceptibility assessments primarily serve practical applications such as regional engineering construction management, mineral resource development, transportation infrastructure layout, and urban and rural planning. Their core objective is to identify and manage potential geological hazard risks in advance, given the continuous impact of engineering activities on the geological environment. With the increasing intensity of engineering construction in mountainous areas, various road construction, mining, site leveling, and underground engineering activities are becoming highly concentrated in both space and time. The disturbances to geological environmental stability caused by these activities are gradually shifting from localized, short-term impacts to long-term, cumulative effects.

[0003] However, in current practice, geological hazard susceptibility assessments often rely primarily on static geological conditions or historical hazard distributions. The characterization of engineering activities remains largely at the level of simple statistics or empirical judgments, failing to reflect the differentiated disturbance effects of different types of engineering activities in different geological environmental units. Furthermore, there is a general lack of systematic analysis of the accumulation, evolution, and amplification of disturbances caused by engineering activities over time, resulting in a disconnect between assessment results and actual engineering progress stages, making it difficult to meet the needs of dynamic risk management and forward-looking decision-making.

[0004] In the context of increasingly compressed construction cycles and ever-increasing control requirements, there is a need for a method that can integrate the spatial distribution characteristics of engineering activities, the carrying capacity of the geological environment, and historical disaster response information, and dynamically evaluate the susceptibility of geological disasters through a data-driven approach. This method can provide more refined and reliable decision-making basis for engineering planning, construction control, and risk prevention. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a geological hazard susceptibility assessment method based on machine learning models to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The geological hazard susceptibility assessment method based on machine learning models includes the following steps:

[0008] S1. Obtain historical datasets for the target area, including historical engineering activity locations and types, historical geological disaster locations, and basic geological environment data;

[0009] S2. Calculate the spatial kernel density values ​​of various historical engineering activity sites in different geological environment units according to the type of engineering activity;

[0010] S3. Conduct a correlation analysis between the spatial kernel density values ​​of various engineering activities within the geological environment unit and the frequency of historical geological disasters to generate a geological sensitivity coefficient matrix corresponding to various engineering activities.

[0011] S4. Based on the geological sensitivity coefficient matrix of various engineering activities, the intensity of engineering activity points is spatially weighted to generate a quantitative data sequence of engineering activity impact intensity.

[0012] S5. Perform time accumulation and trend analysis on the data sequence of engineering activity impact intensity of various geological environment units, extract the trigger feature indicators of the degree of engineering activity disturbance accumulation, growth rate and fluctuation amplitude, and construct the input feature training set;

[0013] S6. Label the input feature training set with geological disaster triggering status labels and input them into the machine learning-based evaluation model for training;

[0014] S7. The evaluation model outputs the geological hazard triggering potential score of each geological environment unit, classifies the hazard susceptibility risk level, and outputs the risk zoning results.

[0015] As a further aspect of the present invention, in step S1, obtaining the historical dataset of the target area, including historical engineering activity locations and types, historical geological disaster locations, and basic geological environment data, specifically includes:

[0016] Collect data on historical engineering activity sites and historical geological disaster sites within the target area;

[0017] Based on the time and intensity characteristics of the disturbance to the geological body caused by engineering activities, the engineering activity type of each historical engineering activity site is marked, and the types include continuous disturbance type, intermittent disturbance type, and concentrated burst type.

[0018] Extract basic geological environmental data of the target area, including information on rock and soil types, slope structural units, and fault distances.

[0019] Based on the rock and soil types and slope structures in the geological environment data of the target area, the target area is divided into several geological environment units;

[0020] The historical engineering activity sites, historical geological disaster sites, and basic geological environment data that have been tagged with type are registered and associated based on geological environment units, and integrated into a historical dataset.

[0021] As a further aspect of the present invention, in step S2, calculating the spatial kernel density values ​​of various historical engineering activity sites within different geological environment units according to the type of engineering activity specifically includes:

[0022] For three types of engineering activities—persistent disturbance, intermittent disturbance, and concentrated outbreak—all historical engineering activity locations for each type of engineering activity within the target area are extracted.

[0023] A spatial density raster layer covering the target area is generated using kernel density estimation.

[0024] Spatial overlay is performed between the spatial density raster layer corresponding to each type of engineering activity and the vector layer of geological environment units divided based on soil and rock type and slope structure;

[0025] For each geological environment unit, the average pixel value of the spatial density raster for each type of engineering activity within the unit is calculated, and the calculation result is used as the spatial kernel density value of that type of engineering activity within the geological environment unit.

[0026] As a further aspect of the present invention, in step S3, the spatial kernel density values ​​of various engineering activities within a geological environment unit are correlated with the frequency of historical geological disasters to generate a geological sensitivity coefficient matrix corresponding to various engineering activities. This specifically includes:

[0027] Using geological environmental units as the basic unit of analysis and the coverage area of ​​geological environmental units as the benchmark, the historical disaster occurrence frequency of each geological environmental unit is calculated and integrated into a historical disaster occurrence frequency sequence.

