Geological disaster meteorological risk early warning method and system based on machine learning

Abstract

The invention relates to the technical field of data processing, and discloses a geological disaster meteorological risk early warning method and system based on machine learning. The method comprises the following steps: acquiring rainfall intensity, soil saturation, underground water level change and slope runoff coefficient by a multi-source sensor, and constructing a geological disaster meteorological data set; performing sensitivity weight distribution on the meteorological factors according to geological conditions to obtain a weight matrix; carrying out weighted fusion on the weight matrix and meteorological time series data, and extracting features through a geological constraint long-short-term memory network to obtain a risk probability vector; dynamically adjusting an early warning threshold value based on the safety coefficient change rate; and carrying out Bayesian fusion on the risk probability vector and an adaptive early warning threshold to obtain a graded early warning result. The technical problem that an existing geological disaster early warning technology lacks a multivariate meteorological factor intelligent weight distribution and geological condition adaptive threshold adjustment mechanism is solved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a geological disaster meteorological risk early warning method and system based on machine learning. Background Art

[0002] Existing geological disaster early warning technologies primarily rely on traditional statistical models and simple threshold-based methods, using single rainfall monitoring data to construct rainfall intensity-duration (ID) curves for geological disaster warning. These methods typically employ fixed warning thresholds and combine historical disaster statistics with expert experience to assess the risk of geological hazards such as landslides and debris flows in specific areas. Some technologies also incorporate simple machine learning algorithms, such as support vector machines and random forests, to analyze and process multidimensional meteorological data.

[0003] However, existing technologies have significant shortcomings: first, traditional methods mainly rely on a single rainfall data source, ignoring the complex coupling relationship between multiple meteorological factors such as soil saturation, groundwater level changes, and slope runoff, resulting in low warning accuracy; second, existing machine learning methods use general algorithms, lack special designs for geological disaster characteristics, and fail to fully consider the sensitivity differences of various meteorological factors under different geological conditions; third, traditional early warning systems use fixed warning thresholds and cannot be adaptively adjusted according to dynamic changes in slope stability, reducing the timeliness and accuracy of warnings.

[0004] Based on the above analysis, the core problem of existing technologies lies in the lack of an intelligent fusion mechanism for multiple meteorological factors and an adaptive weight allocation strategy for geological conditions. Due to the significant differences in the contribution of various meteorological factors to disaster triggering under different geological environments, the traditional unified weight allocation method cannot accurately reflect this difference, which in turn affects the accuracy of the early warning model. At the same time, the fixed threshold early warning mechanism ignores the dynamic evolution of slope stability. When the safety factor continues to decline, the early warning system cannot adjust its sensitivity in time, resulting in a delayed warning. Therefore, there is an urgent need for a machine learning early warning method that can realize intelligent weight allocation of multiple meteorological factors, integrate geophysical constraints, and have the ability to adaptively adjust the threshold according to changes in slope stability. Summary of the Invention

[0005] This application provides a geological disaster meteorological risk warning method and system based on machine learning, which is used to solve the technical problems in existing geological disaster warning technologies, such as the lack of intelligent weight allocation of multiple meteorological factors and adaptive threshold adjustment mechanism for geological conditions.

[0006] In the first aspect, the present application provides a geological disaster meteorological risk warning method based on machine learning, which includes: real-time collection and processing of rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient in the target area through multi-source sensors to obtain a geological disaster meteorological data set; sensitivity weight allocation processing of each meteorological factor in the geological disaster meteorological data set according to geological condition classification to obtain a geological disaster sensitivity weight matrix; weighted fusion processing of the geological disaster sensitivity weight matrix and meteorological time series data, and nonlinear feature extraction of the fused data through a geological constraint long and short-term memory network to obtain a geological disaster risk probability vector; dynamic adjustment processing of the warning threshold according to the safety factor change rate, and increasing the threshold sensitivity when the safety factor change rate continues to decrease to obtain an adaptive warning threshold parameter; Bayesian fusion processing of the geological disaster risk probability vector and the adaptive warning threshold parameter to obtain a graded geological disaster warning result.

[0007] In a second aspect, the present application provides a geological disaster meteorological risk early warning system based on machine learning, the geological disaster meteorological risk early warning system based on machine learning includes: The acquisition module is used to collect and process the rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient of the target area in real time through multi-source sensors to obtain a geological disaster meteorological data set; An allocation module is used to perform sensitivity weight allocation processing on each meteorological factor in the geological disaster meteorological data set according to geological condition classification to obtain a geological disaster sensitivity weight matrix; A weighting module is used to perform weighted fusion processing on the geological hazard sensitivity weight matrix and the meteorological time series data, and to extract nonlinear features of the fused data through a geological constraint long short-term memory network to obtain a geological hazard risk probability vector; An adjustment module is used to dynamically adjust the warning threshold according to the safety factor change rate, and when the safety factor change rate continues to decrease, the threshold sensitivity is increased to obtain an adaptive warning threshold parameter; The fusion module is used to perform Bayesian fusion processing on the geological disaster risk probability vector and the adaptive warning threshold parameter to obtain a graded geological disaster warning result.

[0008] In a third aspect, a geological disaster meteorological risk warning device based on machine learning is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the geological disaster meteorological risk warning device based on machine learning executes the above-mentioned geological disaster meteorological risk warning method based on machine learning.

[0009] In a fourth aspect, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned geological disaster meteorological risk warning method based on machine learning.

[0010] In the technical solution provided by this application, multi-source sensors are used to collect multiple meteorological factors such as rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient in real time, breaking through the limitation of traditional geological disaster early warning technology that relies only on single rainfall data, establishing a comprehensive geological disaster meteorological data set, and significantly improving the integrity and reliability of the data source. By assigning sensitivity weights to each meteorological factor according to geological conditions, this application innovatively constructs a geological disaster sensitivity weight matrix, which can accurately quantify the different contributions of each meteorological factor to disaster triggering under different geological environments, solving the problem of low warning accuracy caused by the use of fixed weight allocation in traditional methods. The geological constraint long short-term memory network is used as the core algorithm. By incorporating the geophysical constraint mechanism, the neural network can learn characteristic patterns that conform to the principles of geological engineering. Compared with the general time series prediction algorithm, the network can more accurately capture the inducing laws of geological disasters, and the extracted geological disaster risk probability vector has stronger physical interpretation and prediction reliability.

