A traffic engineering regional landslide hazard prediction method fusing intelligent algorithm and rainfall threshold model

Abstract

The application provides a traffic engineering regional landslide hazard prediction method fusing intelligent algorithm and rainfall threshold model, relates to the traffic engineering safety technical field, and first collects geological environment interpretation data sets and historical landslide event record sets of a target traffic engineering region, then calls a geographic information system platform to perform spatial overlay analysis on the geological environment interpretation data sets, divides environment factor grid cell arrays and extracts geological environment element parameter sets. Then, a rainfall-induced landslide event case knowledge base is constructed, historical landslide events, environment factor grid cells and rainfall process record data are associated, and an associated case data entry set is formed. Then, a landslide event sample set is constructed, a preset intelligent algorithm model is inputted for training, a spatial susceptibility probability prediction model is outputted, a spatial susceptibility probability distribution map covering the whole range of the target region is generated, and the accuracy of landslide hazard prediction is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of traffic engineering safety technology, and more specifically, to a method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models. Background Technology

[0002] Landslides are a significant threat to the safety of transportation projects and the stability of surrounding areas during construction and operation. The geological conditions in transportation engineering areas are complex and diverse, and significantly influenced by meteorological factors such as rainfall. Accurate prediction of landslide risks in these areas is crucial for ensuring the safe operation of transportation projects and minimizing economic losses and casualties.

[0003] Currently, methods for predicting landslide hazard in transportation engineering areas have certain limitations. Traditional prediction methods often focus on the analysis of single factors, such as considering only geological conditions or rainfall, failing to comprehensively integrate the impact of geological environmental factors and rainfall triggering factors on landslide occurrence. While some methods attempt to combine multiple factors, they lack scientific rigor and systematicity in data processing and model construction, making it difficult to accurately reflect the complex mapping relationship between environmental factors and landslide probability. Furthermore, existing methods are not precise enough in spatial analysis, unable to generate spatial susceptibility probability distribution maps covering the entire transportation engineering area, thus failing to meet the needs of accurate landslide hazard prediction and comprehensive assessment in practical engineering. Summary of the Invention

[0004] In view of the aforementioned problems, this invention provides a method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models, the method comprising:

[0005] Collect a set of geological environment interpretation data of the target transportation engineering area and a set of historical landslide event records of the target transportation engineering area. The set of historical landslide event records includes the geographical coordinates of the occurrence of each historical landslide event, the time of occurrence of the event, and the rainfall process record data corresponding to the time of occurrence of the event.

[0006] The geographic information system platform of the target traffic engineering area is invoked to perform spatial overlay analysis on the geological environment interpretation data set, dividing the geological environment interpretation data set into an array of environmental factor raster cells with spatial continuity characteristics, and extracting the set of geological environment element parameters corresponding to each environmental factor raster cell;

[0007] A knowledge base of cases of rainfall-induced landslide events is constructed. Each historical landslide event in the historical landslide event record set is associated with the corresponding environmental factor raster unit according to the geographical coordinate information of the event occurrence, and associated with the corresponding rainfall process record data according to the time information of the event occurrence, forming a set of associated case data entries containing environmental background parameters of landslide events and rainfall triggering parameters of landslide events.

[0008] A landslide event sample set construction operation is performed on the associated case data entry set. Environmental factor raster cells belonging to the location of landslide events in the associated case data entry set are marked as positive sample cells, and environmental factor raster cells that have not experienced landslide events in the associated case data entry set are marked as negative sample cells, thereby generating a landslide event sample set containing feature data of positive sample cells and feature data of negative sample cells.

[0009] The landslide event sample set is input into a preset intelligent algorithm model for model training, and a spatial susceptibility probability prediction model that can characterize the mapping relationship between environmental factor parameters and landslide occurrence probability is output. The spatial susceptibility probability prediction model is then used to calculate the spatial susceptibility probability value of each environmental factor grid cell in the environmental factor grid cell array, generating a spatial susceptibility probability distribution map covering the entire range of the target traffic engineering area.

[0010] Based on the above, by collecting geological environment interpretation data sets and historical landslide event records of the target transportation engineering area, and calling the geographic information system platform to perform spatial overlay analysis on the geological environment interpretation data sets, the data sets are divided into an array of environmental factor raster units with spatial continuity characteristics. Corresponding sets of geological environment element parameters are extracted, and a knowledge base of rainfall-induced landslide event cases is constructed. Historical landslide events are associated with environmental factor raster units and rainfall process records, forming a set of associated case data entries containing environmental background parameters of landslide events and rainfall triggering parameters. By constructing a landslide event sample set and inputting it into a preset intelligent algorithm model for training, a spatial susceptibility probability prediction model that accurately represents the mapping relationship between environmental factor parameters and landslide occurrence probability is output. This spatial susceptibility probability prediction model is used to generate a spatial susceptibility probability distribution map covering the entire target transportation engineering area, which can intuitively and accurately reflect the landslide hazard in different areas, effectively improving the accuracy of landslide hazard prediction. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the execution flow of the landslide hazard prediction method for traffic engineering areas that integrates intelligent algorithms and rainfall threshold models, provided in an embodiment of the present invention.

[0012] Figure 2This is a schematic diagram of exemplary hardware and software components of a landslide hazard prediction system for traffic engineering areas that integrates intelligent algorithms and rainfall threshold models, provided in an embodiment of the present invention. Detailed Implementation

[0013] Figure 1 This is a flowchart illustrating a landslide hazard prediction method for traffic engineering areas that integrates intelligent algorithms and rainfall threshold models, provided in one embodiment of the present invention. A detailed description follows.

[0014] Step S110: Collect a set of geological environment interpretation data of the target transportation engineering area and a set of historical landslide event records of the target transportation engineering area. The set of historical landslide event records includes the geographical coordinates of each historical landslide event, the time of the event, and the rainfall process records corresponding to the time of the event.

[0015] In this embodiment, the target transportation engineering area is selected as a strip-shaped area extending 20 kilometers on each side of a high-speed railway trunk line in a mountainous area. First, a multi-source, multi-temporal remote sensing image dataset covering the target transportation engineering area is acquired using a satellite remote sensing image interpretation platform. This platform integrates optical imagery and synthetic aperture radar (SAR) imagery. Optical imagery is used to identify surface cover and geological structure features, while SAR imagery is used to acquire surface deformation information. An automatic extraction operation of geological structural linear bodies is performed on the acquired remote sensing image dataset. This operation, based on image edge detection algorithms and Hough transform, identifies and vectorizes linearly distributed geological structural traces in the images, ultimately extracting the fault structure linear body distribution layer and fold structure boundary layer within the target transportation engineering area.

[0016] Next, a digital elevation model (DEM) dataset with a spatial resolution of 12.5 meters was acquired using a DEM data acquisition platform for the target transportation engineering area. Topographic factor derivation calculations were performed on this DEM dataset. Utilizing the spatial analysis module of the geographic information system platform, based on the elevation values ​​of each raster cell and its neighboring cells in the DEM dataset, a difference algorithm was used to calculate the maximum slope change rate, generating a topographic slope factor raster layer. Aspect factor raster layer was generated by calculating the azimuth angle of the slope orientation. Topographic curvature factor raster layer was generated by calculating the slope change rate. Finally, topographic relief factor raster layer was generated by calculating the elevation difference between the highest and lowest points within a preset neighborhood window.

[0017] Then, a geological lithology distribution map at a scale of 1:100,000 was obtained from the geological map digitization platform for the target transportation engineering area. This geological lithology distribution map detailed the rock stratigraphic units of different ages and origins. A lithological unit attribute extraction operation was performed on this geological lithology distribution map, vectorizing each rock stratigraphic unit. Based on a preset lithology classification coding table, each vector polygon was assigned rock type attribute parameters, rock structure attribute parameters, and rock weathering degree attribute parameters, forming a geological lithology distribution raster layer containing the above attribute information.

[0018] Next, a set of hydrogeological condition data for the target transportation engineering area is obtained through a hydrogeological survey database. This set of data includes borehole water level observation data, spring flow data, and hydrogeological profiles. Spatialization of hydrogeological parameters is then performed on this set of data. Based on the principles of groundwater dynamics and combined with digital elevation model data, groundwater flow field simulation is conducted to generate a groundwater burial depth raster layer and a groundwater runoff modulus raster layer.

[0019] The latest land use type data set for the target transportation engineering area is obtained from the land use status survey database. This land use type data set includes types such as cultivated land, forest land, and construction land. A land use type classification and coding operation is performed on this data set. Based on the preset land use classification and coding system, the vectorized land use patches are converted into land use type code values, generating a land use type raster layer.

[0020] Subsequently, the generated fault structure linear distribution layer, fold structure boundary layer, topographic slope factor raster layer, topographic aspect factor raster layer, topographic curvature factor raster layer, topographic relief factor raster layer, geological lithology distribution raster layer, groundwater burial depth raster layer, groundwater runoff modulus raster layer, and land use type raster layer are uniformly projected onto the same coordinate system, and spatial registration and raster overlay operations are performed to ensure that all layers have the same spatial extent and raster resolution. This overlay operation uses the raster algebra operation function of the geographic information system platform to fuse the information of multiple layers into a multi-band raster dataset. Each spatial location point in this multi-band raster dataset corresponds to a vector containing multi-dimensional geological environmental element parameters; this is the geological environmental interpretation dataset.

[0021] Simultaneously, a list of historical landslide events in the target transportation engineering area was obtained from a geological disaster historical record database. Each record in this list includes the latitude and longitude coordinates of the landslide location (determined by on-site investigation or remote sensing interpretation), the date (year, month, day, hour, and minute) of the landslide, daily rainfall data for the 15 consecutive days prior to the landslide, daily rainfall data for the day of the landslide, and daily rainfall data for the 7 consecutive days following the landslide. The rainfall data was obtained from meteorological station monitoring records within the area. Data cleaning was performed on this historical landslide event record list. First, the spatial query function of the geographic information system platform was used to remove invalid records whose latitude and longitude coordinates of the landslide location exceeded the boundary of the target transportation engineering area. Second, the daily rainfall data was checked for completeness, removing records with missing data for any day. Finally, the remaining valid records after cleaning were aggregated to generate a cleaned historical landslide event record set.

[0022] Step S120: Call the geographic information system platform of the target traffic engineering area to perform spatial overlay analysis on the geological environment interpretation data set, divide the geological environment interpretation data set into an array of environmental factor raster cells with spatial continuity characteristics, and extract the set of geological environment element parameters corresponding to each environmental factor raster cell.

[0023] Step S121: Receive the fault structure linear body distribution layer from the geological environment interpretation data set through the rasterization processing module of the geographic information system platform, and convert the fault structure linear body distribution layer into a fault structure linear body distance raster layer with uniform spatial resolution. The value of each raster cell in the fault structure linear body distance raster layer is the Euclidean distance value of the center point of the raster cell from the nearest fault structure linear body.

[0024] In this embodiment, the geographic information system (GIS) platform loads a fault structure linear volume distribution layer containing multiple fault line vectors. After receiving this vector layer, the GIS platform's rasterization module first creates an empty raster frame covering the entire target traffic engineering area, with a spatial resolution of 30 meters and M rows and N columns. Next, for each raster cell in this empty raster frame, the shortest Euclidean distance from its center point coordinates to all fault structure linear volume line segments in the vector layer is calculated. This calculation process involves traversing all fault line segments, calculating the vertical distance from the center point to each segment or the distance to the endpoint, and taking the minimum value. This minimum value is used as the pixel value of the raster cell. After all raster cells have completed this calculation, a fault structure linear volume distance raster layer is generated, where the value of each raster cell represents the potential spatial distance affected by the fault structure at that location.

[0025] Step S122: Receive the fold structure boundary layer from the geological environment interpretation dataset through the rasterization processing module of the geographic information system platform, and convert the fold structure boundary layer into a fold structure distance raster layer with uniform spatial resolution. The value of each raster cell in the fold structure distance raster layer is the Euclidean distance from the center point of the raster cell to the nearest fold structure boundary.