[0028] For all types of engineering activities, the spatial kernel density values ​​of all geological environment units are extracted and integrated into a kernel density sequence;

[0029] The correlation coefficient between the historical disaster occurrence frequency sequence and the kernel density sequence is calculated as the overall spatial correlation strength between each type of engineering activity and geological disaster;

[0030] Based on the obtained correlation coefficients and the degree of deviation of the kernel density values ​​of each geological environment unit from the average kernel density value of the kernel density sequence, the sensitivity coefficient of the geological environment unit under each type of engineering activity is calculated.

[0031] The sensitivity coefficients of all geological environment units are collected to construct a geological sensitivity coefficient matrix for different types of engineering activities.

[0032] As a further aspect of the present invention, in step S4, based on the geological sensitivity coefficient matrix of various engineering activities, the intensity of engineering activity locations is spatially weighted to generate a quantified data sequence of engineering activity impact intensity, specifically including:

[0033] Obtain the intensity of engineering activities corresponding to each historical engineering activity point;

[0034] Based on the geological environment unit to which the engineering activity site belongs, the corresponding geological sensitivity coefficient matrix is ​​selected in combination with the corresponding engineering activity type;

[0035] Multiply the intensity of engineering activities at each engineering activity site by the corresponding sensitivity coefficient to obtain the historical engineering activity impact intensity value after geological environment sensitivity correction;

[0036] The impact intensity values ​​of all historical engineering activities within each geological environment unit are calculated to obtain a data sequence of the impact intensity of historical engineering activities of different types.

[0037] As a further aspect of the present invention, in S5, the time accumulation and trend analysis of the engineering activity impact intensity data sequence of various geological environment units are performed to extract the triggering characteristic indicators of the cumulative degree of engineering activity disturbance, growth rate, and fluctuation amplitude, and the construction of the input feature training set specifically includes:

[0038] The data sequence of engineering activity impact intensity corresponding to a single geological environment unit is divided into continuous time periods according to a fixed time window;

[0039] Within each time period, the sum of the impact intensity values ​​of engineering activities falling within that time period is calculated as the cumulative disturbance, the average rate of change is calculated as the growth rate, and the standard deviation of the intensity values ​​is calculated as the fluctuation amplitude.

[0040] The calculated engineering activity type, geological environment unit identifier, cumulative disturbance, growth rate, and fluctuation amplitude of the geological environment unit in the current time period are combined with the corresponding parameter values ​​of the geological environment basic data of the geological environment unit into a data row. All data rows together constitute the input feature training set.

[0041] As a further aspect of the present invention, step S6, which involves labeling the input feature training set with geological disaster triggering state tags and inputting them into a machine learning-based evaluation model for training, specifically includes:

[0042] Based on the geological environment unit and time period corresponding to each row of data in the input feature training set, query the spatial location and occurrence time records of historical geological disaster sites;

[0043] Determine the historical geological disaster location records within the spatial range of the geological environment unit within a prediction time window of a set length after the end of the time period, and label the input feature training set with binarized trigger state labels;

[0044] The labeled input feature training set and the corresponding trigger state labels are input into the selected machine learning algorithm for model training evaluation.

[0045] As a further aspect of the present invention, in S7, based on the newly added engineering activities, the evaluation model outputs the geological hazard triggering potential score of each geological environment unit, classifies the hazard susceptibility risk level, and outputs the risk zoning results, specifically including:

[0046] For newly added engineering activities within the target area, input features are generated based on the type and location of the engineering activities, combined with the intensity of the engineering activities, and then input into the evaluation model.

[0047] The trigger discrimination score output by the assessment model is converted into the geological hazard triggering potential score of each geological environment unit. Based on the preset classification interval, the risk level is divided to obtain the geological hazard susceptibility risk zoning results.

[0048] The technical effects and advantages of the geological hazard susceptibility assessment method based on machine learning models in this invention are as follows:

[0049] By distinguishing different types of engineering activities and quantifying their spatial impact intensity at the geological environment unit scale, this study effectively characterizes the differentiated disturbance effects of engineering activities on geological stability, avoiding evaluation biases. By constructing a geological sensitivity coefficient matrix and correcting for the intensity of engineering activities, the evaluation results reflect the different responses of different geological environments to engineering disturbances. Furthermore, by combining time accumulation and trend characteristics, the study identifies the triggering potential of engineering activity disturbances, giving the evaluation results forward-looking and phased discriminative capabilities. Finally, based on a machine learning model, the study outputs triggering potential scores and performs risk zoning, effectively improving the refinement and decision support value of geological hazard susceptibility assessment. This approach is applicable to various practical application scenarios such as engineering planning, construction control, and regional risk management.

[0050] Overall, by introducing machine learning models to systematically model the relationship between engineering activity disturbances and geological environment responses, the assessment of geological hazard susceptibility has been transformed from static analysis to dynamic evaluation. Compared with traditional methods that rely solely on geological conditions or historical hazard distribution, this approach can more realistically reflect the evolutionary characteristics of geological hazard risks under conditions of continuous engineering intervention. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the geological hazard susceptibility assessment method based on a machine learning model according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1

[0054] Figure 1 The present invention provides a geological hazard susceptibility assessment method based on a machine learning model, which includes the following steps:

[0055] S1. Obtain historical datasets for the target area, including historical engineering activity locations and types, historical geological disaster locations, and basic geological environment data;

[0056] S2. Calculate the spatial kernel density values ​​of various historical engineering activity sites in different geological environment units according to the type of engineering activity;

[0057] S3. Conduct a correlation analysis between the spatial kernel density values ​​of various engineering activities within the geological environment unit and the frequency of historical geological disasters to generate a geological sensitivity coefficient matrix corresponding to various engineering activities.