[0011] The dynamic adjustment mechanism of the safety factor change rate of the present application realizes the adaptive optimization of the warning threshold, and automatically increases the threshold sensitivity when the safety factor change rate continues to decrease, which effectively solves the technical problem that the traditional fixed threshold warning system cannot respond to the dynamic changes of slope stability, and significantly improves the timeliness and accuracy of the warning. Bayesian fusion processing probabilistically fuses the machine learning prediction results with the physical model calculation results, combines the learning ability of the data-driven method and the physical constraints of the mechanism model, and the resulting graded geological disaster warning results have higher credibility and practicality. Especially in the specific application field of geological disaster meteorological risk warning, the gated modulation mechanism of the geological constraint long short-term memory network can automatically adjust the information processing strategy according to different geological conditions, so that the features learned by the network better reflect the disaster risk characteristics under different geological environments such as granite weathering soil and shale fracture zone. Compared with traditional machine learning methods, the warning accuracy under complex geological-meteorological coupling conditions is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1This is a schematic diagram of an embodiment of a geological disaster meteorological risk early warning method based on machine learning in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a geological disaster meteorological risk early warning system based on machine learning in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of a geological disaster meteorological risk warning device based on machine learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a method and system for early warning of geological disaster meteorological risks based on machine learning. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0015] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the geological disaster meteorological risk early warning method based on machine learning includes: Step S101: Using multi-source sensors, real-time data collection and processing is performed on the rainfall intensity, soil saturation, groundwater level change, and slope runoff coefficient of the target area to obtain a geological disaster meteorological data set; Step S102: performing sensitivity weighting processing on each meteorological factor in the geological hazard meteorological data set according to the geological condition classification to obtain a geological hazard sensitivity weight matrix; Step S103: Perform weighted fusion processing on the geological hazard sensitivity weight matrix and the meteorological time series data, perform nonlinear feature extraction on the fused data through the geological constraint long short-term memory network, and obtain the geological hazard risk probability vector; Step S104: dynamically adjust the warning threshold according to the safety factor change rate, and increase the threshold sensitivity when the safety factor change rate continues to decrease, to obtain an adaptive warning threshold parameter; Step S105: Perform Bayesian fusion processing on the geological disaster risk probability vector and the adaptive warning threshold parameter to obtain a graded geological disaster warning result.

[0016] It is understandable that the execution subject of this application can be a geological disaster meteorological risk warning system based on machine learning, or a terminal or server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0017] Specifically, multi-source sensor data collection is the data foundation of the entire early warning system. When the automatic weather station collects rainfall intensity data, the sensor records the rainfall once an hour. By comparing and screening with the preset effective rainfall intensity range, the outliers are eliminated to form an effective rainfall intensity sequence. The soil moisture sensor monitors soil saturation in real time. The original saturation value needs to be processed with boundary constraints according to the physical reasonable range to ensure that the data is within a reasonable range. The groundwater level monitor continuously collects groundwater level change data. By calculating the difference between the water levels at adjacent time points and dividing it by the time interval, a groundwater level change rate sequence is obtained. The slope flow monitoring equipment collects runoff data, divides the measured runoff volume by the corresponding rainfall to calculate the runoff coefficient value, and forms a slope runoff coefficient sequence.

[0018] Constructing a geological disaster sensitivity weight matrix, which is one of the core innovations of the present invention. Based on the geological exploration data of the target area, the slope, lithology, and soil type are classified and identified to obtain geological condition classification labels. Based on the geological condition classification labels, historical geological disaster cases are classified and statistically processed, and the correlation coefficients between different meteorological factors and disaster occurrence under various geological conditions are calculated. The gradient descent algorithm is an optimization algorithm that iteratively optimizes the weight parameters by calculating the gradient of the loss function and iteratively updating the weight parameters until convergence. The sensitivity weight allocation process is based on the optimized weight parameters, and weights are allocated to rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficients to construct a matrix structure with geological conditions as rows and meteorological factors as columns.

[0019] Implement feature extraction for a geologically constrained long-short-term memory network. A LSTM is a special recurrent neural network capable of memorizing long-term dependencies and controlling information flow through forget gates, input gates, and output gates. The geologically constrained LSTM incorporates geophysical constraints based on the standard LSTM. The geological condition embedding vector is encoded using an embedding layer. The slope classification code, lithology code vector, and soil type code are concatenated and combined to construct a comprehensive geological condition code sequence. Geophysical constraint modulation modulates the three gates of the network based on the geological condition embedding vector, enabling the network to learn characteristic patterns that conform to geophysical laws.

[0020] Dynamic calculation of the safety factor is achieved through the Bishop arc sliding method. The Bishop arc sliding method is a commonly used slope stability analysis method in geological engineering. This method assumes that the sliding surface is an arc and calculates the safety factor based on the principle of moment balance. The rate of change of the safety factor is obtained by performing differential operations on the real-time safety factor in a continuous time period, and calculating the ratio of the difference in safety factors between adjacent time points to the time interval. When the rate of change of the safety factor is continuously negative and the absolute value exceeds the preset drop threshold, the sensitivity adjustment mechanism is triggered, and the baseline warning threshold is adjusted upward by a percentage. The soil moisture correction coefficient is calculated by using the ratio of the current saturation to the field water holding capacity saturation as a correction factor to correct the moisture impact on the adjusted warning threshold.

[0021] Bayesian fusion processing probabilistically fuses machine learning predictions with physical model calculations. The physical model risk probability is calculated using a normal distribution based on the real-time safety factor, with the difference between the safety factor and the baseline value divided by the standard deviation as the input parameter. The Monte Carlo sampling method, a numerical calculation method based on random sampling, calculates the variance of the prediction results by performing multiple random sampling on the effective cohesion and internal friction angle parameters, thereby deriving a quantitative uncertainty probability. Weighted linear combination processing probabilistically fuses the geological hazard risk probability vector, the physical model risk probability, and the quantitative uncertainty probability according to preset weighting coefficients.

[0022] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The rainfall intensity data is collected and processed hourly by the automatic weather station, and the collected data is compared and screened with the set effective rainfall intensity range to obtain the effective rainfall intensity sequence; The soil saturation data is monitored and processed in real time through soil moisture sensors. The collected original saturation values ​​are bounded according to the preset physical reasonable range to obtain a standard soil saturation sequence. The groundwater level change data is continuously collected and processed by the groundwater level monitoring instrument, and the water level difference between adjacent time points is calculated and divided by the time interval to obtain the groundwater level change rate series; The runoff data is collected and processed based on the slope flow monitoring equipment, and the runoff coefficient value is calculated by dividing the measured runoff volume by the corresponding rainfall to obtain the slope runoff coefficient sequence; The effective rainfall intensity series, standard soil saturation series, groundwater level change rate series, and slope runoff coefficient series are time-synchronized and aligned to obtain a geological disaster meteorological dataset.