[0026] Similar to step S121, the rasterization module of the geographic information system platform receives the vector map layer of the fold structure boundary, creates an empty raster frame with the same spatial extent and resolution as the fault layer mentioned above, and performs Euclidean distance calculation. For each raster cell, the shortest distance from its center point to all fold structure boundary polygons is calculated, and this distance value is assigned to the corresponding raster cell. Finally, a fold structure distance raster layer is generated, where the value of each raster cell reflects the spatial proximity of that location to the fold structure boundary.

[0027] Step S123: The rasterization processing module of the geographic information system platform receives the topographic slope factor raster layer, topographic aspect factor raster layer, topographic curvature factor raster layer, and topographic relief factor raster layer from the geological environment interpretation data set. The spatial resolution of the topographic slope factor raster layer, the topographic aspect factor raster layer, the topographic curvature factor raster layer, and the topographic relief factor raster layer is unified to generate a slope factor raster layer, the aspect factor raster layer, the curvature factor raster layer, and the relief factor raster layer with unified spatial resolution.

[0028] In this embodiment, the original spatial resolution of the aforementioned generated terrain slope factor raster layer, terrain aspect factor raster layer, terrain curvature factor raster layer, and terrain relief factor raster layer is 12.5 meters. To maintain a consistent spatial resolution with the constructed layers, the geographic information system platform invokes the raster resampling module and employs a bilinear interpolation algorithm to resample the four layers. For each target 30-meter resolution raster cell, the bilinear interpolation algorithm calculates a new value by taking the weighted average of the values ​​of the four nearest neighbor raster cells covered by that cell in the original 12.5-meter resolution raster. After resampling, four terrain slope factor raster layers, terrain aspect factor raster layers, terrain curvature factor raster layers, and terrain relief factor raster layers with a uniform 30-meter spatial resolution are generated.

[0029] Step S124: Receive the geological lithology distribution raster layer from the geological environment interpretation data set through the rasterization processing module of the geographic information system platform, perform a lithology type coding reclassification operation on the geological lithology distribution raster layer, and convert the rock type attribute parameters, rock structure attribute parameters, and rock weathering degree attribute parameters of each raster unit in the geological lithology distribution raster layer into lithology type coding values ​​to generate a lithology type coding raster layer.

[0030] The rasterization module of the Geographic Information System (GIS) platform receives a geological lithology distribution raster layer. Each raster cell in this layer is associated with three attribute parameters: rock type, rock structure, and rock weathering degree. To simplify the multidimensional attributes into a single code that is easy for the model to process, a pre-defined lithology comprehensive coding mapping table is loaded using a reclassification tool. This mapping table uniquely maps a specific combination of rock type, structure, and weathering degree to an integer code value. For example, granite, massive structure, and moderate weathering correspond to code 101; sandstone, layered structure, and weak weathering correspond to code 202. The reclassification tool traverses each raster cell in the geological lithology distribution raster layer, reads its three attribute values, looks up the corresponding code in the mapping table, and writes the code to the corresponding position in the newly generated raster layer. After the traversal is complete, a lithology type coded raster layer is generated.

[0031] Step S125: Receive the groundwater burial depth raster layer, groundwater runoff modulus raster layer, and land use type raster layer from the geological environment interpretation data set through the rasterization processing module of the geographic information system platform. Perform spatial resolution unification processing on the groundwater burial depth raster layer, the groundwater runoff modulus raster layer, and the land use type raster layer to generate a groundwater burial depth raster layer, runoff modulus raster layer, and land use type raster layer with unified spatial resolution.

[0032] The original spatial resolution of the groundwater depth raster layer and the groundwater runoff modulus raster layer may be 90 meters, while the original resolution of the land use type raster layer may be 10 meters. The geographic information system platform calls the raster resampling module to resample these three layers with a target resolution of 30 meters. For the continuous groundwater depth and runoff modulus layers, bilinear interpolation is used; for the discrete land use type coding layer, the nearest neighbor method is used to ensure that the coded values ​​are not invalidated due to interpolation. After processing, three raster layers with a uniform spatial resolution of 30 meters are generated: a groundwater depth raster layer, a runoff modulus raster layer, and a land use type raster layer.

[0033] Step S126: Perform spatial coordinate system unification processing on the fault structure linear body distance raster layer, the fold structure morphology distance raster layer, the slope factor raster layer, the aspect factor raster layer, the curvature factor raster layer, the undulation factor raster layer, the lithology type coding raster layer, the groundwater depth raster layer, the runoff modulus raster layer, and the land use type raster layer to generate a multi-layer overlay raster dataset with the same spatial range, the same spatial resolution, and the same projected coordinate system.

[0034] The Geographic Information System (GIS) platform loads all ten raster layers processed through the above steps into a single view. First, the platform performs a coordinate system unification check. If differences in the projected coordinate systems of the layers are found, a projection transformation tool is used to convert all layers to a preset unified projected coordinate system. Next, the platform performs spatial extent clipping, using the smallest bounding rectangle of all layers as a reference to clip all layers to the exact same geographic extent. Since the previous steps have unified the spatial resolution of all layers to 30 meters, the ten layers now have identical row count M and column count N. The platform combines these ten layers into a multi-band raster dataset, where each band corresponds to a geological environmental factor. The values ​​of each spatial location (i.e., the raster cell in the i-th row and j-th column) on different bands collectively constitute the multi-dimensional geological environmental element parameters for that location. This multi-band raster dataset is the multi-layer overlay raster dataset.

[0035] Step S127: Based on the raster row and column number configuration parameters of the multi-layer overlay raster dataset, divide the multi-layer overlay raster dataset into an environmental factor raster cell array. Each environmental factor raster cell in the environmental factor raster cell array corresponds to a unique set of raster row and column index numbers.

[0036] The Geographic Information System (GIS) platform reads the metadata of a multi-layer overlay raster dataset to obtain its row count parameter M and column count parameter N. Based on these two parameters, the platform logically constructs an M-row, N-column array of environmental factor raster cells. Each element in the array, i.e., an environmental factor raster cell, is uniquely identified by its row index variable i (ranging from 1 to M) and column index variable j (ranging from 1 to N) within the array. This index number serves as the spatial location identifier for that raster cell.

[0037] Step S128: Traverse each environmental factor raster cell in the environmental factor raster cell array, and extract the fault distance attribute value in the fault structure linear body distance raster layer, the fold distance attribute value in the fold structure morphology distance raster layer, the slope attribute value in the slope factor raster layer, the slope aspect attribute value in the aspect factor raster layer, the curvature attribute value in the curvature factor raster layer, the relief attribute value in the relief factor raster layer, the lithology code attribute value in the lithology type code raster layer, the groundwater depth attribute value in the groundwater depth raster layer, the runoff modulus attribute value in the runoff modulus raster layer, and the land use type code attribute value in the land use type raster layer, to generate the geological environmental element parameter set corresponding to the environmental factor raster cell.

[0038] The Geographic Information System (GIS) platform sequentially obtains the row index i and column index j of each environmental factor raster cell. For the current raster cell, it accesses the multi-layer overlay raster dataset and reads the values ​​of the environmental factor raster cell in ten bands: fault distance attribute value from the fault structure linear body distance raster layer band; fold distance attribute value from the fold structure morphology distance raster layer band; slope attribute value from the slope factor raster layer band; slope aspect attribute value from the slope aspect raster layer band; curvature attribute value from the curvature factor raster layer band; relief attribute value from the relief factor raster layer band; lithology code attribute value from the lithology type code raster layer band; groundwater depth attribute value from the groundwater depth raster layer band; runoff modulus attribute value from the runoff modulus raster layer band; and land use type code attribute value from the land use type raster layer band. These ten attribute values ​​are organized into a set of geological environmental element parameters corresponding to the environmental factor raster cell, and the set of geological environmental element parameters is associated with and stored with the row index i and column index j of the raster cell.

[0039] Step S130: Construct a knowledge base of rainfall-induced landslide event cases, associate each historical landslide event in the historical landslide event record set with the corresponding environmental factor raster unit according to the geographical coordinate information of the event occurrence, and associate it with the corresponding rainfall process record data according to the event occurrence time information, forming a set of associated case data entries containing environmental background parameters of landslide events and rainfall triggering parameters of landslide events.

[0040] Step S131: Extract the geographic coordinates of each historical landslide event from the historical landslide event record set, input the geographic coordinates of the event into the spatial query module of the geographic information system platform, calculate the spatial inclusion relationship between the geographic coordinates of the event and each environmental factor raster cell in the environmental factor raster cell array through the spatial query module, and determine the target environmental factor raster cell containing the geographic coordinates of the event.

[0041] The spatial query module of the Geographic Information System (GIS) platform receives the geographic coordinates of a historical landslide event, including longitude (X) and latitude (Y). This coordinate point is then spatially overlaid with the entire environmental factor raster array. Since each raster cell in the environmental factor raster array corresponds to a defined geographic extent (determined by the coordinates of its top-left corner, spatial resolution, and row and column indices), the spatial query module quickly locates the unique raster cell containing the coordinate point (X, Y) by calculating the geometric inclusion relationship between the coordinate point (X, Y) and the geographic extent of each raster cell. This raster cell is the target environmental factor raster cell associated with this historical landslide event, and its corresponding raster row and column index number (i, j) is obtained.

[0042] Step S132: Extract the event occurrence time information of each historical landslide event from the historical landslide event record set. Based on the event occurrence time information, extract the daily rainfall sequence data within a consecutive preset number of days before the landslide event as the pre-rainfall process data from the rainfall process record data of the historical landslide event. Extract the daily rainfall data on the day the landslide event occurred as the triggering rainfall process data. Extract the daily rainfall sequence data within a consecutive preset number of days after the landslide event as the post-rainfall process data.

[0043] Extract the event occurrence time information from records of the same historical landslide event, including the specific year, month, day, and hour. Based on this date, extract the daily rainfall data for the 15 consecutive days prior to the landslide from the associated rainfall records, arranging them chronologically to form the pre-landslide rainfall data sequence. Extract the daily rainfall data for the day the landslide occurred as the trigger rainfall data. Extract the daily rainfall data for the seven consecutive days following the landslide, arranging them chronologically to form the post-landslide rainfall data sequence.

[0044] Step S133: Use the set of geological environmental element parameters corresponding to the target environmental factor raster unit as the environmental background parameter set of the historical landslide event, and combine the previous rainfall process data, the triggered rainfall process data and the subsequent rainfall process data into the rainfall trigger parameter set of the historical landslide event.

[0045] The set of geological environmental element parameters corresponding to the target environmental factor raster cells determined in step S131 is directly used as the environmental background parameter set for this historical landslide event. At the same time, the data sequences of the preceding rainfall process, the triggered rainfall process, and the subsequent rainfall process extracted in step S132 are encapsulated into a rainfall trigger parameter set containing three subsets.

[0046] Step S134: Link and store the geographical coordinates of each historical landslide event, the time of the event, the set of environmental background parameters, and the set of rainfall triggering parameters to generate a record of rainfall-induced landslide event case data containing landslide event identifier field, spatial location field, time field, environmental factor field, and rainfall process field.

[0047] The data management module of the Geographic Information System (GIS) platform generates a unique identifier for this historical landslide event, serving as the landslide event identification field. Subsequently, the event's geographic coordinates, time of occurrence, environmental background parameters, and rainfall triggering parameters are stored in the spatial location field, time field, environmental factor field, and rainfall process field, respectively. These five fields together constitute a complete record of a rainfall-induced landslide event case.

[0048] Step S135: Perform a spatial location consistency check on the rainfall-induced landslide event case data records, and check whether the environmental factor raster cell corresponding to the spatial location field has the same raster row and column index number as the spatial source raster cell of each geological environmental element parameter in the environmental background parameter set. Mark the rainfall-induced landslide event case data records that pass the spatial location consistency check as valid case data records.

[0049] The data management module performs validation on each generated case data record. It parses the raster row and column index numbers of the target environmental factor raster cell from the spatial location field. From the environmental background parameter set, it arbitrarily selects a parameter (e.g., slope attribute value), traces back its source layer and corresponding raster cell, and obtains the index number of that raster cell. It compares whether these two index numbers are completely identical. If they are identical, it indicates that the spatial association of the record is accurate, and the record is marked as a valid case data record; if they are inconsistent, it is marked as an invalid record and discarded.