[0058] S4. Based on the geological sensitivity coefficient matrix of various engineering activities, the intensity of engineering activity points is spatially weighted to generate a quantitative data sequence of engineering activity impact intensity.

[0059] S5. Perform time accumulation and trend analysis on the data sequence of engineering activity impact intensity of various geological environment units, extract the trigger feature indicators of the degree of engineering activity disturbance accumulation, growth rate and fluctuation amplitude, and construct the input feature training set;

[0060] S6. Label the input feature training set with geological disaster triggering status labels and input them into the machine learning-based evaluation model for training;

[0061] S7. The evaluation model outputs the geological hazard triggering potential score of each geological environment unit, classifies the hazard susceptibility risk level, and outputs the risk zoning results.

[0062] In step S1, historical datasets of the target area are obtained, including historical engineering activity locations and types, historical geological disaster locations, and basic geological environment data.

[0063] The data on engineering activities and geological disaster events that have occurred within the target area for geological hazard susceptibility assessment were collected and organized. Engineering activity location data were obtained through engineering construction archives, construction records, and project completion documents. All collected engineering activity locations included clear spatial location identifiers, represented using a unified coordinate system to ensure consistency in subsequent spatial matching. Simultaneously, the start and end times, duration of operations, and intensity of engineering implementation for each engineering activity location were compiled and archived. The intensity of engineering implementation was quantified through methods such as construction scale, frequency of operations, or division of construction stages; for example, the number of operations per unit time or the number of continuous operation days were used as examples of methods to represent the intensity of engineering activities. Based on regional geological hazard survey results, hazard census reports, and historical disaster records, the locations of landslides, collapses, and other geological hazards that have occurred within the target area were compiled, ensuring that each geological hazard location corresponds to a clear occurrence time and spatial location. To avoid data omissions, only hazard locations with clear occurrence times and spatial locations and complete records were selected for inclusion in the historical geological hazard dataset.

[0064] Each engineering activity site is tagged with its activity type. Based on the duration and intensity of the disturbance to the geological body, the activities are classified. Activities with long durations, continuous processes, and long-term disturbances are labeled as continuous disturbance activities, such as large-scale continuous site leveling or long-term mining operations. Activities with periodic occurrences, clear intervals, and phased release of disturbances are labeled as intermittent disturbance activities, such as phased road excavation. Activities with concentrated disturbances, short durations, but high instantaneous intensity are labeled as concentrated burst activities, such as blasting operations carried out in a short period. These tagged activities use a unified set of rules to ensure consistent tagged classification for similar activities across different regions. Simultaneously with the tagged activities, basic geological environmental data for the target area is extracted. The basic geological environmental data includes information on soil and rock types, slope structural units, and fault distances. Soil and rock types are classified and labeled based on regional geological data. Slope structural units are divided according to slope aspect, slope gradient, and structural combination characteristics. Fault distance information is obtained by calculating the spatial distance between engineering activity points or disaster points and known faults. All of the above basic geological environmental data are compiled using a unified data standard and maintain a spatial reference system consistent with that of engineering activity points and geological disaster points.

[0065] Based on the extracted geological environment data, the target area is divided into geological environment units. Using soil and rock types and slope structures as the primary criteria, spatial areas with similar or highly similar soil and rock properties and slope structure characteristics are grouped into the same geological environment unit. During the division process, the continuity and integrity of the geological environment unit boundaries are strictly maintained to avoid fragmented or overlapping units, ensuring that each geological environment unit has a clear and stable spatial coverage. After the geological environment unit division is completed, historical engineering activity locations and historical geological hazard locations, which have been marked with engineering activity types, are assigned to their corresponding geological environment units according to their spatial location. In practice, spatial matching is used to determine whether a location is within the boundary of a geological environment unit, and a unique correspondence between the location and the geological environment unit is established accordingly. For locations near the unit boundary, they are uniformly assigned to the nearest geological environment unit based on their spatial center location. Subsequently, the engineering activity location data, geological hazard location data, and geological environment data are integrated at the geological environment unit level to form a historical dataset indexed by geological environment units.

[0066] In S2, the spatial kernel density values ​​of various historical engineering activity sites within different geological environment units are calculated according to the type of engineering activity.

[0067] For historical engineering activity location data with completed activity type labeling, three categories of engineering activities were extracted: continuous disturbance, intermittent disturbance, and concentrated outbreak. Using the activity type labeling results as the sole criterion, all historical engineering activity locations within the target area belonging to the same activity type were uniformly grouped into the same dataset, ensuring that each dataset contains only locations of a single type of engineering activity. This classification and extraction process maintains the precise location of the original locations at the spatial coordinate level, without reducing or merging the number or distribution range of locations, thus fully preserving the true spatial distribution characteristics of various engineering activities within the target area.