[0023] Specifically, when the automatic weather station collects and processes rainfall intensity data on an hourly basis, the rainfall intensity sensor records the rainfall value once an hour. The collected raw data needs to be compared and screened with the set effective rainfall intensity range. The comparison and screening process is implemented through a numerical comparison algorithm, comparing each collected data point with the preset upper and lower thresholds one by one. Abnormal data points that exceed the range are marked and eliminated, and the retained data points are arranged in chronological order to form an effective rainfall intensity sequence. When the soil moisture sensor monitors and processes soil saturation data in real time, the sensor probe is inserted into the soil to measure the soil moisture content. The raw saturation value is converted into a digital signal through the sensor's resistance change or frequency change. Boundary constraint processing refers to comparing the collected raw saturation value with a preset physical reasonable range. The physical reasonable range is determined by soil type and geological conditions. Values ​​outside the reasonable range are adjusted to the boundary value. The adjusted data are arranged in a time series to form a standard soil saturation sequence.

[0024] The groundwater level monitor continuously collects and processes groundwater level change data, which includes two data processing processes: water level difference calculation and change rate calculation. The water level difference calculation is obtained by subtracting the water level readings at adjacent time points. The change rate calculation divides the water level difference by the time interval, which is usually one hour. The result of the division operation represents the rate of change of the groundwater level per unit time. The continuous change rate values ​​form a groundwater level change rate sequence in chronological order.

[0025] When slope flow monitoring equipment collects and processes runoff data, flow sensors measure the flow rate through a specific section. The runoff coefficient is calculated by dividing the measured runoff by the rainfall in the corresponding time period. The physical meaning of the division operation is the ratio of rainfall to surface runoff. The runoff coefficient reflects the permeability and runoff generation capacity of the slope. The calculated runoff coefficient values ​​are arranged in chronological order to form a slope runoff coefficient sequence. Time synchronization alignment processing aligns data sequences from four different sources according to a unified time base. The alignment process first establishes a unified time axis and then matches each data sequence according to the corresponding timestamp. When a certain data point is missing at a certain time point, linear interpolation is used to fill the missing value. Linear interpolation calculates the missing value by taking a weighted average of the data values ​​of adjacent time points. The aligned data forms a unified data matrix structure. The rows of the matrix represent time points and the columns represent different meteorological factors. This data matrix is ​​the geological disaster meteorological data set.

[0026] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Classify and identify the slope, lithology, and soil type based on the geological exploration data of the target area to obtain geological condition classification labels; Based on the geological condition classification labels, historical geological disaster cases were classified and statistically processed, and the correlation coefficients between different meteorological factors and disaster occurrence under different geological conditions were calculated to obtain the factor correlation coefficient matrix; The factor correlation coefficient matrix is ​​normalized according to the geological condition dimension, and the weight parameters are iteratively optimized using the gradient descent algorithm to obtain the optimized weight parameters; Based on the optimized weight parameters, sensitivity weights are assigned to rainfall intensity, soil saturation, groundwater level change, and slope runoff coefficient. A matrix structure with geological conditions as rows and meteorological factors as columns is constructed to obtain the initial sensitivity weight matrix. The initial sensitivity weight matrix is ​​verified by expert knowledge based on geological engineering experience to obtain the geological hazard sensitivity weight matrix.

[0027] Specifically, the geological condition classification and identification processing is based on the geological exploration data of the target area. The slope classification and identification converts the continuous slope value into a classification label by measuring the surface inclination angle and discretizing it according to the preset interval. The lithology classification and identification determines the rock type through the mineral composition analysis and physical property testing of the rock samples. The soil type classification and identification determines the soil classification through soil particle analysis, plasticity index determination, bearing capacity test and other means. The three types of geological condition information are combined to form a geological condition classification label, which serves as the index basis for subsequent data processing.

[0028] The statistical processing of historical geological disaster case classification is based on geological condition classification labels. Past geological disaster events such as landslides, mudslides, and collapses are grouped according to the geological conditions of the locations where they occurred. Each geological condition group contains corresponding information such as the time of disaster occurrence, meteorological conditions, and disaster scale. Correlation coefficient calculation is a statistical method for measuring the strength of the linear relationship between two variables. The calculation process involves calculating the covariance between each meteorological factor and the number of disaster occurrences, and then dividing it by the product of the standard deviations of the two variables. The correlation coefficient ranges from -1 to +1, with positive values ​​indicating positive correlation and negative values ​​indicating negative correlation. The larger the absolute value, the stronger the correlation. The correlation coefficients between different meteorological factors and disaster occurrence under various geological conditions are arranged in a matrix form to form a factor correlation coefficient matrix.

[0029] Normalization is an important step in data preprocessing. The factor correlation coefficient matrix is ​​normalized according to the geological condition dimension. Normalization is calculated by subtracting the minimum correlation coefficient under each geological condition from each correlation coefficient, and then dividing it by the difference between the maximum and minimum correlation coefficients. The normalized values ​​range between zero and one, eliminating the impact of numerical magnitude differences under different geological conditions. The gradient descent algorithm is a commonly used optimization algorithm in machine learning. It determines the gradient direction by calculating the partial derivative of the loss function with respect to the weight parameters. The weight parameters are updated in the opposite direction of the gradient. The update step size is controlled by the learning rate. The iterative process continues until the loss function converges or the preset number of iterations is reached. After iterative optimization, the optimized weight parameters are obtained.

[0030] The sensitivity weighting process is based on optimizing weight parameters. It quantifies the sensitivity weights of four meteorological factors—rainfall intensity, soil saturation, groundwater level change, and slope runoff coefficient—under different geological conditions. The principle of weighting is that meteorological factors with stronger correlations receive higher weights. Weights are calculated by multiplying the optimized weight parameters with the correlation coefficients. A matrix structure is constructed with geological conditions as rows and meteorological factors as columns. Each matrix element represents the sensitivity weight of a specific meteorological factor under specific geological conditions. This matrix structure facilitates subsequent matrix operations and data indexing. The constructed matrix is ​​called the initial sensitivity weight matrix.

[0031] Expert knowledge verification, based on professional experience and theoretical knowledge in the field of geological engineering, verifies and adjusts the rationality of the weight distribution in the initial sensitivity weight matrix. The verification process includes checking whether the weight distribution is consistent with the mechanism of geological hazard occurrence. For example, the weight of groundwater level change should be higher in high-permeability soil conditions, while the weight of rainfall intensity should be higher in low-permeability soil conditions. Verification processing uses expert experience to correct unreasonable weight values. Correction methods include increasing, decreasing, or redistributing weight values. After the verification is completed, the geological hazard sensitivity weight matrix is ​​obtained.