[0050] Step S136: Collect all valid case data records to form a rainfall-induced landslide event case knowledge base. Each valid case data record in the rainfall-induced landslide event case knowledge base contains the corresponding relationship between the environmental background parameters of the landslide event and the rainfall triggering parameters of the landslide event.

[0051] All valid case data records that pass the verification in step S135 are collected to form a database set, namely the rainfall-induced landslide event case knowledge base. Each record in this rainfall-induced landslide event case knowledge base defines a complete correspondence between a specific historical landslide event and its geological environmental background (environmental background parameter set) and rainfall-inducing conditions (rainfall triggering parameter set).

[0052] Step S137: Traverse each valid case data record in the rainfall-induced landslide event case knowledge base, extract the geological environmental element parameters of the environmental background parameter set in the valid case data record to form an environmental background parameter vector, extract the previous rainfall process data sequence value and the triggered rainfall process data value of the rainfall trigger parameter set in the valid case data record to form a rainfall trigger parameter vector, and concatenate the environmental background parameter vector and the rainfall trigger parameter vector to generate the associated case data entry of the valid case data record. All associated case data entries corresponding to the valid case data records constitute an associated case data entry set.

[0053] The data processing module iterates through each valid case data record in the rainfall-induced landslide event case knowledge base. For the current record, it first extracts the following attribute values ​​from its environmental factor field, in the order of ten preset geological environmental elements: fault distance, fold distance, slope, aspect, curvature, relief, lithology coding, groundwater depth, runoff modulus, and land use type coding. These are then arranged into a 10-dimensional environmental background parameter vector. Next, it extracts 15 daily rainfall values ​​from the previous rainfall process data sequence and the triggering rainfall process data value (one value) from its rainfall process field, arranging them into a 16-dimensional rainfall triggering parameter vector. Finally, the environmental background parameter vector and the rainfall triggering parameter vector are concatenated dimensionally to form a 26-dimensional comprehensive feature vector, which serves as the associated case data entry for this record. This process is repeated for all records, and all generated associated case data entries together constitute the associated case data entry set.

[0054] Step S140: Perform a landslide event sample set construction operation on the associated case data entry set, mark the environmental factor raster cells in the associated case data entry set that belong to the location of the landslide event as positive sample cells, and mark the environmental factor raster cells in the associated case data entry set that have not experienced a landslide event as negative sample cells, thereby generating a landslide event sample set containing the feature data of positive sample cells and the feature data of negative sample cells.

[0055] Step S141: Extract the raster row and column index number of the environmental factor raster cell corresponding to each associated case data entry in the associated case data entry set, and summarize the raster row and column index numbers corresponding to all associated case data entries to form a landslide event occurrence raster index number set.

[0056] The data processing module traverses the set of associated case data entries, parsing the raster row and column index number of the corresponding environmental factor raster cell from each entry (i.e., the index of the landslide location obtained by tracing back from the association process of that entry). These index numbers are collected into a set, and duplicates are removed, ultimately forming a set of landslide event occurrence raster index numbers, which records the spatial locations of all historical landslide events.

[0057] Step S142: Extract the raster row and column index numbers of all environmental factor raster cells in the environmental factor raster cell array to form a complete set of raster index numbers for the entire region.

[0058] From the metadata of the environmental factor raster cell array, the number of rows M and columns N are obtained. All possible combinations of row indices i (from 1 to M) and column indices j (from 1 to N) are generated through nested loops to form a complete set of raster index numbers for the entire region. This complete set of raster index numbers contains the spatial identifiers of all locations within the target traffic engineering area.

[0059] Step S143: Remove the raster row and column index numbers belonging to the set of raster index numbers where landslide events occurred from the set of raster index numbers in the whole area through set difference operation to obtain the set of raster index numbers where no landslide events occurred.

[0060] The data processing module performs a set difference operation on the complete set of raster index numbers for the entire region and the set of raster index numbers for landslide events. That is, it removes all elements belonging to the occurrence set from the complete set, and the remaining elements are the index numbers of the raster cells for which no landslide events have occurred, thus forming the set of raster index numbers for which no landslide events have occurred.

[0061] Step S144: Traverse each raster row and column index number in the set of raster index numbers for landslide events, locate the environmental factor raster cell corresponding to the raster row and column index number, extract the values ​​of each geological environmental element parameter corresponding to the environmental factor raster cell from the set of geological environmental element parameters to form a positive sample cell feature vector, associate and store the positive sample cell feature vector with the preset positive sample category label to generate positive sample cell feature data.

[0062] For each index number (i, j) in the set of raster index numbers for landslide events, the corresponding environmental factor raster cell is located. From the set of geological environmental element parameters of this cell, fault distance, fold distance, slope, aspect, curvature, relief, lithology coding, groundwater depth, runoff modulus, and land use type coding attributes are extracted and arranged in a preset order to form a positive sample cell feature vector with a dimension of ten. Simultaneously, a preset category label is associated with this feature vector, for example, the integer value 1, representing "landslide occurred". The pairing of this feature vector and label constitutes a positive sample cell feature data.

[0063] Step S145: Traverse each raster row and column index number in the set of raster index numbers where no landslide event has occurred, locate the environmental factor raster cell corresponding to the raster row and column index number, extract the geological environmental element parameter values ​​corresponding to the environmental factor raster cell from the set of geological environmental element parameters to form a negative sample cell feature vector, associate and store the negative sample cell feature vector with the preset negative sample category label to generate negative sample cell feature data.

[0064] For each index number (i, j) in the set of raster index numbers where no landslide event occurred, perform the same operation as in step S144 to extract its 10 geological environmental element parameter values, forming a negative sample cell feature vector with a dimension of ten. Associate it with a preset category label, such as the integer value 0, representing "no landslide occurred". The pairing of this feature vector with the label constitutes a negative sample cell feature data.

[0065] Step S146: Configure parameters according to the preset ratio between the number of positive sample units and the number of negative sample units, randomly extract a specified number of positive sample unit feature vectors and corresponding positive sample category labels from the positive sample unit feature data to form a positive sample subset, and randomly extract a specified number of negative sample unit feature vectors and corresponding negative sample category labels from the negative sample unit feature data to form a negative sample subset.

[0066] Since the locations where landslides occurred (positive samples) are typically far fewer than the locations where they did not occur (negative samples), a pre-defined positive-to-negative sample ratio, such as 1:1, is used to construct a balanced dataset for model training. Based on this ratio, the data processing module determines the required number of positive and negative samples, T0. From the positive sample unit feature data, T records are randomly selected without replacement using a random number generator to form a positive sample subset. Similarly, T records are randomly selected from the negative sample unit feature data to form a negative sample subset.

[0067] Step S147: Merge the positive sample subset and the negative sample subset to form an initial landslide event sample set.

[0068] The positive and negative sample subsets generated in step S146 are merged to form an initial landslide event sample set containing T sample records.

[0069] Step S148: Perform feature standardization processing on the sample feature vectors of all sample records in the initial landslide event sample set, so that the feature values ​​corresponding to each geological environmental element parameter conform to the preset value distribution range, and generate a standardized landslide event sample set. Each sample record in the standardized landslide event sample set contains a standardized sample feature vector and a corresponding sample category label.

[0070] The data processing module standardizes each dimension (i.e., each geological environmental element) of the feature vectors of all samples in the initial landslide event sample set. For a specific element dimension (e.g., slope), the module calculates the arithmetic mean μ and standard deviation σ of all samples in that dimension. Then, for each sample record, the original value X of that sample in that dimension is calculated by dividing (X - μ) by σ to obtain the new standardized value. This operation transforms the original data into a distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of different dimensions and orders of magnitude. After performing this operation on all ten dimensions, the feature vector of each sample record is updated to the standardized feature vector. Finally, all sample records and their corresponding category labels constitute the standardized landslide event sample set.

[0071] Step S150: Input the landslide event sample set into a preset intelligent algorithm model for model training, output a spatial susceptibility probability prediction model that can characterize the mapping relationship between environmental factor parameters and landslide occurrence probability, and use the spatial susceptibility probability prediction model to calculate the spatial susceptibility probability value of each environmental factor grid cell in the environmental factor grid cell array, generating a spatial susceptibility probability distribution map covering the entire range of the target traffic engineering area.

[0072] Step S151: Obtain the initial model parameter configuration information of the preset intelligent algorithm model. The preset intelligent algorithm model includes input feature dimension parameters, hidden layer node number parameters, hidden layer number parameters, output layer node number parameters, and activation function type parameters.

[0073] In this embodiment, the preset intelligent algorithm model uses a multilayer perceptron neural network architecture. Its initial model parameter configuration information is as follows: the input feature dimension parameter is set to ten, consistent with the number of geological environmental element parameters. The number of hidden layer nodes is set to 64 nodes for the first hidden layer and 32 nodes for the second hidden layer. The number of hidden layer layers is set to two. The number of output layer nodes is set to one, used to output the probability value of landslide occurrence. The activation function type parameter is set as follows: the hidden layers use the modified linear unit activation function, and the output layer uses the logistic activation function.

[0074] Step S152: Divide the landslide event sample set into a model training sample subset and a model validation sample subset. The model training sample subset is used to optimize the model parameters of the preset intelligent algorithm model, and the model validation sample subset is used to evaluate the generalization performance of the preset intelligent algorithm model.

[0075] The data processing module randomly sorts all samples in the standardized landslide event sample set and, according to a preset division ratio, such as seven to three, designates the first 70% of the samples as the model training sample subset for subsequent model parameter optimization; and designates the last 30% of the samples as the model validation sample subset. This model validation sample subset does not participate in parameter updates during training and is only used to evaluate the generalization performance of the model, i.e., to determine whether the model is overfitting.

[0076] Step S153: Input the sample feature vector of each sample record in the training sample subset of the model into the input layer of the preset intelligent algorithm model, perform nonlinear transformation processing on the sample feature vector through the hidden layer of the preset intelligent algorithm model, extract the high-order combined feature representation of the sample feature vector, perform probability mapping processing on the high-order combined feature representation through the output layer of the preset intelligent algorithm model, and output the landslide occurrence probability prediction value corresponding to the sample record.

[0077] Model training begins. First, a sample record is taken from the training sample subset, and its standardized 10-dimensional feature vector is used as input to the input layer of the multilayer perceptron model. The 10 nodes of the input layer each receive one of the 10 dimensions of the feature vector. Next, the data propagates forward to the first hidden layer. This hidden layer contains 64 nodes, each connected to all ten nodes of the input layer, with a trainable weight parameter on each connection. For each node in the first hidden layer, the model calculates the weighted sum of all dimensions of the input feature vector, adds the node's bias parameter, and then inputs the result into the rectified linear unit activation function. This activation function sets negative values ​​to zero, leaves positive values ​​unchanged, and outputs the node's activation value. The activation values ​​of all 64 nodes in the first hidden layer form a 64-dimensional vector. This vector is then input to the second hidden layer. The second hidden layer contains 32 nodes, each connected to all 64 nodes of the first hidden layer. Similarly, the model calculates the sum of weights and biases, and performs a nonlinear transformation through a modified linear unit activation function, outputting a 32-dimensional vector. This vector represents the higher-order combined feature representation of the sample feature vectors. Finally, this 32-dimensional vector is input to the output layer. The output layer has only one node, which is connected to all 32 nodes in the second hidden layer. The model calculates the sum of weights and biases of the input at this node and inputs it to the logistic activation function. This activation function compresses the input values ​​to the interval (0, 1), and the output value is the predicted probability of landslide occurrence for that sample record.

[0078] Step S154: Calculate the loss function value between the predicted landslide occurrence probability value and the sample category label of the sample record. Based on the loss function value, calculate the gradient information of each model parameter in the preset intelligent algorithm model through the backpropagation algorithm. Update the model parameters of the preset intelligent algorithm model based on the gradient information.