[0068] Spatial density calculations are performed for each type of engineering activity location. Using the spatial location of each activity location as the center, a kernel density estimation method is used to continuously represent the spatial concentration of engineering activities. During kernel density estimation, a smoothing kernel function is uniformly used to extend the spatial influence of each activity location, ensuring that each activity location has a continuous impact on its surrounding space. The spatial resolution of the kernel density estimation is uniformly set according to the scale of the target area; for example, a fixed grid side length is used to regularly divide the target area, ensuring that the generated spatial density results are comparable across different types of engineering activities. Through the above processing, spatial density raster layers covering the entire target area are generated for persistent disturbance, intermittent disturbance, and concentrated burst engineering activities. These spatial density raster layers reflect the spatial distribution intensity of the corresponding engineering activity type within the target area in continuous numerical form.

[0069] After generating spatial density raster layers for different engineering activity types, the spatial density raster layers corresponding to each engineering activity type are spatially overlaid with vector layers of geological environment units based on soil and rock types and slope structures. In practice, the spatial density raster layers and geological environment unit layers use the same spatial coordinate reference. The coverage area of ​​each geological environment unit in the spatial density raster layer is determined through spatial overlay. This overlay process ensures that each geological environment unit corresponds to a clearly defined set of spatial density raster pixels, and that there is no pixel overlap or duplicate assignment between different geological environment units. After spatial overlay, for each geological environment unit, the spatial density raster of persistent disturbance, intermittent disturbance, and concentrated burst-type engineering activities is statistically calculated. By traversing all raster pixels falling within the spatial range of the geological environment unit, the spatial density value of the corresponding engineering activity type is read, and the average value of all pixel values ​​within the unit is calculated. The resulting average value is used as the spatial kernel density value of that engineering activity type within the geological environment unit.

[0070] In S3, the spatial kernel density values ​​of various engineering activities within the geological environment unit are correlated with the frequency of historical geological disasters to generate a geological sensitivity coefficient matrix corresponding to various engineering activities.

[0071] Using the already defined geological environmental units as a unified spatial analysis scale, the occurrence of historical geological hazards was statistically processed. First, the historical geological hazard locations assigned to each geological environmental unit were summarized and statistically analyzed to determine the total number of geological hazards occurring in that unit during the complete historical observation period. To eliminate the influence of spatial differences between different geological environmental units on the statistical results, the coverage area of ​​the geological environmental unit was used as a unified benchmark to normalize the number of hazard occurrences. This was achieved by calculating the ratio of the number of hazard occurrences to the corresponding spatial area of ​​the geological environmental unit, making the hazard occurrences of geological environmental units at different area scales comparable. Through this processing, the historical hazard occurrence frequency values ​​corresponding to each geological environmental unit were obtained and arranged according to the spatial index order of the geological environmental units, forming a historical hazard occurrence frequency sequence. Simultaneously, for all identified engineering activity types, the corresponding spatial kernel density values ​​within each geological environmental unit were extracted and organized. The spatial kernel density values ​​of persistent disturbance type, intermittent disturbance type, and concentrated outbreak type engineering activities were read separately within each geological environmental unit and arranged according to the geological environmental unit index order consistent with the historical hazard occurrence frequency sequence. Using the above methods, kernel density sequences of engineering activities corresponding to the frequency sequences of historical disasters are generated.

[0072] After obtaining the historical disaster occurrence frequency sequence and the kernel density sequence corresponding to each type of engineering activity, a quantitative analysis of the spatial correlation between the two is conducted. Using the geological environment unit as the smallest control unit, the disaster occurrence frequency values ​​within the same geological environment unit are paired one-to-one with the kernel density values ​​of the corresponding engineering activity type, forming a set of data pairs reflecting the relationship between the spatial distribution of engineering activities and the occurrence of geological disasters. Based on this, correlation analysis is performed on the data pairs. By calculating the Pearson correlation coefficient between the disaster occurrence frequency sequence and the engineering activity kernel density sequence, the overall spatial correlation between the intensity of the spatial distribution of engineering activities and the degree of geological disaster occurrence is quantified. The correlation coefficient is used to characterize the overall impact trend of a certain type of engineering activity on the occurrence of geological disasters at the target area scale; its magnitude reflects the strength of the positive or negative correlation between the spatial distribution of this type of engineering activity and the occurrence of disasters. The above correlation analysis is performed separately for each type of engineering activity, thereby obtaining the spatial correlation strength indicators corresponding to persistent disturbance type, intermittent disturbance type, and concentrated outbreak type engineering activities. By establishing a clear correspondence between the types of engineering activities and their corresponding spatial correlation strength, the disaster impact characteristics of different types of engineering activities at the overall scale can be quantitatively expressed, laying the foundation for the subsequent differential calculation of sensitivity coefficients.