[0032] In a specific embodiment, the process of executing step S103 may specifically include the following steps: The geological disaster meteorological data set is segmented into sequences according to time windows to construct meteorological time series data with continuous time steps; According to the geological hazard sensitivity weight matrix, each meteorological factor in the meteorological time series data is weighted and multiplied element by element to obtain a weighted meteorological characteristic sequence. The geological condition information of slope, lithology and soil type is vector-encoded through the embedding layer to obtain the geological condition embedding vector; Based on the geological condition embedding vector, the forget gate, input gate, and output gate of the geological constraint long short-term memory network are modulated by geophysical constraints. The weighted meteorological feature sequence is input into the modulated network for temporal feature learning to obtain the geological constraint feature vector. The geological constraint feature vector is processed through a fully connected layer and a softmax activation function for probability distribution calculation to obtain a geological hazard risk probability vector.

[0033] Specifically, the time window sequence segmentation process segments the geological disaster meteorological dataset according to the preset time window length. The time window refers to the length of the continuous time series used to train the neural network. The segmentation process is implemented through the sliding window technology. The sliding window starts from the starting time point of the dataset and moves forward according to a fixed step size. Each window position extracts continuous data of corresponding length to form an independent time series sample. These samples constitute the meteorological time series data of continuous time steps. The structure of the time series data is a three-dimensional tensor. The first dimension represents the number of samples, the second dimension represents the time step, and the third dimension represents the number of meteorological factor features. The element-by-element weighted product processing adjusts the weights of various meteorological factors in the meteorological time series data according to the geological hazard sensitivity weight matrix. The weighted product refers to the operation of multiplying the weight values ​​in the weight matrix with the corresponding meteorological factor values ​​one by one. The multiplication process needs to consider the matching relationship of geological conditions. Each time series sample corresponds to specific geological conditions. The weight value of the corresponding geological condition row in the weight matrix is ​​multiplied element-by-element with the meteorological factor value in the sample. The result of the product operation is the adjusted meteorological factor value, which reflects the importance of different meteorological factors under specific geological conditions. The weighted data is reorganized according to the original time series structure to form a weighted meteorological feature sequence.

[0034] The embedding layer vector encoding process converts discrete geological condition information into continuous numerical vectors. The embedding layer is a common feature representation method in deep learning, mapping discrete identifiers into dense vectors through a trainable weight matrix. The slope information encoding process first discretizes the continuous slope value into a graded discretization according to a preset interval. Each slope grade corresponds to a unique integer identifier, which is converted into a fixed-dimensional vector representation through the embedding layer. The lithology information encoding process converts lithology categories such as granite, shale, and sandstone into a one-hot encoding form. A one-hot encoding is a binary vector with only one position set to one and all other positions set to zero. The one-hot encoding is further converted into a dense vector through the embedding layer. The soil type information encoding process assigns numerical identifiers to categories such as clay, sand, and silt. The identifiers are also converted into vector representations through the embedding layer. The vector representations of the three geological conditions are combined through a concatenation operation to form a geological condition embedding vector.

[0035] The geophysical constraint modulation process adjusts the gating mechanism of the long-short-term memory network (LSTM) based on the geological condition embedding vector. The LSTM network contains three gating structures: a forget gate, an input gate, and an output gate. Each gate structure controls the flow of information using a sigmoid activation function. The forget gate determines which information is discarded from the cell state. Its calculation involves a linear combination of the current input, the previous hidden state, and the geological condition embedding vector. This linear combination is achieved by matrix multiplication of the weight matrix and the input vector. The product is processed by a sigmoid function to obtain a gating value between zero and one. The input gate determines which new information is stored in the cell state. The calculation process of the input gate is similar to that of the forget gate, but uses different weight matrices and bias parameters. The output gate controls which parts of the cell state are output. The calculation of the output gate also involves the geological condition embedding vector. The modulation process modifies the weight parameters of each gate using the geological condition embedding vector, enabling the network to adjust its information processing strategy based on different geological conditions. The weighted meteorological feature sequence is input into the modulated network for forward propagation. The network learns the temporal pattern through multi-level feature transformation to obtain the geological constraint feature vector. The probability distribution calculation process converts the geological constraint feature vector into a risk probability output. The fully connected layer is a linear transformation layer in the neural network. The linear output is calculated by matrix multiplication of the weight matrix and the input vector. The linear output plus the bias term forms the output of the fully connected layer. The softmax activation function is a commonly used activation function in multi-classification problems. The softmax function converts any real vector into a probability distribution. The conversion process first calculates the exponential function for each element of the input vector and then normalizes all the exponential values. Normalization is achieved by dividing each exponential value by the sum of all exponential values. The output of the softmax function is a probability vector. Each element in the vector represents the probability value of the corresponding category, and the sum of all probability values ​​is equal to one. The geological constraint feature vector is linearly transformed by the fully connected layer. The transformed vector is calculated by the softmax activation function to obtain the geological hazard risk probability vector. The probability vector contains the probability values ​​of multiple risk levels.

[0036] In a specific embodiment, the process of performing vector encoding processing on the geological condition information of slope, lithology, and soil type through the embedding layer may specifically include the following steps: The slope values ​​are discretized in a hierarchical manner, and the continuous slope values ​​are classified and coded according to the preset slope intervals to obtain the slope classification code; Perform numerical mapping of rock types according to lithologic types, convert granite, shale, and sandstone lithologic categories into one-hot encoding to obtain lithologic encoding vectors; Soil types are coded and labeled based on soil classification standards, with clay, sand, and silt being numerically identified to obtain soil type codes. The slope classification code, lithology code vector, and soil type code are combined and processed to construct a geological condition comprehensive code sequence to obtain a geological condition input vector; The geological condition input vector is linearly transformed and nonlinearly activated through a trainable weight matrix to obtain a geological condition embedding vector.