[0079] The model training module obtains the predicted landslide probability output from step S153, along with the true class label (0 or 1) of the sample record. A binary cross-entropy loss function is used to calculate the difference between the predicted value and the true label, yielding the loss function value. This loss function value quantifies the quality of the model's current prediction. Next, the backpropagation algorithm is initiated, starting from the output layer. Using the chain rule from calculus, the partial derivatives of the loss function value with respect to each trainable parameter in the model (i.e., the connection weights and biases between nodes in each layer) are calculated in reverse. These partial derivatives constitute the gradient information. The gradient of each parameter indicates how to adjust that parameter in the current direction to most quickly reduce the loss function value. The model optimizer (e.g., an adaptive moment estimator) updates all parameters according to the calculated gradient information and a preset learning rate, adjusting the parameter values ​​along the direction of gradient descent.

[0080] Step S155: Repeat the operation of inputting the sample feature vector of the sample record in the sample subset of the model training sample into the preset intelligent algorithm model and updating the model parameters until the loss function value converges to the preset convergence threshold range, stop the model parameter update operation, and use the preset intelligent algorithm model in the current model parameter state as the spatial susceptibility probability prediction model.

[0081] The model training module repeatedly executes steps S153 and S154, iteratively updating the model parameters using each sample record in the training sample subset. Each complete traversal of all training samples constitutes a training epoch. After each epoch, the module calculates the average loss function value across the entire training sample subset. When the average loss function value decreases by less than a preset convergence threshold (e.g., one-thousandth) over multiple consecutive training epochs, or when the preset maximum number of training epochs is reached, the model is considered converged, and training stops. At this point, all parameter values ​​stored in the current model (i.e., the trained weights and biases) are fixed; this model with fixed parameters is the spatial susceptibility probability prediction model.

[0082] Step S156: Input the sample feature vector of each sample record in the model validation sample subset into the spatial susceptibility probability prediction model, obtain the landslide occurrence probability prediction value of the validation sample output by the spatial susceptibility probability prediction model, calculate the validation loss function value between the landslide occurrence probability prediction value of the validation sample and the sample category label of the sample record in the model validation sample subset, and when the validation loss function value exceeds the preset validation loss threshold, readjust the initial model parameter configuration information of the preset intelligent algorithm model and retrain the model.

[0083] During or after model training, a subset of validation samples is used to evaluate the model's generalization ability. The feature vector of each sample in this subset is input into the trained spatial susceptibility probability prediction model to obtain its output landslide probability prediction value. The loss function value between these prediction values ​​and the true label is calculated to obtain the validation loss function value. If this value is significantly higher than the loss function value on the model training sample subset, or higher than the preset validation loss threshold, it indicates that the model has overfitting or underfitting problems. In this case, it is necessary to readjust the initial model parameter configuration information in step S151, such as adjusting the number of hidden layers, the number of nodes, the learning rate, or introducing a regularization term, and then re-execute the entire training process from steps S152 to S155 until a model with a validation loss function value that meets the requirements is obtained.

[0084] Step S157: Obtain the set of geological environmental element parameters corresponding to each environmental factor raster cell in the environmental factor raster cell array, and arrange and combine the values ​​of each geological environmental element parameter in the set of geological environmental element parameters according to the feature order of the sample feature vector to generate the prediction feature vector of the environmental factor raster cell.

[0085] After model training is complete, spatial prediction is performed. The data processing module iterates through all raster cells in the environmental factor raster cell array. For each cell, its set of geological environmental element parameters is obtained. Following the exact same order as during model training (i.e., fault distance, fold distance, slope, aspect, curvature, relief, lithology code, groundwater depth, runoff modulus, and land use type code), the above attribute values ​​are arranged into a ten-dimensional prediction feature vector.

[0086] Step S158: Input the predicted feature vector of each environmental factor grid cell into the spatial susceptibility probability prediction model, obtain the landslide occurrence probability prediction value of the environmental factor grid cell output by the spatial susceptibility probability prediction model, and use the landslide occurrence probability prediction value as the spatial susceptibility probability value of the environmental factor grid cell.

[0087] For each environmental factor grid cell, its corresponding predicted feature vector is input into the pre-trained spatial susceptibility probability prediction model. The model performs forward propagation calculations according to its trained weights and biases (the process is the same as step S153), ultimately outputting a value between 0 and 1, which is the predicted probability of a landslide occurring in that grid cell. This predicted probability value is recorded as the spatial susceptibility probability value for that cell.

[0088] Step S159: Generate a spatial susceptibility probability distribution map based on the raster row and column index number of each environmental factor raster cell and its corresponding spatial susceptibility probability value. The pixel value of each raster position in the spatial susceptibility probability distribution map corresponds to the spatial susceptibility probability value of the environmental factor raster cell in which the raster position is located.

[0089] The Geographic Information System (GIS) platform receives the raster row and column indices of all environmental factor raster cells and their corresponding spatial susceptibility probability values. The platform creates a new raster layer with the same number of rows M and columns N as the multi-layer overlaid raster dataset. For each index number (i, j), the spatial susceptibility probability value at that location is assigned as the pixel value of the new layer. After all locations are assigned values, a single-band raster layer covering the entire target traffic engineering area is generated; this single-band raster layer is the spatial susceptibility probability distribution map.

[0090] Step S210: Obtain the real-time rainfall monitoring data sequence from meteorological monitoring stations and the rainfall inversion data sequence from meteorological satellite cloud images of the target traffic engineering area. Perform spatiotemporal fusion processing on the real-time rainfall monitoring data sequence from meteorological monitoring stations and the rainfall inversion data sequence from meteorological satellite cloud images to generate an hourly rainfall raster layer sequence covering the entire target traffic engineering area.

[0091] Step S211: Obtain the real-time rainfall monitoring data sequence reported by meteorological monitoring stations within and around the target traffic engineering area within a preset distance range in a preset time interval through the meteorological monitoring data receiving interface. The real-time rainfall monitoring data sequence includes the geographical coordinates of each meteorological monitoring station, the monitoring time information of the station, and the rainfall value corresponding to the monitoring time of the station.

[0092] This step acquires high-precision point-like rainfall data. Through a meteorological monitoring data receiving interface, real-time or near-real-time rainfall monitoring data sequences reported by all automatic weather stations within the target transportation engineering area and a 50-kilometer radius around it are acquired from the past 24 hours. Each record in this rainfall monitoring data sequence includes the station's geographical coordinates (longitude, latitude), the station's monitoring time (year, month, day, hour), and the corresponding hourly rainfall value (in millimeters) at that monitoring time.

[0093] Step S212: Obtain the meteorological satellite cloud image rainfall inversion data sequence covering the target transportation engineering area through the meteorological satellite data receiving interface. The meteorological satellite cloud image rainfall inversion data sequence includes the raster geographic coordinate range information of each cloud image raster unit, cloud image acquisition time information, and rainfall inversion value corresponding to the cloud image acquisition time.

[0094] Simultaneously, infrared and visible light cloud image data from geostationary meteorological satellites (such as Fengyun-4) covering the target transportation engineering area are acquired through a meteorological satellite data receiving interface. These cloud image data are then processed using a pre-defined satellite rainfall retrieval algorithm (such as a rainfall intensity retrieval model based on cloud top brightness temperature) to generate a meteorological satellite cloud image rainfall retrieval data sequence. Each record in this meteorological satellite cloud image rainfall retrieval data sequence is a raster layer containing the raster geographic coordinate range information of each cloud image raster unit, the cloud image acquisition time information (the hour corresponding to the meteorological station monitoring time), and the rainfall value (in millimeters) obtained from satellite retrieval at that acquisition time.

[0095] Step S213: The real-time rainfall monitoring data sequence is grouped according to the station monitoring time information to generate a set of rainfall values ​​for each meteorological monitoring station corresponding to each monitoring time. Each set of rainfall values ​​for each meteorological monitoring station includes the station geographic coordinates and rainfall values ​​of each meteorological monitoring station at that monitoring time.

[0096] First, the real-time rainfall monitoring data sequence is processed, and the data is grouped according to the monitoring time information of the stations (i.e., hours). For each specific hour, such as 14:00 on July 15, 2023, the records reported by all meteorological monitoring stations within that hour are collected to form a set, which includes the geographical coordinates of each meteorological monitoring station at that time and its corresponding measured rainfall value.

[0097] Step S214: The meteorological satellite cloud image rainfall inversion data sequence is grouped according to the cloud image acquisition time information to generate a satellite cloud image rainfall inversion raster layer corresponding to each acquisition time. Each satellite cloud image rainfall inversion raster layer contains the raster geographic coordinate range information and rainfall inversion value of each cloud image raster unit at that acquisition time.

[0098] The module performs the same grouping process on the meteorological satellite cloud image rainfall inversion data sequence, grouping it according to the cloud image acquisition time information (hour). For the same hour, for example, 14:00 on July 15, 2023, the raster layer for that time is extracted to generate a satellite cloud image rainfall inversion raster layer. Each raster cell in this satellite cloud image rainfall inversion raster layer contains its geographic coordinate range and the rainfall value obtained from satellite inversion at that location.

[0099] Step S215: Perform spatiotemporal matching processing on the set of rainfall values ​​from meteorological monitoring stations corresponding to each monitoring time and the satellite cloud image rainfall inversion raster layer corresponding to that monitoring time. Extract the rainfall inversion values ​​of the corresponding cloud image raster unit for each meteorological monitoring station in the satellite cloud image rainfall inversion raster layer, and calculate the deviation between the rainfall monitoring values ​​and the rainfall inversion values ​​for each meteorological monitoring station.

[0100] For the same hour (e.g., 14:00 on July 15, 2023), the module performs spatiotemporal matching between the set of rainfall data from meteorological monitoring stations generated in step S213 and the satellite cloud image rainfall inversion raster layer generated in step S214. For each meteorological monitoring station in the set, based on its geographical coordinates, the module locates its corresponding cloud image raster cell in the satellite cloud image rainfall inversion raster layer and extracts the rainfall inversion value for that raster cell. Then, the difference between the measured rainfall value of the station and the extracted satellite inversion value is calculated; this difference is the rainfall deviation value for that station within that hour.

[0101] Step S216: Calculate the spatial distribution variation function of the deviation values ​​based on the deviation values ​​of all meteorological monitoring stations, and perform spatial interpolation processing on the spatial distribution variation function of the deviation values ​​using the Kriging interpolation method to generate a rainfall deviation correction raster layer covering the entire range of the target traffic engineering area.

[0102] The module collects the geographic coordinates of all meteorological monitoring stations and their corresponding rainfall deviations for the given hour, forming a discrete point set. Based on these points, the module calculates the spatial distribution variability function of the deviations, which describes how the deviations vary with spatial distance. Then, using ordinary kriging interpolation, with this variability function as a model, it performs spatial interpolation predictions for all locations without meteorological stations within the target transportation engineering area, generating a continuous surface covering the entire area with a spatial resolution of 30 meters—the rainfall deviation correction raster layer. Each raster cell value in this rainfall deviation correction raster layer represents the amount of deviation that needs to be corrected for the satellite-retrieved rainfall at that location within that hour.

[0103] Step S217: Overlay and correct the satellite cloud image rainfall inversion raster layer with the rainfall deviation correction raster layer. Add the rainfall inversion value of each cloud image raster unit in the satellite cloud image rainfall inversion raster layer to the deviation correction value of the corresponding raster unit in the rainfall deviation correction raster layer to generate the corrected hourly rainfall raster layer at the acquisition time.

[0104] The module adds the original satellite cloud image rainfall inversion raster layer to the newly generated rainfall deviation correction raster layer, cell by cell. For each raster cell, the satellite inversion value is added to the deviation correction value to obtain the final hourly rainfall value for that cell within that hour. This operation combines the accuracy of station data with the spatial continuity of satellite data to generate a corrected hourly rainfall raster layer.

[0105] Step S218: Arrange the corrected hourly rainfall raster layers corresponding to each collection time in chronological order of the collection times to generate an hourly rainfall raster layer sequence.

[0106] Steps S213 to S217 are repeated for each hour within the preset time interval (e.g., the past 24 hours) in step S210 to obtain the corrected hourly rainfall raster layer for each hour. Finally, the above layers are sorted according to the chronological order of the acquisition time to form a time series, namely the hourly rainfall raster layer sequence.