[0073] The sensitivity of geological environmental units under different engineering activities is further refined. First, for each type of engineering activity, the average spatial kernel density of that type of activity is calculated across all geological environmental units. This average reflects the overall distribution level of that type of engineering activity within the target area. Then, for each geological environmental unit, the deviation between its corresponding engineering activity kernel density value and the average kernel density value of that type of engineering activity is compared. The degree of deviation determines whether the geological environmental unit is in a disturbance state above or below the regional average under the influence of that type of engineering activity. The obtained overall spatial correlation strength of the engineering activity type and the kernel density deviation of the geological environmental units are weighted and comprehensively calculated using custom weights to obtain the sensitivity coefficient of each geological environmental unit under the conditions of that type of engineering activity. This sensitivity coefficient characterizes the difference in response of different geological environmental units to disasters under the same type of engineering activity. The above sensitivity coefficient calculation process is performed for each type of engineering activity and each geological environmental unit individually. Finally, all calculated sensitivity coefficients are summarized and organized according to the correspondence between engineering activity types and geological environmental units to construct a geological sensitivity coefficient matrix corresponding to different types of engineering activities. This sensitivity coefficient matrix records the disaster sensitivity characteristics under the combined effects of differences in engineering activity types and differences in geological environment units in a structured form.

[0074] In step S4, based on the geological sensitivity coefficient matrix of various engineering activities, the intensity of engineering activity locations is spatially weighted to generate a quantitative data sequence of engineering activity impact intensity.

[0075] For historical engineering activity sites that have completed spatial registration and type labeling, the intensity of each engineering activity site is obtained. The intensity of the engineering activity is determined based on the actual implementation of the activity, specifically using objective indicators that reflect the degree of disturbance to the geological body, such as blasting intensity and vibration intensity. For continuous disturbance-type engineering activities, the intensity is characterized by the duration of continuous operation, for example, the number of consecutive construction days. For intermittent disturbance-type engineering activities, the intensity is characterized by the cumulative operation time of each operation phase. For concentrated burst-type engineering activities, the intensity is characterized by the scale or number of operations within the concentrated operation phase. Subsequently, for each historical engineering activity site, its corresponding geological environment unit is determined based on its spatial location, and the corresponding sensitivity coefficient is selected from the constructed geological sensitivity coefficient matrix, taking into account the type of engineering activity at that site. In the specific implementation process, the corresponding engineering activity type index is determined by the type marking information of the engineering activity point, and the row and column position of the point in the sensitivity coefficient matrix is ​​located by the geological environment unit identifier, thereby uniquely determining the geological sensitivity coefficient value corresponding to the engineering activity point.

[0076] After matching the intensity of engineering activities with the geological sensitivity coefficient, the impact intensity correction calculation is performed for each historical engineering activity location. Specifically, the intensity of the engineering activity at each location is multiplied by its corresponding geological sensitivity coefficient. This multiplication yields a dimensionless impact intensity value for the engineering activity, corrected for geological sensitivity. This impact intensity value comprehensively reflects both the intensity of the disturbance caused by the engineering activity itself and the response of the local geological environment to the disturbance, resulting in differentiated impacts of the same engineering activity intensity in different geological environmental units. After calculating the impact intensity of a single engineering activity location, all historical engineering activity locations within each geological environmental unit are traversed. During this traversal, the impact intensity values ​​of engineering activities belonging to the same geological environmental unit and of the same type are recorded sequentially according to the chronological order of the engineering activities, forming an impact intensity data sequence reflecting the change of engineering activity disturbances over time within that geological environmental unit. For different types of engineering activities, corresponding historical engineering activity impact intensity data sequences are constructed, ensuring that each geological environmental unit corresponds to multiple impact intensity time sequences distinguishing different types of engineering activities.

[0077] In step S5, the time accumulation and trend analysis of the impact intensity data sequence of engineering activities on various geological environment units are performed, and the trigger feature indicators of the cumulative degree of engineering activity disturbance, growth rate and fluctuation amplitude are extracted to construct the input feature training set.

[0078] Taking a single geological environment unit as the smallest analytical object, the corresponding engineering activity impact intensity data sequence is subjected to time-structured processing. In the specific implementation, firstly, based on the time identifier corresponding to each data point in the engineering activity impact intensity data sequence, the data sequence is divided into windows of fixed time length. The fixed time length is uniformly set according to the actual continuous characteristics of the disturbance caused by engineering activities to the geological environment; for example, a monthly or quarterly time window is used as an example, ensuring that each time window can completely cover a period of engineering activity with continuous disturbance significance. The time windows are arranged continuously in chronological order to ensure the integrity and continuity of the engineering activity impact intensity data in the time dimension, without skipping or merging intermediate time periods. After completing the time window division, statistical calculations are performed on the engineering activity impact intensity data falling within each time window. Firstly, all engineering activity impact intensity values ​​within the time window are accumulated, and the accumulated result is used as the cumulative disturbance amount corresponding to that time window, representing the overall disturbance level exerted by engineering activities on the geological environment during that time period. Subsequently, the changes in the intensity values ​​of engineering activities within the same time window were analyzed. By calculating the variation amplitude between adjacent intensity values ​​and taking their average, the growth rate index corresponding to this time window was obtained to reflect the increasing trend of engineering activity disturbances over time. Simultaneously, the dispersion of the intensity values ​​of all engineering activities within this time window was statistically analyzed, and the standard deviation of the intensity values ​​was used as a fluctuation amplitude index to characterize the stability of engineering activity disturbances during this period.