[0037] Specifically, the process of generating the geological condition embedding vector involves five core data encoding and transformation steps, each of which involves specific data processing algorithms and numerical calculations. The slope classification discretization process converts continuous slope measurements into classification identifiers. Classification discretization involves dividing the continuous numerical space into several non-overlapping intervals, each corresponding to a discrete category. Slope values ​​are obtained using topographic survey equipment, with the measurement results representing the angle between the ground surface and the horizontal plane. Pre-set slope intervals are divided according to geological engineering standards, including gentle slope, moderate slope, steep slope, and dangerous slope. The classification encoding process is implemented using a conditional judgment algorithm that compares the slope value with the boundary values ​​of each interval and assigns the slope value to the corresponding interval. Each interval is assigned a unique integer code, and the resulting encoding forms the slope classification code. The lithology numerical mapping process converts qualitative rock type information into quantitative numerical representations. Lithology types are determined through geological exploration and rock sample analysis, including rock properties such as mineralogy, structural characteristics, and formation environment. One-hot encoding is a commonly used encoding method for categorical variables in machine learning. The encoding process creates a binary vector for each lithologic category. The vector length equals the total number of lithologic categories. Positions corresponding to each category are set to one, and all other positions are set to zero. Granite, shale, and sandstone are encoded using different binary vector patterns. The encoding process is achieved by creating a zero vector and setting the corresponding positions to one. This encoding maintains the independence of lithologic categories and avoids artificial numerical correlations. The lithologic encoding vector contains the classification information for each lithologic type.

[0038] Soil labeling and coding converts soil types into numerical identifiers based on soil classification standards. Soil classification standards are determined based on physical properties such as particle composition, plasticity, and bearing capacity. Clay, sand, and silt, for example, possess distinct engineering geological characteristics. Labeling and coding involves assigning a unique integer identifier to each soil type. Identifiers are assigned according to predetermined coding rules. The coding process is implemented through a lookup table or mapping dictionary, mapping soil type names to corresponding numerical identifiers. The resulting encoding forms a soil type code, preserving the discreteness and uniqueness of the soil types.

[0039] The splicing and combination process combines the coding results of the three geological conditions into a unified data structure. The splicing operation involves concatenating multiple vectors or values ​​in a specific order into a longer vector. The slope classification code, as a scalar value, is first converted to vector form. The lithology code vector maintains its original vector structure, and the soil type code is also converted to vector form. The three vectors are then spliced ​​together end-to-end. The length of the spliced ​​vector is equal to the sum of the lengths of the component vectors. The splicing order follows a fixed sequence of slope, lithology, and soil. The splicing result constructs a comprehensive geological condition coding sequence, which contains all the classification information of the geological conditions and forms the geological condition input vector.

[0040] Linear transformation and nonlinear activation processing convert the geological condition input vector into a dense feature representation. Linear transformation is achieved through the matrix product of a trainable weight matrix and the input vector. The dimension of the weight matrix is ​​determined by the length of the input vector and the target embedding dimension. The matrix product calculation process includes row-by-row and column-by-column multiplication and addition operations. The result of the operation plus the bias vector forms the output of the linear transformation. Nonlinear activation processing nonlinearly maps the linear transformation result through activation functions such as ReLU, tanh, and sigmoid. The choice of activation function affects the numerical distribution and expressive power of the embedded vector. The activated vector becomes the geological condition embedding vector, which has fixed dimension and continuous numerical features and can represent the complex correlations between geological conditions.

[0041] For example: The geological exploration results of a monitoring point in a mountainous area show that the slope is 35 degrees, the rock type is granite, and the soil type is weathered soil. The slope classification discretization processing classifies 35 degrees into the medium slope category, and the corresponding classification code is an integer value of two. In the lithology numerical mapping processing, the unique hot coding length of the three lithology categories of granite, shale, and sandstone is three. The coding vector corresponding to granite is a vector with the first position being one and the remaining positions being zero, while shale and sandstone correspond to vectors with different positions being one. The soil labeling coding processing classifies weathered soil into a specific category, and the corresponding numerical identifier is an integer value of one. The splicing combination processing converts the slope code into a single-element vector, and connects it end-to-end with the lithology coding vector and the soil coding vector. The spliced ​​geological condition input vector contains all the coding information of the slope, lithology, and soil type. During the linear transformation process, the input vector is multiplied by the weight matrix. The parameters of the weight matrix are learned through the machine learning training process. The transformation result is processed by the ReLU activation function, which sets negative values ​​to zero and keeps positive values ​​unchanged. The processed vector becomes the embedded vector of the geological conditions of the monitoring point. The value of the embedded vector reflects the geological characteristics of the medium-slope granite weathered soil slope.

[0042] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Based on the Bishop arc sliding method, the slope stability is calculated in real time. The current safety factor value is calculated according to the effective cohesion, internal friction angle, soil weight, sliding mass thickness, and slope angle parameters to obtain the real-time safety factor. Perform differential calculation on the real-time safety factor within a continuous time period, calculate the ratio of the difference of the safety factor between adjacent time points to the time interval, and obtain the safety factor change rate; Compare the safety factor change rate with the preset drop threshold value, and when the change rate is continuously negative and the absolute value exceeds the threshold, the sensitivity adjustment mechanism is triggered to obtain a threshold adjustment trigger signal; According to the threshold adjustment trigger signal, the baseline warning threshold is adjusted upward by a percentage, and the three-level warning thresholds of danger, warning, and caution are adjusted upward simultaneously to obtain the adjusted warning threshold; The adjusted warning threshold is corrected for moisture impact based on the soil moisture correction coefficient. The ratio of current saturation to field water holding capacity saturation is calculated as the correction factor to obtain the adaptive warning threshold parameter.

[0043] Specifically, the Bishop arc sliding method's real-time calculation process is based on slope stability theory. This method assumes the sliding surface is an arc and calculates the slope's safety factor using the principle of moment balance. The safety factor calculation requires key parameters such as effective cohesion, internal friction angle, soil mass, sliding mass thickness, and slope angle. Effective cohesion is measured through triaxial compression or direct shear tests and reflects the bond strength between soil particles. The internal friction angle is determined through soil mechanics testing and represents the magnitude of internal frictional resistance. Soil mass is obtained through field sampling and laboratory density testing. The sliding mass thickness is determined through geological exploration and borehole data, and the slope angle is obtained through topographic surveys. The Bishop arc sliding method divides the sliding mass into several vertical segments. The mass, bottom inclination, and effective stress of each segment are calculated separately. The safety factor is calculated as the ratio of the anti-sliding torque to the sliding torque. The calculation process employs an iterative approach. Because the safety factor itself appears in the denominator of the calculation formula, the iterative calculation begins with an assumed initial safety factor value and gradually modifies it until convergence. Upon convergence, the real-time safety factor is obtained. Differential processing performs numerical analysis on the real-time safety factor over a continuous time period. Differential operations are a fundamental method for calculating the rate of change of a function in numerical analysis. The rate of change of the safety factor is calculated by subtracting the safety factor values ​​at adjacent time points to obtain the difference. This difference reflects the magnitude of the change in the safety factor per unit time. The time interval is typically set to one hour. The rate of change is calculated by dividing the difference by the time interval. The result of the division indicates the hourly rate of change of the safety factor. A positive value indicates an increase in the safety factor, while a negative value indicates a decrease. The absolute value of the rate of change reflects the severity of the change. The rates of change at multiple consecutive time points form a rate of change sequence, which reflects the dynamic evolution of slope stability.