[0107] Step S220: Extract the continuous hourly rainfall value sequence of each grid cell in the hourly rainfall raster layer sequence within a preset time window, calculate the cumulative rainfall value of the grid cell within the preset time window based on the continuous hourly rainfall value sequence, and generate the cumulative rainfall time series data of the grid cell.

[0108] For each spatial raster cell in the hourly rainfall raster layer sequence, the data processing module extracts a continuous sequence of hourly rainfall values ​​within a preset time window along its time dimension. For example, this preset time window might be the 72 hours preceding the current moment. From the corresponding time period of the raster cell in the sequence, 72 rainfall values ​​are extracted chronologically. These values ​​are then summed to obtain the cumulative rainfall value for that raster cell within the 72-hour window. This process is repeated for each time point in the sequence, thereby generating a time-series data of cumulative rainfall for that raster cell.

[0109] Step S230: Obtain the preset rainfall infiltration coefficient configuration parameters, and calculate the previous effective cumulative rainfall value sequence of the grid cell within the preset time window by multiplying the cumulative rainfall value at each time point in the cumulative rainfall time series data with the rainfall infiltration coefficient configuration parameters.

[0110] A preset rainfall permeability coefficient configuration parameter is obtained. This parameter is a constant between 0.72 and 0.86, and its specific value is determined through statistical correlation analysis described later, representing the proportion of rainwater infiltrating into the soil and rock mass. For each time point in the cumulative rainfall time series data generated in step S220, the module multiplies the cumulative rainfall value at that time point by the permeability coefficient to obtain the previous effective cumulative rainfall value at that time point. This operation is performed on all time points to generate the previous effective cumulative rainfall value sequence for that raster cell within a preset time window.

[0111] Step S240: Extract the hourly rainfall value corresponding to the target time for each grid cell in the hourly rainfall raster layer sequence as the trigger rainfall value for that grid cell, and extract the previous effective cumulative rainfall value corresponding to the target time from the previous effective cumulative rainfall value sequence as the target previous effective rainfall value for that grid cell.

[0112] The module defines the current prediction time as the target time. For each raster cell, the value of the hourly rainfall raster layer corresponding to the target time is extracted from the hourly rainfall raster layer sequence and used as the trigger rainfall value for that raster cell. At the same time, the cumulative value corresponding to the target time is extracted from the previous effective cumulative rainfall value sequence of that raster cell generated in step S230 and used as the target previous effective rainfall value for that raster cell.

[0113] Step S250: Construct a rainfall duration parameter extraction function to determine the rainfall duration parameter value of each grid cell based on the time length parameter of the preset time window.

[0114] The rainfall duration parameter is defined as the duration of the rainfall process that triggers the landslide. In this embodiment, the rainfall duration parameter is associated with the length of the preset time window in step S220. A rainfall duration parameter extraction function is constructed, which directly uses the length of the preset time window (e.g., 72 hours) as the rainfall duration parameter value for each grid cell. Therefore, the rainfall duration parameter value for each grid cell is equal to the length of the preset time window.

[0115] Step S260: Input the target effective rainfall value and rainfall duration parameter value of each grid cell into a preset continuous probability rainfall threshold model. The preset continuous probability rainfall threshold model includes a logistic regression function expression with the rainfall duration parameter value as the independent variable and the target effective rainfall value as the independent variable. Calculate the landslide time probability value of the grid cell at the target time using the logistic regression function expression.

[0116] The module invokes a preset continuous probability rainfall threshold model. The core of this model is a binary logistic regression function, whose expression is: the probability of a landslide occurring at a given time equals one divided by (one plus an exponential function value of the natural constant, where the exponent of this exponential function is a0 plus a1 multiplied by the rainfall duration parameter value plus a2 multiplied by the target preceding effective rainfall value). Here, a0, a1, and a2 are model coefficients, obtained by fitting maximum likelihood estimates using the rainfall duration parameter values ​​and target preceding effective rainfall values ​​from a knowledge base of historical landslide events induced by rainfall. Substituting the target preceding effective rainfall value and rainfall duration parameter value for each grid cell into this logistic regression function, the calculated result is the probability of a landslide occurring in that grid cell under the current rainfall conditions.

[0117] Step S270: Multiply the landslide time occurrence probability value of each grid cell with the spatial susceptibility probability value of that grid cell in the spatial susceptibility probability distribution map to generate the comprehensive landslide hazard probability value of that grid cell.

[0118] For each grid cell, the module reads the spatial susceptibility probability value of that cell from the spatial susceptibility probability distribution map generated in step S158, and obtains the landslide occurrence probability value of that cell from step S260. Multiplying these two probability values ​​yields the comprehensive landslide hazard probability value for that grid cell. This product operation combines spatial susceptibility (intrinsic stability) with rainfall triggering (external dynamics), reflecting the comprehensive risk of a landslide occurring in that cell at the current moment.

[0119] Step S280: Generate a comprehensive probability distribution map of landslide risk covering the entire target traffic engineering area based on the grid row and column index number of each grid cell and its corresponding comprehensive probability value of landslide risk.

[0120] The Geographic Information System (GIS) platform receives the raster row and column index numbers of all raster cells and their corresponding comprehensive landslide hazard probability values. The platform creates a new raster layer, assigning the comprehensive landslide hazard probability value at each index number (i, j) as the pixel value. The resulting single-band raster layer is a comprehensive landslide hazard probability distribution map covering the entire target transportation engineering area.

[0121] Step S310: Obtain a vector map layer showing the distribution of traffic engineering facilities within the target traffic engineering area. The vector map layer includes the geographic coordinate boundary information and type attribute information of each traffic engineering facility.

[0122] In this embodiment, to assess the impact of a landslide on critical infrastructure, a vector map layer showing the distribution of transportation engineering facilities within the target transportation engineering area is obtained. This vector map layer includes all planned or existing transportation engineering facilities within the area, such as high-speed railway lines, stations, tunnel entrances, and bridge piers. Each facility element stores its precise geographic coordinate boundary information in the form of vector polygons or polylines, and is associated with facility type attribute information, such as "high-speed railway roadbed," "viaduct," and "tunnel."

[0123] Step S320: Perform a spatial overlay analysis on the traffic engineering facility distribution vector layer and the landslide hazard comprehensive probability distribution map to identify traffic engineering facilities that have spatial intersection with the grid cells in the landslide hazard comprehensive probability distribution map where the landslide hazard comprehensive probability value exceeds a preset probability threshold, and mark them as a set of high-risk traffic engineering facilities.

[0124] The geographic information system platform loads the traffic engineering facility distribution vector layer and the comprehensive probability distribution map of landslide hazard generated in step S280. The platform sets a preset probability threshold, for example, 0.3. The platform performs spatial overlay analysis, filtering out all raster cells in the comprehensive probability distribution map of landslide hazard with a probability value greater than or equal to 0.3, forming a set of high-risk raster cells. Then, the platform performs a spatial intersection query with the traffic engineering facility distribution vector layer to identify all traffic engineering facilities that spatially overlap (i.e., intersect or contain) with the high-risk raster cells. These identified facilities are marked, forming a set of high-risk traffic engineering facilities.

[0125] Step S330: Extract the geographic coordinate boundary information of each traffic engineering facility in the set of high-risk traffic engineering facilities, and calculate the statistical characteristic parameters of the comprehensive probability value of landslide risk of all grid cells within the grid cell range covered by the geographic coordinate boundary information of the traffic engineering facility. The statistical characteristic parameters include the maximum value parameter of comprehensive probability of landslide risk, the average value parameter of comprehensive probability of landslide risk, and the standard deviation parameter of comprehensive probability of landslide risk.

[0126] For each transportation engineering facility in the set of high-risk transportation engineering facilities, the platform extracts its geographic coordinate boundary information (a polygon). Using this polygon, the platform crops the comprehensive probability distribution map of landslide hazard, extracting all raster cells and their probability values ​​within the area covered by the facility. Then, the platform calculates the statistical characteristic parameters of these probability values: the maximum value parameter, which is the maximum probability value of all raster cells in the area; the average value parameter, which is the arithmetic mean of the probability values ​​of all raster cells in the area; and the standard deviation parameter, which is a measure of the dispersion of probability values ​​in the area.

[0127] Step S340: Based on the traffic engineering facility type attribute information of each traffic engineering facility, obtain the risk level classification threshold parameter of the traffic engineering facility from the preset facility type risk level mapping table, compare the maximum value parameter of the comprehensive probability of landslide hazard with the risk level classification threshold parameter, and determine the landslide risk level identifier of each traffic engineering facility.

[0128] A pre-defined facility type risk level mapping table is loaded. This table defines different risk level classification threshold parameters based on the importance of the facility. For example, for "high-speed railway subgrade," the high-risk threshold is set to 0.2; for "temporary construction access road," the high-risk threshold is set to 0.5. The platform queries this table and obtains the corresponding risk level classification threshold parameter based on the facility type attribute information of the current facility. Subsequently, the maximum value parameter of the facility calculated in step S330 is compared with this threshold. If the maximum value parameter is greater than or equal to the threshold, the landslide risk level of the facility is identified as "high risk"; otherwise, it may be identified as "medium risk" or "low risk."

[0129] Step S350: Generate a landslide risk warning list for each traffic engineering facility based on the landslide risk level identifier and the geographical coordinate boundary information of that traffic engineering facility. The landslide risk warning list for each traffic engineering facility in the high-risk traffic engineering facility set includes facility identifier information, facility geographical coordinate boundary information, landslide risk level identifier, and corresponding comprehensive probability statistical characteristic parameters of landslide hazard for each traffic engineering facility.

[0130] Information on each facility in the set of high-risk transportation engineering facilities is compiled to generate a structured early warning list. Each entry in the early warning list corresponds to a transportation engineering facility and includes its unique facility identification information, geographic coordinate boundary information for location, landslide risk level identification determined in step S340, and statistical characteristic parameters such as the maximum value parameter, average value parameter, and standard deviation parameter of the area where the facility is located, calculated in step S330.

[0131] Step S360: Push the landslide risk warning list of the transportation engineering facilities to the risk warning information receiving terminal of the transportation engineering management department.

[0132] Through a pre-defined data interface, the generated landslide risk warning list for transportation engineering facilities is pushed in a standard format (such as a vector file compatible with a geographic information system platform) to the risk warning information receiving terminal designated by the transportation engineering management department, so that the management department can take timely measures.

[0133] Step S410: Obtain a vector map layer showing the distribution of residential areas within the target transportation engineering area. The vector map layer shows the geographic coordinate boundary information of each residential area and the population size attribute information of each residential area.

[0134] Meanwhile, to assess the threat of landslides to residents' lives, a vector map layer of residential areas within the target transportation engineering area was obtained. This vector map layer includes all natural villages and township settlements within the area. Each settlement stores its precise geographic coordinate boundary information in the form of a vector polygon, and is associated with population attribute information obtained from census data or administrative statistics.

[0135] Step S420: Perform a spatial overlay analysis on the vector map of the residential area distribution and the comprehensive probability distribution map of landslide hazard to identify residential areas that have spatial intersection with the grid cells in the comprehensive probability distribution map of landslide hazard that exceed a preset probability threshold, and mark them as a set of high-risk residential areas.

[0136] Similar to step S320, the platform performs spatial overlay analysis on the vector map of residential area distribution and the comprehensive probability distribution map of landslide hazard. Using the same preset probability threshold, high-risk raster cells are selected, and all residential areas that spatially intersect with these high-risk raster cells are identified, forming a set of high-risk residential areas.

[0137] Step S430: Extract the geographic coordinate boundary information of each residential area in the set of high-risk residential areas, and calculate the weighted average of the comprehensive probability values ​​of landslide risk of all grid cells within the grid cell range covered by the geographic coordinate boundary information of the residential area. The weight coefficient of each grid cell is obtained by inverse distance weighting based on the Euclidean distance between the geographic coordinates of the grid cell and the geographic center coordinates of the residential area, thereby generating the weighted comprehensive probability value of landslide risk for the residential area.