[0079] After calculating the cumulative disturbance, growth rate, and fluctuation amplitude indices for each time window of a single geological environment unit, the results are structured and organized to construct an input feature training set for machine learning training. For each geological environment unit, a complete data row is constructed according to unified data organization rules, based on a set of triggering feature indices formed within each time window. This data row contains at least the following: the engineering activity type identifier corresponding to the current time window, the unique identifier of the geological environment unit, and the calculated cumulative disturbance, growth rate, and fluctuation amplitude indices. The engineering activity type identifier is directly derived from the type labeling results of engineering activity points and is used to distinguish the impact characteristics of different disturbance mechanisms on the geological environment. The geological environment unit identifier is used to clarify the spatial unit range corresponding to this data row, ensuring that spatial attributes are not confused during model training. After filling in the triggering feature indices, the basic geological environment data parameters corresponding to this geological environment unit are synchronously introduced into the data row. These basic geological environment data parameters include the classification identifier corresponding to the soil and rock type, slope structure unit characteristics, and fault distance information, all filled in with the same expression as before to avoid inconsistent parameter scales or ambiguous meanings. The aforementioned basic geological environmental data and triggering characteristic indicators are combined within the same data row, enabling this data row to simultaneously reflect the disturbance state of engineering activities and the carrying capacity characteristics of the geological environment. By repeatedly executing the above data row construction process on all geological environmental units and their corresponding time windows within the target area, an input feature training set consisting of multiple data rows is ultimately formed.

[0080] In step S6, the input feature training set is labeled with geological disaster triggering status labels and then input into the machine learning-based evaluation model for training.

[0081] For the constructed input feature training set, the geological environment unit and time period corresponding to each row of data are analyzed one by one. The geological environment unit identifier information and the start and end range of the corresponding time period for each row of data in the input feature training set are read to clarify the spatial range and time interval described by the sample. Subsequently, based on the geological environment unit identifier, the spatial boundary range of the geological environment unit within the target area is determined, and historical geological disaster location records within this spatial range are extracted under a unified spatial reference system. After completing the spatial range limitation, the occurrence time information in the historical geological disaster location records is time-matched. For the end time period corresponding to each row of sample data, a prediction time window of a set length is extended forward, and historical geological disaster events occurring within this prediction time window are filtered. The length of the prediction time window is set according to a unified rule, such as using a fixed number of years as the prediction time span, to ensure the consistency of label judgment standards between different samples. Through the above dual spatial and temporal constraints, the set of historical geological disaster events corresponding to the current sample is accurately selected.

[0082] After completing the spatial and temporal screening of historical geological disaster events, a trigger status label determination operation is performed on each row of data in the input feature training set. For each row of samples, corresponding to the geological environment unit and the prediction time window, it is determined whether there is at least one historical geological disaster location record falling within the spatial range of the geological environment unit within the prediction time window. If there is at least one historical geological disaster location record that meets both the spatial range and time window conditions, the geological environment unit corresponding to the sample is determined to have entered a geological disaster trigger state after that time period, and the sample is labeled with a positive trigger state. If no historical geological disaster location record that meets the conditions is found within the corresponding prediction time window, the geological environment unit corresponding to the sample is determined not to have entered a trigger state, and the sample is labeled with a non-trigger state. The above trigger status labels are recorded using a binary method to clearly distinguish whether a geological disaster trigger event has occurred in the geological environment unit corresponding to the sample within a specific time period. The entire labeling process is performed on each row of data in the input feature training set to ensure that each row of sample data has a unique and clear trigger status label.

[0083] The machine learning evaluation model adopts a classification model structure. Its input layer receives the feature information contained in each row of data in the aforementioned input feature training set. The feature information includes engineering activity type identifier, geological environment unit identifier, cumulative disturbance amount, growth rate, fluctuation amplitude, and corresponding basic geological environment parameters. The model's output layer is used to output the trigger discrimination result of the corresponding input sample, which is used to characterize the discrimination tendency of the sample to trigger a geological disaster event within the prediction time window. The input feature training set labeled with trigger status is divided into a training subset and a validation subset. The training subset is used for iterative updates of model parameters, and the validation subset is used for monitoring model performance and judging stability during training. During training, the input feature data in the training subset is fed into the model input layer in batches. The input features are processed by the model's internal discrimination structure, and the corresponding discrimination output is generated in the output layer. Subsequently, the model output result is compared with the corresponding trigger status label. The model training error is calculated based on the magnitude of the deviation between the two, and the update operation of the model's internal parameters is triggered accordingly. The above training process is continuously executed according to the preset training rounds until the model's discrimination result on the validation subset reaches a stable state, specifically:

[0084] The model's internal parameters include the weights of the input features, the discriminant coefficients used to determine the relationships between feature combinations, and the bias parameters used to adjust the model's output bias. During each training round, after the input feature training data passes through the model, the model generates a corresponding discriminant output based on the current parameter configuration. This output is then compared with the actual trigger state labels to determine the direction and magnitude of the discrimination error under the current model parameter configuration. During the parameter update phase, based on the feedback results of the discrimination error, the model's internal weights are gradually adjusted. Features that contribute significantly to the trigger state receive higher weights in subsequent training, while features that contribute less or in the opposite direction have their weights reduced accordingly. Simultaneously, the model's internal bias parameters are updated to correct the overall discrimination threshold, enabling the model to form clearer discrimination boundaries between different categories of samples. This parameter update process is executed immediately after each batch of training samples completes its discrimination, ensuring that the model parameters gradually converge during training. After all training rounds are completed, the model's internal parameter configuration is fixed, forming a trained machine learning evaluation model. When the evaluation model receives new input feature data, it generates a corresponding discriminant output value through its internal discriminant structure. The discriminant output value serves as the geological hazard triggering potential score.