[0044] The comparison and judgment process compares the rate of change of the safety factor with the preset drop threshold. The comparison process is implemented through a conditional judgment algorithm. The drop threshold is determined based on geological engineering experience and historical disaster statistics. The setting of the threshold needs to balance the sensitivity and accuracy of the warning. A threshold that is too small will lead to frequent false alarms, and a threshold that is too large will reduce the timeliness of the warning. The comparison and judgment includes a logical AND operation of two conditions. The first condition checks whether the rate of change is negative. A negative value indicates that the safety factor is on a downward trend. The second condition checks whether the absolute value of the rate of change exceeds the preset threshold. Exceeding the threshold indicates that the rate of decline is too fast. When both conditions are met at the same time, the sensitivity adjustment mechanism is triggered. The triggering process is implemented through Boolean logic operations. When the conditions are met, a true value is output, and when the conditions are not met, a false value is output. The true value signal serves as the threshold adjustment trigger signal.

[0045] The percentage increase process adjusts the baseline warning threshold numerically based on the threshold adjustment trigger signal. A percentage increase refers to increasing the threshold value by a fixed percentage. The baseline warning thresholds include three levels: danger, warning, and caution. Each level corresponds to a different risk probability range, with the danger level threshold being the lowest and the caution level the highest. The percentage increase process is performed simultaneously on all three thresholds. The increase ratio is determined by the severity of the safety factor reduction. The increase calculation is calculated by multiplying the original threshold by the increase coefficient. The increase coefficient is usually set to a value greater than one. The increased threshold increases the warning sensitivity, enabling the warning system to issue an alert earlier. The adjusted warning threshold reflects the current deterioration of slope stability.

[0046] The moisture impact correction process further adjusts the adjusted warning threshold based on the soil moisture correction factor. The soil moisture correction factor reflects the impact of soil moisture conditions on slope stability. The current saturation is obtained through real-time monitoring of the soil moisture sensor. The field capacity saturation is the maximum moisture content that the soil can maintain under gravity. This parameter is determined through soil physical testing. The ratio calculation is to divide the current saturation by the field capacity saturation to obtain a correction factor. A correction factor greater than one indicates that the soil moisture content exceeds the normal level, while a correction factor less than one indicates that the soil is relatively dry. The correction process divides the adjusted warning threshold by the correction factor to obtain the corrected value. The division operation takes into account the impact of soil moisture on slope stability. The threshold is lowered at high moisture content and increased at low moisture content. The corrected value becomes the adaptive warning threshold parameter.

[0047] In a specific embodiment, the process of executing step S105 may specifically include the following steps: Based on the real-time safety factor, the risk probability of the physical model is calculated using normal distribution. The difference between the safety factor and the benchmark value is divided by the standard deviation as the input parameter to obtain the risk probability of the physical model. The geological parameters were randomly disturbed by Monte Carlo sampling method, and the variance of the prediction results was calculated by random sampling of effective cohesion and internal friction angle parameters multiple times to obtain the uncertainty quantification probability. Perform weighted linear combination processing on geological hazard risk probability vector, physical model risk probability and uncertainty quantification probability, perform probability fusion calculation according to preset weight coefficients, and obtain fused risk probability; The fusion risk probability is graded and judged according to the adaptive warning threshold parameters. The fusion probability is compared with the four-level thresholds of danger, warning, attention, and safety step by step to obtain the warning level judgment result. Based on the warning level determination results, the warning information is formatted and outputted to generate a warning report containing risk level, probability value, and duration forecast, and obtain the graded geological disaster warning results.

[0048] Specifically, the normal distribution calculation process quantifies the risk probability of the physical model based on the real-time safety factor. The normal distribution is a continuous probability distribution commonly used in probability theory, characterized by a bell-shaped curve. Its distribution parameters include the mean and standard deviation. The baseline value is typically set as the critical value of the safety factor. When the safety factor equals the baseline value, the slope is in a critical stability state. The difference between the safety factor and the baseline value reflects the degree to which the current stability deviates from the critical state. A positive difference indicates good stability, while a negative difference indicates insufficient stability. The standard deviation parameter is determined through statistical analysis of historical safety factor data and reflects the degree of dispersion in the safety factor. A larger standard deviation indicates greater fluctuations in the safety factor. The calculation process of dividing the difference by the standard deviation is called normalization. The normalized value serves as the input parameter for the normal distribution. The normal distribution calculation uses a table lookup or numerical integration method to obtain the corresponding cumulative probability. This probability indicates the likelihood that the safety factor will be lower than the current value. The probability value reflects the risk level predicted by the physical model. The Monte Carlo sampling method randomly perturbs the geological parameters. The Monte Carlo method is a numerical calculation technique based on random sampling, which obtains numerical solutions through a large number of random experiments. Effective cohesion and internal friction angle are key geological parameters that affect slope stability. These parameters are subject to uncertainty in engineering practice. Sources of uncertainty include measurement errors, spatial variability, and differences in test conditions. Random perturbation processing is achieved by adding random noise to the original values ​​of the parameters. Random noise usually follows a normal distribution or a uniform distribution, and the amplitude of the noise is determined by the uncertainty level of the parameters. Multiple random sampling processes generate a large number of parameter combinations, each of which corresponds to a safety factor calculation. The results of multiple calculations form a probability distribution of the safety factor. The variance of the predicted results is obtained by calculating the sum of the squares of the differences between all results and the mean and dividing it by the number of samples. The variance reflects the degree of uncertainty in the predicted results. The larger the variance, the lower the reliability of the prediction. The uncertainty quantification probability is obtained by comparing the variance with a preset threshold.

[0049] Weighted linear combination processing fuses probabilistic information from three different sources. Weighted linear combination is a fundamental method for multi-source information fusion. The geological hazard risk probability vector is derived from machine learning predictions using a geologically constrained long-short-term memory network, the physical model risk probability is derived from mechanical calculations using the Bishop arc sliding method, and the uncertainty quantification probability is derived from statistical analysis using Monte Carlo sampling. These three types of probabilistic information have different physical meanings and reliability levels. Preset weight coefficients reflect the importance and credibility of each information source. The determination of these weight coefficients requires comprehensive consideration of factors such as the accuracy, stability, and applicability of each method. The sum of these weight coefficients is equal to one to ensure the normalized nature of the probabilities. Probabilistic fusion calculations are performed by summing the products of each probability and its corresponding weight coefficient. The resulting fused probability integrates the data-driven capabilities of machine learning, the mechanistic explanatory power of the physical model, and the uncertainty quantification capabilities of statistical methods. The fused risk probability reflects the comprehensive judgment of information from multiple sources.