[0138] For each residential area in the set of high-risk residential areas, the platform extracts its geographic coordinate boundary information. It then calculates the coordinates of the geometric center point of the residential area. Next, it extracts the probability values ​​of all raster cells within the area covered by the residential area. For each raster cell within the area, it calculates the Euclidean distance from its center point to the geometric center point of the residential area. Then, it calculates a weighting coefficient based on this distance, which is inversely proportional to the square of the distance; that is, raster cells that are closer to each other have higher weights. The weighted average of all raster cells is calculated as the weighted comprehensive probability value of landslide hazard for the residential area, which better reflects the potential danger level of the core area of ​​the residential area.

[0139] Step S440: Based on the population size attribute information of each residential area and the weighted comprehensive probability value of landslide risk for that residential area, calculate the estimated number of people exposed to landslide risk in each residential area. The estimated number of people exposed to landslide risk is the product of the population size attribute information of the residential area and the weighted comprehensive probability value of landslide risk.

[0140] For each high-risk residential area, the platform multiplies its population size attribute information with the weighted landslide hazard probability value calculated in step S430 to obtain a numerical value, which is the estimated number of people in the area who may be threatened by a landslide. For example, if a residential area has 1,000 people and the weighted probability is 0.1, then the estimated number of people at risk of exposure is 100.

[0141] Step S450: Link and store the geographic coordinate boundary information, weighted comprehensive probability value of landslide risk, and estimated number of people exposed to landslide risk for each residential area to generate a set of landslide risk assessment data for residential areas.

[0142] The geographical coordinate boundary information of all high-risk residential areas, the corresponding weighted comprehensive probability value of landslide hazard, and the estimated number of people exposed to risk are stored as a complete record to form a set of landslide risk assessment data for residential areas.

[0143] Step S460: Sort all residential areas in the high-risk residential area set according to the weighted comprehensive probability value of landslide hazard in the landslide risk assessment data set of the residential area, and generate a landslide risk ranking list of residential areas arranged in descending order of risk level.

[0144] A sorting algorithm was applied, using the "weighted comprehensive probability value of landslide hazard" from the landslide risk assessment dataset for residential areas as the key field, to rank all records in descending order. Residential areas with higher probability values ​​were ranked higher. This sorted list demonstrates the relative risk faced by each residential area.

[0145] Step S470: Push the landslide risk ranking list of the residential area to the emergency response command terminal of the emergency management department.

[0146] The generated landslide risk ranking list for residential areas will be pushed to the emergency response command terminal designated by the emergency management department.

[0147] For example, step S510: Obtain a multi-period time-series remote sensing image dataset of the target traffic engineering area, wherein the multi-period time-series remote sensing image dataset contains high-resolution remote sensing images of multiple time phases collected at fixed time intervals within a historical time interval.

[0148] To achieve dynamic monitoring of landslide events and self-enhancing of the model, this method also includes an iterative model optimization step. First, high-resolution optical or synthetic aperture radar remote sensing images of the target traffic engineering area, collected monthly over the past three years, are acquired at multiple temporal intervals to form a multi-period time-series remote sensing image dataset. This dataset contains image records of the land surface at different time points.

[0149] Step S520: Perform image change detection and analysis on the multi-period time-series remote sensing image dataset, calculate the surface deformation of each pixel between remote sensing images of different time phases; identify pixel areas with surface deformation greater than a preset deformation threshold as suspected slope deformation areas, form a set of suspected slope deformation areas, and extract the deformation center point coordinate information, deformation start time phase information, and deformation end time phase information of each suspected slope deformation area.

[0150] A geographic information system (GIS) platform loads a multi-period time-series remote sensing image dataset. Using differential interferometry (DII) with synthetic aperture radar (SAR), adjacent SAR image pairs are processed to generate interferograms, allowing the extraction of surface deformation for each pixel between the two time phases. For optical images, digital image correlation (DIC) techniques are used to analyze the displacement of corresponding feature points in images from different time phases and calculate the deformation. A preset deformation threshold (e.g., 10 mm) is set, and all pixels with deformation exceeding this threshold are identified as points experiencing significant deformation. Connectivity analysis is performed on these points, grouping spatially connected deformation points into a single deformation region, forming a set of suspected slope deformation regions. For each suspected slope deformation region, its geometric center is extracted as the coordinate information of the deformation center point, and the time when significant deformation begins in the region (deformation initiation phase information) and the time when the deformation reaches its peak (deformation termination phase information) are recorded.

[0151] Step S530: Extract the geographical coordinates of all historical landslide events within the historical time interval from the rainfall-induced landslide event case knowledge base, and calculate the minimum Euclidean distance between the coordinates of the deformation center point of each suspected slope deformation area and the geographical coordinates of all historical landslide events.

[0152] For each suspected slope deformation area identified in step S520, the data processing module extracts the geographic coordinates of all historical landslide events from the rainfall-induced landslide event case knowledge base. Then, it calculates the Euclidean distance between the coordinates of the deformation center point of the deformation area and the geographic coordinates of each historical landslide event, and finds the minimum value among these distances.

[0153] Step S540: Mark suspected slope deformation areas with minimum Euclidean distance less than a preset spatial distance threshold as actual landslide deformation areas, and obtain the event occurrence time information of the historical landslide event that is closest to the actual landslide deformation area from the rainfall-induced landslide event case knowledge base, and use it as the actual landslide occurrence time of the actual landslide deformation area.

[0154] A preset spatial distance threshold is set, for example, 50 meters. For each suspected slope deformation area, if its minimum Euclidean distance to a historical landslide event is less than the preset spatial distance threshold, then the area is determined to have indeed experienced a landslide and is marked as a real landslide deformation area. Simultaneously, from the rainfall-induced landslide event case knowledge base, the historical landslide event corresponding to the minimum distance is identified, and its occurrence time information is used as the actual landslide occurrence time for this real landslide deformation area.

[0155] Step S550: Extract the environmental factor raster cell corresponding to the actual landslide deformation area, obtain the geological environmental element parameter set of the environmental factor raster cell from the geological environmental element parameter set, obtain the rainfall process record data corresponding to the actual landslide occurrence time in the actual landslide deformation area from the rainfall-induced landslide event case knowledge base, and combine the geological environmental element parameter set of the environmental factor raster cell with the rainfall process record data to generate a new associated case data entry.

[0156] For each marked actual landslide deformation area, the platform determines its corresponding environmental factor raster cell based on its geographical location. The geological environmental element parameter set for that raster cell is extracted from the geological environmental element parameter set generated in step S128. Simultaneously, rainfall process records (i.e., rainfall on the triggering day and the preceding 15 days) corresponding to the actual landslide occurrence time in that area are extracted from the rainfall-induced landslide event case knowledge base. The aforementioned geological environmental element parameter set and rainfall process record data are combined, and a new associated case data entry is generated according to the method in step S137.

[0157] Step S560: Add the newly added associated case data entries to the associated case data entry set to form an expanded associated case data entry set containing landslide events automatically identified from remote sensing images.

[0158] All related case data entries generated by automatically identifying and verifying new landslide events through image change detection are added to the original set of related case data entries, thereby expanding the sample library and forming an expanded set of related case data entries.

[0159] Step S570: Re-execute the landslide event sample set construction operation using the expanded associated case data entry set to generate an expanded landslide event sample set containing the features of landslide event samples automatically identified from remote sensing images.

[0160] Following steps S140 to S148, the expanded set of associated case data entries is processed, and the landslide event sample set construction operation is re-executed. This includes relabeling positive and negative samples, balancing sample ratios, and feature standardization, ultimately generating an expanded landslide event sample set that includes the features of the newly discovered landslide event samples.

[0161] Step S580: Input the expanded landslide event sample set into the spatial susceptibility probability prediction model for incremental model training, and use the backpropagation algorithm to update the model parameters of the spatial susceptibility probability prediction model to generate an optimized spatial susceptibility probability prediction model that integrates landslide event samples automatically identified from remote sensing images.

[0162] An expanded set of landslide event samples is used as new training data and input into the existing spatial susceptibility probability prediction model for incremental training. Starting with the existing parameters, the model uses the new sample data and further optimizes the parameters through backpropagation to better fit the old and new data. This process avoids retraining from scratch, making it more efficient. After training, a higher-performing optimized spatial susceptibility probability prediction model is generated.

[0163] Step S590: Recalculate the environmental factor grid cell array using the spatial susceptibility probability prediction optimization model to generate an updated spatial susceptibility probability distribution map.

[0164] Using the newly generated spatial susceptibility probability prediction optimization model, all environmental factor raster cells are re-predicted, and an updated spatial susceptibility probability distribution map is generated according to steps S157 to S159. This spatial susceptibility probability distribution map incorporates the latest landslide event information and reflects the latest state of regional landslide susceptibility.

[0165] Step S5100: Perform spatial difference analysis between the updated spatial susceptibility probability distribution map and the original spatial susceptibility probability distribution map, and calculate the change in spatial susceptibility probability value of each environmental factor grid cell; identify environmental factor grid cells whose spatial susceptibility probability value changes are greater than the preset probability change threshold, and form a set of dynamically enhanced landslide susceptibility areas.

[0166] The updated spatial susceptibility probability distribution map generated in step S590 is subtracted grid-by-grid from the original spatial susceptibility probability distribution map generated in step S159 to obtain a change distribution map. The value of each grid cell in this change distribution map represents the change in susceptibility probability at that location. A preset probability change threshold is set, for example, 0.05. All grid cells with an absolute change value greater than 0.05 are identified; these cells represent areas where landslide susceptibility has changed significantly. These are then grouped to form a set of dynamically enhanced landslide susceptibility areas for further analysis or focused monitoring.

[0167] In one exemplary embodiment, a landslide hazard prediction system for traffic engineering areas is provided, which integrates intelligent algorithms and rainfall threshold models. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2 As shown, it includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a landslide hazard prediction method for traffic engineering areas that integrates intelligent algorithms and rainfall threshold models. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of the traffic engineering area landslide hazard prediction system that integrates intelligent algorithms and rainfall threshold models. It can also be an external keyboard, touchpad, or mouse, etc.