[0085] In S7, based on the newly added engineering activities, the evaluation model outputs the geological disaster triggering potential score of each geological environment unit, classifies the disaster susceptibility risk level, and outputs the risk zoning results.

[0086] For newly added engineering activities within the target area, a data processing workflow consistent with historical engineering activities is employed. First, spatial location information, activity type identifiers, and intensity descriptions of the new engineering activities are acquired. Then, based on the spatial location of the new engineering activity points, they are assigned to pre-defined geological environmental units, ensuring spatial consistency with the existing geological environmental unit system. After point assignment, for each new engineering activity, a corresponding geological sensitivity coefficient is selected according to its activity type, and the activity intensity is corrected to generate an impact intensity value. Subsequently, the impact intensity value is organized according to a predetermined time window rule, extracting characteristic indicators such as cumulative disturbance, growth rate, and fluctuation amplitude consistent with the historical training phase. Simultaneously, the geological environmental unit identifier, activity type identifier, and corresponding geological environmental parameters such as soil and rock type and slope structure unit are combined to form structurally complete and dimensionally consistent input feature data. The above input feature data strictly follows the feature composition order and data format of the training phase to ensure semantic consistency when inputting into the evaluation model. Finally, the newly constructed engineering activity input features are input one by one into the trained evaluation model, the model inference process is executed, and the corresponding trigger discrimination score output is obtained.

[0087] After obtaining the trigger discrimination score output by the evaluation model, the trigger discrimination score is subjected to uniform scaling processing to convert it into a geological hazard triggering potential score. The trigger discrimination score originates from the output of the discrimination function within the evaluation model. This output value characterizes the relative tendency of new engineering activities to trigger geological hazard events under the current combination of geological environmental conditions and disturbance characteristics. To facilitate subsequent risk classification management, the trigger discrimination score is normalized, mapping the discrimination scores corresponding to different geological environmental units to a unified numerical range, which serves as the geological hazard triggering potential score.

[0088] After calculating the trigger potential score, the geological hazard trigger potential score is classified into levels according to pre-defined grading interval rules. The grading intervals are flexibly set according to engineering management needs; for example, the trigger potential score interval can be divided into low trigger potential zones, medium trigger potential zones, and high trigger potential zones, mapping the trigger potential score corresponding to each geological environmental unit to a clear risk level identifier. Finally, the risk level results of each geological environmental unit are integrated with its spatial range to generate the geological hazard susceptibility risk zoning results for the target area. This risk zoning result uses the geological environmental unit as the basic expression unit, intuitively reflecting the distribution of geological hazard susceptibility risks in different areas under the conditions of new engineering activities, providing clear spatial guidance for engineering activity control and risk prevention.

[0089] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0090] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0091] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0092] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0093] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0095] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0097] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing geological hazard susceptibility based on a machine learning model, characterized in that, Includes the following steps: S1. Obtain historical datasets for the target area, including historical engineering activity locations and types, historical geological disaster locations, and basic geological environment data; S2. Calculate the spatial kernel density values ​​of various historical engineering activity sites in different geological environment units according to the type of engineering activity; S3. Conduct a correlation analysis between the spatial kernel density values ​​of various engineering activities within the geological environment unit and the frequency of historical geological disasters to generate a geological sensitivity coefficient matrix corresponding to various engineering activities. S4. Based on the geological sensitivity coefficient matrix of various engineering activities, the intensity of engineering activity points is spatially weighted to generate a quantitative data sequence of engineering activity impact intensity. S5. Perform time accumulation and trend analysis on the data sequence of engineering activity impact intensity of various geological environment units, extract the trigger feature indicators of the degree of engineering activity disturbance accumulation, growth rate and fluctuation amplitude, and construct the input feature training set; S6. Label the input feature training set with geological disaster triggering status labels and input them into the machine learning-based evaluation model for training; S7. The evaluation model outputs the geological hazard triggering potential score of each geological environment unit, classifies the hazard susceptibility risk level, and outputs the risk zoning results.

2. The geological hazard susceptibility assessment method based on a machine learning model according to claim 1, characterized in that, In step S1, the historical dataset of the target area is obtained, including historical engineering activity locations and types, historical geological disaster locations, and basic geological environment data, specifically including: Collect data on historical engineering activity sites and historical geological disaster sites within the target area; Based on the time and intensity characteristics of the disturbance to the geological body caused by engineering activities, the engineering activity type of each historical engineering activity site is marked, and the types include continuous disturbance type, intermittent disturbance type, and concentrated burst type. Extract basic geological environmental data of the target area, including information on rock and soil types, slope structural units, and fault distances. Based on the rock and soil types and slope structures in the geological environment data of the target area, the target area is divided into several geological environment units; The historical engineering activity sites, historical geological disaster sites, and basic geological environment data that have been tagged with type are registered and associated based on geological environment units, and integrated into a historical dataset.