[0050] The hierarchical judgment process classifies the fused risk probability according to the adaptive warning threshold parameters. Hierarchical judgment is a key step in risk assessment. The adaptive warning threshold parameters include four levels of thresholds: danger, warning, caution, and safety. Each threshold corresponds to a different risk probability range. The threshold setting is determined based on historical disaster statistics and engineering experience. The step-by-step comparison process starts with the highest risk level. First, it determines whether the fused probability exceeds the danger level threshold. If it exceeds, it is determined to be a danger level. If it does not exceed, it continues to determine whether it exceeds the warning level threshold, and so on until the specific risk level is determined. The comparison process is implemented through a conditional judgment algorithm. The algorithm adopts an if-else logic structure. Each condition corresponds to a risk level judgment. The judgment result represents the warning level in numerical or textual form. The warning level judgment result lays the foundation for the subsequent generation of warning information.

[0051] Formatted output processing generates standardized warning information based on the warning level determination results. Formatted output is a standardized process for information expression. Warning information includes key elements such as risk level, probability value, and duration forecast. The risk level is directly derived from the warning level determination results, the probability value is derived from the calculation results of the integrated risk probability, and the duration forecast is based on meteorological forecast data and statistical analysis of historical disaster durations. The warning report generation process includes steps such as information collection, formatting, and content verification. Information collection summarizes the results of each processing step, formatting organizes the information content according to a predetermined template structure, and content verification verifies the integrity and consistency of the information. The graded geological hazard warning results are output as structured data or documents. The output content includes a clear identification of the warning level, a quantitative value of the risk probability, and a specific range of the warning's effective duration. The output format facilitates rapid understanding and response by emergency management departments.

[0052] The above describes the geological disaster meteorological risk warning method based on machine learning in the embodiment of the present application. The following describes the geological disaster meteorological risk warning system based on machine learning in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the geological disaster meteorological risk early warning system based on machine learning includes: The acquisition module is used to collect and process the rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient of the target area in real time through multi-source sensors to obtain a geological disaster meteorological data set; An allocation module is used to perform sensitivity weight allocation processing on each meteorological factor in the geological disaster meteorological data set according to geological condition classification to obtain a geological disaster sensitivity weight matrix; A weighting module is used to perform weighted fusion processing on the geological hazard sensitivity weight matrix and the meteorological time series data, and to extract nonlinear features of the fused data through a geological constraint long short-term memory network to obtain a geological hazard risk probability vector; An adjustment module is used to dynamically adjust the warning threshold according to the safety factor change rate, and when the safety factor change rate continues to decrease, the threshold sensitivity is increased to obtain an adaptive warning threshold parameter; The fusion module is used to perform Bayesian fusion processing on the geological disaster risk probability vector and the adaptive warning threshold parameter to obtain a graded geological disaster warning result.

[0053] above Figure 2 The geological disaster meteorological risk warning system based on machine learning in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The geological disaster meteorological risk warning equipment based on machine learning in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0054] Reference Figure 3 In an embodiment of the present invention, a geological disaster meteorological risk warning device based on machine learning is also provided. The geological disaster meteorological risk warning device based on machine learning can be a server, and its internal structure can be as follows: Figure 3As shown. The geological disaster meteorological risk warning device based on machine learning includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the geological disaster meteorological risk warning device based on machine learning includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the geological disaster meteorological risk warning device based on machine learning is used to store the corresponding data in this embodiment. The network interface of the geological disaster meteorological risk warning device based on machine learning is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0055] Those skilled in the art will understand that Figure 3 The structure shown in is only a block diagram of part of the structure related to the solution of the present invention, and does not constitute a limitation on the machine learning-based geological disaster meteorological risk warning device to which the solution of the present invention is applied.

[0056] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the steps of the geological disaster meteorological risk warning method based on machine learning.

[0057] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0058] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a machine learning-based geological disaster and meteorological risk warning device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A geological disaster meteorological risk early warning method based on machine learning, characterized in that: The method comprises: Multi-source sensors are used to collect and process the rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient in the target area in real time to obtain a geological disaster meteorological data set. Perform sensitivity weighting processing on each meteorological factor in the geological disaster meteorological data set according to the geological condition classification to obtain a geological disaster sensitivity weight matrix; The geological hazard sensitivity weight matrix is ​​weightedly fused with meteorological time series data, and nonlinear features are extracted from the fused data using a geological constraint long short-term memory network to obtain a geological hazard risk probability vector; Dynamically adjust the warning threshold according to the safety factor change rate, and increase the threshold sensitivity when the safety factor change rate continues to decrease to obtain an adaptive warning threshold parameter; The geological disaster risk probability vector is subjected to Bayesian fusion processing with the adaptive warning threshold parameter to obtain a graded geological disaster warning result.

2. The geological disaster meteorological risk early warning method based on machine learning according to claim 1 is characterized in that: The multi-source sensors are used to collect and process the rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient of the target area in real time to obtain a geological disaster meteorological data set, including: The rainfall intensity data is collected and processed hourly by the automatic weather station, and the collected data is compared and screened with the set effective rainfall intensity range to obtain the effective rainfall intensity sequence; The soil saturation data is monitored and processed in real time through soil moisture sensors. The collected original saturation values ​​are bounded according to the preset physical reasonable range to obtain a standard soil saturation sequence. The groundwater level change data is continuously collected and processed by the groundwater level monitoring instrument, and the water level difference between adjacent time points is calculated and divided by the time interval to obtain the groundwater level change rate series; The runoff data is collected and processed based on the slope flow monitoring equipment, and the runoff coefficient value is calculated by dividing the measured runoff volume by the corresponding rainfall to obtain the slope runoff coefficient sequence; The effective rainfall intensity sequence, standard soil saturation sequence, groundwater level change rate sequence, and slope runoff coefficient sequence are subjected to time synchronization alignment processing to obtain a geological disaster meteorological dataset.