[0168] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models, characterized in that, The method includes: Collect a set of geological environment interpretation data of the target transportation engineering area and a set of historical landslide event records of the target transportation engineering area. The set of historical landslide event records includes the geographical coordinates of the occurrence of each historical landslide event, the time of occurrence of the event, and the rainfall process record data corresponding to the time of occurrence of the event. The geographic information system platform of the target traffic engineering area is invoked to perform spatial overlay analysis on the geological environment interpretation data set, dividing the geological environment interpretation data set into an array of environmental factor raster cells with spatial continuity characteristics, and extracting the set of geological environment element parameters corresponding to each environmental factor raster cell; A knowledge base of cases of rainfall-induced landslide events is constructed. Each historical landslide event in the historical landslide event record set is associated with the corresponding environmental factor raster unit according to the geographical coordinate information of the event occurrence, and associated with the corresponding rainfall process record data according to the time information of the event occurrence, forming a set of associated case data entries containing environmental background parameters of landslide events and rainfall triggering parameters of landslide events. A landslide event sample set construction operation is performed on the associated case data entry set. Environmental factor raster cells belonging to the location of landslide events in the associated case data entry set are marked as positive sample cells, and environmental factor raster cells that have not experienced landslide events in the associated case data entry set are marked as negative sample cells, thereby generating a landslide event sample set containing feature data of positive sample cells and feature data of negative sample cells. The landslide event sample set is input into a preset intelligent algorithm model for model training, and outputs a spatial susceptibility probability prediction model that can characterize the mapping relationship between environmental factor parameters and landslide occurrence probability. The spatial susceptibility probability prediction model is then used to calculate the spatial susceptibility probability value of each environmental factor grid cell in the environmental factor grid cell array, generating a spatial susceptibility probability distribution map covering the entire range of the target traffic engineering area. The construction of a rainfall-induced landslide event case knowledge base involves associating each historical landslide event in the historical landslide event record set with its corresponding environmental factor raster unit based on the geographical coordinates of the event occurrence, and with the corresponding rainfall process record data based on the time of the event occurrence. This forms a set of associated case data entries containing environmental background parameters and rainfall-triggered parameters of the landslide event, including: Extract the geographic coordinates of each historical landslide event from the historical landslide event record set, input the geographic coordinates of the event into the spatial query module of the geographic information system platform, calculate the spatial inclusion relationship between the geographic coordinates of the event and each environmental factor raster cell in the environmental factor raster cell array through the spatial query module, and determine the target environmental factor raster cell containing the geographic coordinates of the event. Extract the event occurrence time information of each historical landslide event from the historical landslide event record set. Based on the event occurrence time information, extract the daily rainfall sequence data within a consecutive preset number of days before the landslide event as the pre-rainfall process data from the rainfall process record data of the historical landslide event. Extract the daily rainfall data on the day the landslide event occurred as the triggering rainfall process data. Extract the daily rainfall sequence data within a consecutive preset number of days after the landslide event as the post-rainfall process data. The set of geological environmental element parameters corresponding to the target environmental factor raster unit is used as the environmental background parameter set of the historical landslide event, and the data of the previous rainfall process, the data of the triggered rainfall process, and the data of the subsequent rainfall process are combined to form the rainfall trigger parameter set of the historical landslide event. The geographical coordinates of each historical landslide event, the time of the event, the set of environmental background parameters, and the set of rainfall triggering parameters are associated and stored to generate a record of rainfall-induced landslide event case data containing landslide event identifier fields, spatial location fields, time fields, environmental factor fields, and rainfall process fields. A spatial location consistency check is performed on the rainfall-induced landslide event case data records to verify whether the environmental factor raster cell corresponding to the spatial location field has the same raster row and column index number as the spatial source raster cell of each geological environmental element parameter in the environmental background parameter set. The rainfall-induced landslide event case data records that pass the spatial location consistency check are marked as valid case data records. All valid case data records are aggregated to form a rainfall-induced landslide event case knowledge base. Each valid case data record in the rainfall-induced landslide event case knowledge base contains the corresponding relationship between the environmental background parameters of the landslide event and the rainfall triggering parameters of the landslide event. Each valid case data record in the rainfall-induced landslide event case knowledge base is traversed. The geological environmental element parameters of the environmental background parameter set in the valid case data record are extracted to form an environmental background parameter vector. The preceding rainfall process data sequence values ​​and the triggered rainfall process data values ​​of the rainfall trigger parameter set in the valid case data record are extracted to form a rainfall trigger parameter vector. The environmental background parameter vector and the rainfall trigger parameter vector are concatenated and combined to generate associated case data entries for the valid case data record. All associated case data entries corresponding to all valid case data records constitute an associated case data entry set.

2. The method for predicting landslide hazard in transportation engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 1, characterized in that, The geological environment interpretation data set of the target transportation engineering area and the historical landslide event record set of the target transportation engineering area include: The remote sensing image data set of the target transportation engineering area is obtained through a satellite remote sensing image interpretation platform, and the geological structure linear body automatic extraction operation is performed on the remote sensing image data set to extract the fault structure linear body distribution layer and fold structure boundary layer of the target transportation engineering area. The digital elevation model data set of the target traffic engineering area is obtained through the digital elevation model data acquisition platform. The terrain factor derivation calculation operation is performed on the digital elevation model data set to generate the terrain slope factor raster layer, terrain aspect factor raster layer, terrain curvature factor raster layer and terrain relief factor raster layer of the target traffic engineering area. The geological lithology distribution map of the target transportation engineering area is obtained through a geological map digitization platform. The lithology unit attribute extraction operation is performed on the geological lithology distribution map to extract the rock type attribute parameters, rock structure attribute parameters, and rock weathering degree attribute parameters of each lithology unit, forming a geological lithology distribution raster layer. The hydrogeological condition data set of the target transportation engineering area is obtained through the hydrogeological survey database. The hydrogeological condition data set is then subjected to a spatialization operation of hydrological parameters to generate a groundwater burial depth raster layer and a groundwater runoff modulus raster layer for the target transportation engineering area. The land use type data set of the target transportation engineering area is obtained through the land use status survey database. Land use type classification and coding operation is performed on the land use type data set to generate a land use type raster layer of the target transportation engineering area. Spatial registration and raster overlay operations are performed on the fault structure linear distribution layer, the fold structure morphology boundary layer, the topographic slope factor raster layer, the topographic aspect factor raster layer, the topographic curvature factor raster layer, the topographic relief factor raster layer, the geological lithology distribution raster layer, the groundwater burial depth raster layer, the groundwater runoff modulus raster layer, and the land use type raster layer to generate a geological environment interpretation data set containing multi-dimensional geological environment element parameters at each spatial location point; A list of historical landslide events in the target transportation engineering area is obtained from a geological disaster historical record database. Each record in the historical landslide event record list includes the latitude and longitude coordinates of the landslide event location, the year, month, day, hour and minute of the landslide event, the daily rainfall sequence data within a preset number of consecutive days before the landslide event, the daily rainfall data on the day the landslide event occurred, and the daily rainfall sequence data within a preset number of consecutive days after the landslide event. Data cleaning is performed on the historical landslide event record list to remove invalid records where the latitude and longitude coordinates of the landslide event location exceed the boundary of the target transportation engineering area, and records with missing data in the daily rainfall sequence data, thereby generating a cleaned set of historical landslide event records.

3. The method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 1, characterized in that, The process involves calling the geographic information system platform of the target transportation engineering area to perform spatial overlay analysis on the geological environment interpretation data set. This divides the geological environment interpretation data set into an array of environmental factor raster cells with spatial continuity, and extracts the set of geological environment element parameters corresponding to each environmental factor raster cell, including: The rasterization processing module of the geographic information system platform receives the fault structure linear body distribution layer from the geological environment interpretation data set, and converts the fault structure linear body distribution layer into a fault structure linear body distance raster layer with uniform spatial resolution. The value of each raster cell in the fault structure linear body distance raster layer is the Euclidean distance value of the center point of the raster cell from the nearest fault structure linear body. The rasterization processing module of the geographic information system platform receives the fold structure boundary layer from the geological environment interpretation dataset, and converts the fold structure boundary layer into a fold structure distance raster layer with uniform spatial resolution. The value of each raster cell in the fold structure distance raster layer is the Euclidean distance value of the center point of the raster cell from the nearest fold structure boundary. The geographic information system platform receives the topographic slope factor raster layer, topographic aspect factor raster layer, topographic curvature factor raster layer, and topographic relief factor raster layer from the geological environment interpretation dataset through the rasterization processing module. The topographic slope factor raster layer, the topographic aspect factor raster layer, the topographic curvature factor raster layer, and the topographic relief factor raster layer are processed to unify the spatial resolution, generating a topographic slope factor raster layer, the topographic aspect factor raster layer, the topographic curvature factor raster layer, and the topographic relief factor raster layer with unified spatial resolution. The rasterization processing module of the geographic information system platform receives the geological lithology distribution raster layer from the geological environment interpretation data set, performs a lithology type coding reclassification operation on the geological lithology distribution raster layer, and converts the rock type attribute parameters, rock structure attribute parameters, and rock weathering degree attribute parameters of each raster unit in the geological lithology distribution raster layer into lithology type coding values ​​to generate a lithology type coding raster layer. The geographic information system platform's rasterization processing module receives groundwater burial depth raster layers, groundwater runoff modulus raster layers, and land use type raster layers from the geological environment interpretation dataset. It then performs spatial resolution unification processing on these three raster layers to generate groundwater burial depth raster layers, runoff modulus raster layers, and land use type raster layers with unified spatial resolution. The fault structure linear volume distance raster layer, the fold structure morphology distance raster layer, the slope factor raster layer, the aspect factor raster layer, the curvature factor raster layer, the relief factor raster layer, the lithology type coding raster layer, the groundwater depth raster layer, the runoff modulus raster layer, and the land use type raster layer are subjected to spatial coordinate system unification processing to generate a multi-layer overlay raster dataset with the same spatial range, the same spatial resolution, and the same projected coordinate system. Based on the raster row and column number configuration parameters of the multi-layer overlay raster dataset, the multi-layer overlay raster dataset is divided into an environmental factor raster cell array, and each environmental factor raster cell in the environmental factor raster cell array corresponds to a unique set of raster row and column index numbers. Traverse each environmental factor raster cell in the environmental factor raster cell array, and extract the fault distance attribute value in the fault structure linear body distance raster layer, the fold distance attribute value in the fold structure morphology distance raster layer, the slope attribute value in the slope factor raster layer, the slope aspect attribute value in the aspect factor raster layer, the curvature attribute value in the curvature factor raster layer, the relief attribute value in the relief factor raster layer, the lithology code attribute value in the lithology type code raster layer, the groundwater depth attribute value in the groundwater depth raster layer, the runoff modulus attribute value in the runoff modulus raster layer, and the land use type code attribute value in the land use type raster layer to generate the geological environmental element parameter set corresponding to the environmental factor raster cell.

4. The method for predicting landslide hazard in transportation engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 1, characterized in that, The step of constructing a landslide event sample set by performing a landslide event sample set construction operation on the associated case data entry set involves marking environmental factor raster cells belonging to the location of landslide events in the associated case data entry set as positive sample cells and marking environmental factor raster cells in the associated case data entry set that have not experienced landslide events as negative sample cells, thereby generating a landslide event sample set containing feature data of both positive and negative sample cells. Extract the raster row and column index number of the environmental factor raster cell corresponding to each associated case data entry in the associated case data entry set, and summarize the raster row and column index numbers corresponding to all associated case data entries to form a set of raster index numbers for landslide events. Extract the raster row and column index numbers of all environmental factor raster cells in the environmental factor raster cell array to form a complete set of raster index numbers for the entire region; By performing set difference operations, the raster row and column index numbers belonging to the set of raster index numbers where landslide events occurred are removed from the set of raster index numbers for the entire region, resulting in the set of raster index numbers where no landslide events occurred. Traverse each raster row and column index number in the set of raster index numbers for landslide events, locate the environmental factor raster cell corresponding to the raster row and column index number, extract the values ​​of each geological environmental element parameter corresponding to the environmental factor raster cell from the set of geological environmental element parameters to form a positive sample cell feature vector, associate and store the positive sample cell feature vector with the preset positive sample category label to generate positive sample cell feature data. Traverse each raster row and column index number in the set of raster index numbers where no landslide event has occurred, locate the environmental factor raster cell corresponding to the raster row and column index number, extract the geological environmental element parameter values ​​corresponding to the environmental factor raster cell from the set of geological environmental element parameters to form a negative sample cell feature vector, associate and store the negative sample cell feature vector with the preset negative sample category label to generate negative sample cell feature data. According to the preset ratio of positive sample units to negative sample units, a specified number of positive sample unit feature vectors and corresponding positive sample category labels are randomly extracted from the positive sample unit feature data to form a positive sample subset, and a specified number of negative sample unit feature vectors and corresponding negative sample category labels are randomly extracted from the negative sample unit feature data to form a negative sample subset. The positive sample subset and the negative sample subset are merged to form an initial landslide event sample set; The feature vectors of all sample records in the initial landslide event sample set are subjected to feature standardization processing so that the feature values ​​corresponding to each geological environmental element parameter conform to the preset value distribution range, thereby generating a standardized landslide event sample set. Each sample record in the standardized landslide event sample set contains a standardized sample feature vector and a corresponding sample category label.