3. The geological hazard susceptibility assessment method based on a machine learning model according to claim 2, characterized in that, In S2, the calculation of spatial kernel density values ​​for various historical engineering activity locations within different geological environment units, based on the type of engineering activity, specifically includes: For three types of engineering activities—persistent disturbance, intermittent disturbance, and concentrated outbreak—all historical engineering activity locations for each type of engineering activity within the target area are extracted. A spatial density raster layer covering the target area is generated using kernel density estimation. Spatial overlay is performed between the spatial density raster layer corresponding to each type of engineering activity and the vector layer of geological environment units divided based on soil and rock type and slope structure; For each geological environment unit, the average pixel value of the spatial density raster for each type of engineering activity within the unit is calculated, and the calculation result is used as the spatial kernel density value of that type of engineering activity within the geological environment unit.

4. The geological hazard susceptibility assessment method based on a machine learning model according to claim 1, characterized in that, In step S3, the spatial kernel density values ​​of various engineering activities within a geological environment unit are correlated with the frequency of historical geological disasters to generate a geological sensitivity coefficient matrix corresponding to various engineering activities. This specifically includes: Using geological environmental units as the basic unit of analysis and the coverage area of ​​geological environmental units as the benchmark, the historical disaster occurrence frequency of each geological environmental unit is calculated and integrated into a historical disaster occurrence frequency sequence. For all types of engineering activities, the spatial kernel density values ​​of all geological environment units are extracted and integrated into a kernel density sequence; The correlation coefficient between the historical disaster occurrence frequency sequence and the kernel density sequence is calculated as the overall spatial correlation strength between each type of engineering activity and geological disaster; Based on the obtained correlation coefficients and the degree of deviation of the kernel density values ​​of each geological environment unit from the average kernel density value of the kernel density sequence, the sensitivity coefficient of the geological environment unit under each type of engineering activity is calculated. The sensitivity coefficients of all geological environment units are collected to construct a geological sensitivity coefficient matrix for different types of engineering activities.

5. The geological hazard susceptibility assessment method based on a machine learning model according to claim 1, characterized in that, In step S4, based on the geological sensitivity coefficient matrix of various engineering activities, the intensity of engineering activity sites is spatially weighted to generate a quantified data sequence of engineering activity impact intensity, specifically including: Obtain the intensity of engineering activities corresponding to each historical engineering activity point; Based on the geological environment unit to which the engineering activity site belongs, the corresponding geological sensitivity coefficient matrix is ​​selected in combination with the corresponding engineering activity type; Multiply the intensity of engineering activities at each engineering activity site by the corresponding sensitivity coefficient to obtain the historical engineering activity impact intensity value after geological environment sensitivity correction; The impact intensity values ​​of all historical engineering activities within each geological environment unit are calculated to obtain a data sequence of the impact intensity of historical engineering activities of different types.

6. The geological hazard susceptibility assessment method based on a machine learning model according to claim 1, characterized in that, In step S5, the time accumulation and trend analysis of the impact intensity data sequence of engineering activities on various geological environment units are performed to extract trigger feature indicators of the cumulative degree, growth rate and fluctuation amplitude of engineering activity disturbances, and the construction of the input feature training set specifically includes: The data sequence of engineering activity impact intensity corresponding to a single geological environment unit is divided into continuous time periods according to a fixed time window; Within each time period, the sum of the impact intensity values ​​of engineering activities falling within that time period is calculated as the cumulative disturbance, the average rate of change is calculated as the growth rate, and the standard deviation of the intensity values ​​is calculated as the fluctuation amplitude. The calculated engineering activity type, geological environment unit identifier, cumulative disturbance, growth rate, and fluctuation amplitude of the geological environment unit in the current time period are combined with the corresponding parameter values ​​of the geological environment basic data of the geological environment unit into a data row. All data rows together constitute the input feature training set.

7. The geological hazard susceptibility assessment method based on a machine learning model according to claim 1, characterized in that, In step S6, labeling the input feature training set with geological disaster triggering state tags and inputting them into the machine learning-based evaluation model for training specifically includes: Based on the geological environment unit and time period corresponding to each row of data in the input feature training set, query the spatial location and occurrence time records of historical geological disaster sites; Determine the historical geological disaster location records within the spatial range of the geological environment unit within a prediction time window of a set length after the end of the time period, and label the input feature training set with binarized trigger state labels; The labeled input feature training set and the corresponding trigger state labels are input into the selected machine learning algorithm for model training evaluation.

8. The geological hazard susceptibility assessment method based on a machine learning model according to claim 1, characterized in that, In S7, based on the newly added engineering activities, the evaluation model outputs geological hazard triggering potential scores for each geological environment unit, classifies hazard susceptibility risk levels, and outputs risk zoning results, specifically including: For newly added engineering activities within the target area, input features are generated based on the type and location of the engineering activities, combined with the intensity of the engineering activities, and then input into the evaluation model. The trigger discrimination score output by the assessment model is converted into the geological hazard triggering potential score of each geological environment unit. Based on the preset classification interval, the risk level is divided to obtain the geological hazard susceptibility risk zoning results.

Images