3. The geological disaster meteorological risk early warning method based on machine learning according to claim 1 is characterized in that: The sensitivity weight distribution process is performed on each meteorological factor in the geological disaster meteorological data set according to the geological condition classification to obtain a geological disaster sensitivity weight matrix, including: Classify and identify the slope, lithology, and soil type based on the geological exploration data of the target area to obtain geological condition classification labels; Based on the geological condition classification labels, historical geological disaster cases are classified and statistically processed, and the correlation coefficients between different meteorological factors and disaster occurrence under different geological conditions are calculated to obtain a factor correlation coefficient matrix; Normalizing the factor correlation coefficient matrix according to the geological condition dimension, iteratively optimizing the weight parameters using a gradient descent algorithm to obtain optimized weight parameters; Based on the optimized weight parameters, sensitivity weight distribution processing is performed on rainfall intensity, soil saturation, groundwater level change, and slope runoff coefficient, and a matrix structure with geological conditions as rows and meteorological factors as columns is constructed to obtain an initial sensitivity weight matrix; The initial sensitivity weight matrix is ​​subjected to expert knowledge verification processing based on geological engineering experience to obtain a geological hazard sensitivity weight matrix.

4. The geological disaster meteorological risk early warning method based on machine learning according to claim 1 is characterized in that: The geological hazard sensitivity weight matrix is ​​subjected to weighted fusion processing with meteorological time series data, and nonlinear feature extraction is performed on the fused data through a geological constraint long short-term memory network to obtain a geological hazard risk probability vector, including: Sequentially segmenting the geological disaster meteorological data set according to time windows to construct meteorological time series data with continuous time steps; Performing element-by-element weighted product processing on each meteorological factor in the meteorological time series data according to the geological hazard sensitivity weight matrix to obtain a weighted meteorological feature sequence; The geological condition information of slope, lithology and soil type is vector-encoded through the embedding layer to obtain the geological condition embedding vector; Based on the geological condition embedding vector, the forget gate, input gate, and output gate of the geological constraint long short-term memory network are subjected to geophysical constraint modulation processing, and the weighted meteorological feature sequence is input into the modulated network for time series feature learning to obtain a geological constraint feature vector; The geological constraint feature vector is processed by probability distribution calculation through a fully connected layer and a softmax activation function to obtain a geological disaster risk probability vector.

5. The geological disaster meteorological risk early warning method based on machine learning according to claim 4 is characterized in that: The process of performing vector coding processing on the geological condition information of slope, lithology and soil type through the embedding layer to obtain the geological condition embedding vector includes: The slope values ​​are discretized in a hierarchical manner, and the continuous slope values ​​are classified and coded according to the preset slope intervals to obtain the slope classification code; Perform numerical mapping of rock types according to lithologic types, convert granite, shale, and sandstone lithologic categories into one-hot encoding to obtain lithologic encoding vectors; Soil types are labeled and coded based on soil classification standards, with clay, sand, and silt categories numerically labeled to obtain soil type codes; The slope classification code, lithology code vector, and soil type code are spliced ​​and combined to construct a geological condition comprehensive code sequence to obtain a geological condition input vector; The geological condition input vector is subjected to linear transformation and nonlinear activation processing through a trainable weight matrix to obtain a geological condition embedding vector.

6. The geological disaster meteorological risk early warning method based on machine learning according to claim 1 is characterized in that: The method dynamically adjusts the warning threshold according to the safety factor change rate, increases the threshold sensitivity when the safety factor change rate continues to decrease, and obtains the adaptive warning threshold parameter, including: Based on the Bishop arc sliding method, the slope stability is calculated in real time. The current safety factor value is calculated according to the effective cohesion, internal friction angle, soil weight, sliding mass thickness, and slope angle parameters to obtain the real-time safety factor. Perform differential calculation on the real-time safety factor within a continuous time period, calculate the ratio of the difference of the safety factor between adjacent time points to the time interval, and obtain the safety factor change rate; Compare the safety factor change rate with a preset drop threshold value, and when the change rate is continuously negative and the absolute value exceeds the threshold, trigger the sensitivity adjustment mechanism to obtain a threshold adjustment trigger signal; According to the threshold adjustment trigger signal, the baseline warning threshold is adjusted upward by a percentage, and the three-level warning thresholds of danger, warning, and caution are adjusted upward simultaneously to obtain the adjusted warning threshold; The adjusted warning threshold is corrected for moisture impact based on the soil moisture correction coefficient, and the ratio of the current saturation to the field water holding capacity saturation is calculated as a correction factor to obtain an adaptive warning threshold parameter.

7. The geological disaster meteorological risk early warning method based on machine learning according to claim 6 is characterized in that: The Bayesian fusion processing of the geological disaster risk probability vector and the adaptive warning threshold parameter to obtain a graded geological disaster warning result includes: Performing normal distribution calculation processing on the physical model risk probability based on the real-time safety factor, dividing the difference between the safety factor and the reference value by the standard deviation as an input parameter to obtain the physical model risk probability; The geological parameters were randomly disturbed by Monte Carlo sampling method, and the variance of the prediction results was calculated by random sampling of effective cohesion and internal friction angle parameters multiple times to obtain the uncertainty quantification probability. Performing weighted linear combination processing on the geological hazard risk probability vector, the physical model risk probability, and the uncertainty quantification probability, and performing probability fusion calculation according to preset weight coefficients to obtain a fused risk probability; Performing hierarchical judgment processing on the fusion risk probability according to the adaptive warning threshold parameters, comparing the fusion probability with the four-level thresholds of danger, warning, attention, and safety step by step to obtain the warning level judgment result; Based on the warning level determination result, the warning information is formatted and outputted to generate a warning report including risk level, probability value, and duration prediction to obtain a graded geological disaster warning result.

8. A geological disaster meteorological risk early warning system based on machine learning, characterized by: For implementing the geological disaster meteorological risk early warning method based on machine learning according to any one of claims 1 to 7, the geological disaster meteorological risk early warning system based on machine learning comprises: The acquisition module is used to collect and process the rainfall intensity, soil saturation, groundwater level changes, and slope runoff coefficient of the target area in real time through multi-source sensors to obtain a geological disaster meteorological data set; An allocation module is used to perform sensitivity weight allocation processing on each meteorological factor in the geological disaster meteorological data set according to geological condition classification to obtain a geological disaster sensitivity weight matrix; A weighting module is used to perform weighted fusion processing on the geological hazard sensitivity weight matrix and the meteorological time series data, and to extract nonlinear features of the fused data through a geological constraint long short-term memory network to obtain a geological hazard risk probability vector; An adjustment module is used to dynamically adjust the warning threshold according to the safety factor change rate, and when the safety factor change rate continues to decrease, the threshold sensitivity is increased to obtain an adaptive warning threshold parameter; The fusion module is used to perform Bayesian fusion processing on the geological disaster risk probability vector and the adaptive warning threshold parameter to obtain a graded geological disaster warning result.

9. A geological disaster meteorological risk early warning device based on machine learning, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, it implements the geological disaster meteorological risk warning method based on machine learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the geological disaster meteorological risk early warning method based on machine learning as described in any one of claims 1 to 7.

Images