5. The method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 1, characterized in that, The process involves inputting the landslide event sample set into a preset intelligent algorithm model for model training, outputting a spatial susceptibility probability prediction model that characterizes the mapping relationship between environmental factor parameters and landslide occurrence probabilities, and using this spatial susceptibility probability prediction model to calculate the spatial susceptibility probability value for each environmental factor grid cell in the environmental factor grid cell array, generating a spatial susceptibility probability distribution map covering the entire target traffic engineering area, including: Obtain the initial model parameter configuration information of the preset intelligent algorithm model, which includes input feature dimension parameters, hidden layer node count parameters, hidden layer number parameters, output layer node count parameters, and activation function type parameters; The landslide event sample set is divided into a model training sample subset and a model validation sample subset. The model training sample subset is used to optimize the model parameters of the preset intelligent algorithm model, and the model validation sample subset is used to evaluate the generalization performance of the preset intelligent algorithm model. The sample feature vector of each sample record in the training sample subset of the model is input into the input layer of the preset intelligent algorithm model. The sample feature vector is subjected to nonlinear transformation processing through the hidden layer of the preset intelligent algorithm model to extract the high-order combination feature representation of the sample feature vector. The high-order combination feature representation is subjected to probability mapping processing through the output layer of the preset intelligent algorithm model to output the landslide occurrence probability prediction value corresponding to the sample record. Calculate the loss function value between the predicted landslide occurrence probability and the sample category label of the sample record; calculate the gradient information of each model parameter in the preset intelligent algorithm model based on the loss function value through backpropagation algorithm; and update the model parameters of the preset intelligent algorithm model based on the gradient information. Repeatedly execute the operation of inputting the sample feature vector of the sample record in the sample subset of the model training sample into the preset intelligent algorithm model and updating the model parameters until the loss function value converges to the preset convergence threshold range, stop the model parameter update operation, and use the preset intelligent algorithm model in the current model parameter state as the spatial susceptibility probability prediction model. The sample feature vector of each sample record in the model validation sample subset is input into the spatial susceptibility probability prediction model to obtain the landslide occurrence probability prediction value of the validation sample output by the spatial susceptibility probability prediction model. The validation loss function value between the landslide occurrence probability prediction value of the validation sample and the sample category label of the sample record in the model validation sample subset is calculated. When the validation loss function value exceeds the preset validation loss threshold, the initial model parameter configuration information of the preset intelligent algorithm model is readjusted and the model training operation is re-performed. Obtain the set of geological environmental element parameters corresponding to each environmental factor raster cell in the environmental factor raster cell array, and arrange and combine the values ​​of each geological environmental element parameter in the set of geological environmental element parameters according to the feature order of the sample feature vector to generate the prediction feature vector of the environmental factor raster cell. The predicted feature vector of each environmental factor grid cell is input into the spatial susceptibility probability prediction model to obtain the landslide occurrence probability prediction value of the environmental factor grid cell output by the spatial susceptibility probability prediction model, and the landslide occurrence probability prediction value is used as the spatial susceptibility probability value of the environmental factor grid cell. A spatial susceptibility probability distribution map is generated based on the raster row and column index number of each environmental factor raster cell and its corresponding spatial susceptibility probability value. The pixel value of each raster position in the spatial susceptibility probability distribution map corresponds to the spatial susceptibility probability value of the environmental factor raster cell in which the raster position is located.

6. The method for predicting landslide hazard in transportation engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 1, characterized in that, The method further includes: The real-time monitoring data sequence of rainfall from meteorological monitoring stations and the inversion data sequence of rainfall from meteorological satellite cloud images of the target traffic engineering area are obtained. The real-time monitoring data sequence of rainfall from meteorological monitoring stations and the inversion data sequence of rainfall from meteorological satellite cloud images are spatiotemporally fused to generate a raster layer sequence of hourly rainfall across the entire target traffic engineering area. Extract the continuous hourly rainfall value sequence of each grid cell in the hourly rainfall raster layer sequence within a preset time window, calculate the cumulative rainfall value of the grid cell within the preset time window based on the continuous hourly rainfall value sequence, and generate the cumulative rainfall time series data of the grid cell. Obtain the preset rainfall infiltration coefficient configuration parameters, and calculate the previous effective cumulative rainfall value sequence of the grid cell within the preset time window by multiplying the cumulative rainfall value at each time point in the cumulative rainfall time series data with the rainfall infiltration coefficient configuration parameters. Extract the hourly rainfall value corresponding to the target time for each grid cell in the hourly rainfall raster layer sequence as the trigger rainfall value for that grid cell, and extract the previous effective cumulative rainfall value corresponding to the target time from the previous effective cumulative rainfall value sequence as the target previous effective rainfall value for that grid cell; A rainfall duration parameter extraction function is constructed to determine the rainfall duration parameter value of each grid cell based on the time length parameter of the preset time window; The target effective rainfall value and rainfall duration parameter value of each grid cell are input into a preset continuous probability rainfall threshold model. The preset continuous probability rainfall threshold model includes a logistic regression function expression with the rainfall duration parameter value as the independent variable and the target effective rainfall value as the independent variable. The probability value of landslide occurrence of the grid cell at the target time is calculated through the logistic regression function expression. The landslide time occurrence probability value of each grid cell is multiplied with the spatial susceptibility probability value of that grid cell in the spatial susceptibility probability distribution map to generate the comprehensive landslide hazard probability value of that grid cell. Based on the raster row and column index number of each raster cell and its corresponding comprehensive landslide hazard probability value, a comprehensive landslide hazard probability distribution map covering the entire target traffic engineering area is generated.

7. The method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 6, characterized in that, The process involves acquiring real-time rainfall monitoring data sequences from meteorological monitoring stations and rainfall inversion data sequences from meteorological satellite cloud images of the target transportation engineering area. Then, the real-time rainfall monitoring data sequences from meteorological monitoring stations and the rainfall inversion data sequences from meteorological satellite cloud images are spatiotemporally fused to generate a raster layer sequence of hourly rainfall across the entire target transportation engineering area. This includes: The real-time rainfall monitoring data sequence reported by meteorological monitoring stations within and around the target traffic engineering area within a preset distance range and within a preset time interval is obtained through the meteorological monitoring data receiving interface. The real-time rainfall monitoring data sequence includes the geographical coordinates of each meteorological monitoring station, the monitoring time of the station, and the rainfall value corresponding to the monitoring time of the station. The meteorological satellite cloud image rainfall inversion data sequence covering the target transportation engineering area is obtained through the meteorological satellite data receiving interface. The meteorological satellite cloud image rainfall inversion data sequence includes the raster geographic coordinate range information of each cloud image raster unit, cloud image acquisition time information, and rainfall inversion value corresponding to the cloud image acquisition time. The real-time rainfall monitoring data sequence is grouped according to the station monitoring time information to generate a set of rainfall values ​​for each meteorological monitoring station corresponding to each monitoring time. Each set of rainfall values ​​for each meteorological monitoring station includes the station geographic coordinates and rainfall values ​​of each meteorological monitoring station at that monitoring time. The meteorological satellite cloud image rainfall inversion data sequence is grouped according to the cloud image acquisition time information to generate a satellite cloud image rainfall inversion raster layer corresponding to each acquisition time. Each satellite cloud image rainfall inversion raster layer contains the raster geographic coordinate range information and rainfall inversion value of each cloud image raster unit at that acquisition time. The set of rainfall values ​​from meteorological monitoring stations corresponding to each monitoring time is spatiotemporally matched with the rainfall inversion raster layer of satellite cloud image corresponding to that monitoring time. The rainfall inversion values ​​of each meteorological monitoring station in the corresponding cloud image raster cell in the rainfall inversion raster layer of satellite cloud image are extracted, and the deviation between the rainfall monitoring values ​​and the rainfall inversion values ​​of each meteorological monitoring station is calculated. The spatial distribution variation function of the deviation values ​​is calculated based on the deviation values ​​of all meteorological monitoring stations. The spatial distribution variation function of the deviation values ​​is then spatially interpolated using the Kriging interpolation method to generate a rainfall deviation correction raster layer covering the entire target traffic engineering area. The satellite cloud image rainfall inversion raster layer and the rainfall deviation correction raster layer are superimposed and corrected. The rainfall inversion value of each cloud image raster unit in the satellite cloud image rainfall inversion raster layer is added to the deviation correction value of the corresponding raster unit in the rainfall deviation correction raster layer to generate the corrected hourly rainfall raster layer at the acquisition time. Arrange the corrected hourly rainfall raster layers corresponding to each collection time in chronological order of collection time to generate an hourly rainfall raster layer sequence.

8. The method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 6, characterized in that, After multiplying the landslide occurrence probability value of each grid cell with the spatial susceptibility probability value of that grid cell in the spatial susceptibility probability distribution map to generate the comprehensive landslide hazard probability value of that grid cell, the method further includes: Obtain a vector map layer showing the distribution of traffic engineering facilities within the target traffic engineering area. The vector map layer contains the geographic coordinate boundary information and the type attribute information of each traffic engineering facility. The traffic engineering facilities distribution vector layer is spatially overlaid with the landslide hazard comprehensive probability distribution map to identify traffic engineering facilities that have spatial intersection with the grid cells in the landslide hazard comprehensive probability distribution map whose landslide hazard comprehensive probability value exceeds a preset probability threshold, and these are marked as a set of high-risk traffic engineering facilities. Extract the geographic coordinate boundary information of each traffic engineering facility in the set of high-risk traffic engineering facilities, and calculate the statistical characteristic parameters of the comprehensive probability value of landslide risk of all grid cells within the grid cell range covered by the geographic coordinate boundary information of the traffic engineering facility. The statistical characteristic parameters include the maximum value parameter of comprehensive probability of landslide risk, the average value parameter of comprehensive probability of landslide risk, and the standard deviation parameter of comprehensive probability of landslide risk. Based on the traffic engineering facility type attribute information of each traffic engineering facility, the risk level classification threshold parameter of the traffic engineering facility is obtained from the preset facility type risk level mapping table. The maximum value parameter of the comprehensive probability of landslide hazard is compared with the risk level classification threshold parameter to determine the landslide risk level identifier of each traffic engineering facility. A landslide risk warning list for each transportation engineering facility is generated based on the landslide risk level identifier and the geographical coordinate boundary information of that transportation engineering facility. The landslide risk warning list for each transportation engineering facility in the high-risk transportation engineering facility set includes facility identifier information, facility geographical coordinate boundary information, landslide risk level identifier, and corresponding comprehensive probability statistical characteristic parameters of landslide hazard for each transportation engineering facility. The landslide risk warning list for the transportation engineering facilities will be pushed to the risk warning information receiving terminal of the transportation engineering management department.

9. The method for predicting landslide hazard in traffic engineering areas by integrating intelligent algorithms and rainfall threshold models according to claim 6, characterized in that, After multiplying the landslide occurrence probability value of each grid cell with the spatial susceptibility probability value of that grid cell in the spatial susceptibility probability distribution map to generate the comprehensive landslide hazard probability value of that grid cell, the method further includes: Obtain a vector map layer showing the distribution of residential areas within the target transportation engineering area. The vector map layer includes the geographic coordinate boundary information of each residential area and the population size attribute information of each residential area. The distribution vector layer of the residential area is spatially overlaid with the comprehensive probability distribution map of landslide risk. Residential areas that spatially intersect with grid cells in the comprehensive probability distribution map of landslide risk exceeding a preset probability threshold are identified and marked as a set of high-risk residential areas. Extract the geographic coordinate boundary information of each residential area in the set of high-risk residential areas, and calculate the weighted average of the comprehensive probability values ​​of landslide risk of all grid cells within the grid cell range covered by the geographic coordinate boundary information of the residential area. The weight coefficient of each grid cell is obtained by inverse distance weighting based on the Euclidean distance between the geographic coordinates of the grid cell and the geographic center coordinates of the residential area, thereby generating the weighted comprehensive probability value of landslide risk for the residential area. Based on the population size attribute information of each residential area and the weighted comprehensive probability value of landslide risk for that residential area, the estimated number of people exposed to landslide risk in each residential area is calculated. The estimated number of people exposed to landslide risk is the product of the population size attribute information of the residential area and the weighted comprehensive probability value of landslide risk. The geographic coordinate boundary information, weighted comprehensive probability value of landslide risk, and estimated number of people exposed to landslide risk for each residential area are linked and stored to generate a set of landslide risk assessment data for residential areas. Based on the weighted comprehensive probability value of landslide hazard in the landslide risk assessment dataset of the residential areas, all residential areas in the high-risk residential area set are sorted to generate a landslide risk ranking list of residential areas arranged in descending order of risk level. The landslide risk ranking list of the residential areas will be pushed to the emergency response command terminal of the emergency management department.

Images