Rainfall monitoring data-driven small watershed geological disaster intelligent early warning method

Abstract

The present application relates to the field of geological disaster prevention, and particularly relates to a rainfall monitoring data driven small watershed geological disaster intelligent early warning method, comprising: obtaining historical rainfall data and geological disaster historical record data of a target small watershed for preprocessing; matching geological disaster events to corresponding rainfall events to obtain a rainfall event set in which geological disasters occur and serving as positive samples, and a rainfall event set in which no geological disasters occur and serving as negative samples; using the positive samples and the negative samples to construct a data set for learning, training and testing of a machine learning model; constructing an array of small watershed geological disaster intelligent early warning machine learning models and training, verifying and testing the array using the data set; constructing a target small watershed geological disaster occurrence probability calculation model and an uncertainty analysis model under the influence of a rainfall event, calculating the target small watershed geological disaster occurrence probability and uncertainty under the rainfall event, and performing early warning; and the present application improves the reliability and accuracy of small watershed geological disaster early warning results.

Description

Technical Field

[0001] This invention relates to the field of geological disaster prevention and control, specifically to an intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data. Background Technology

[0002] In recent years, frequent heavy rainfall and severe convective weather have led to a surge in geological disasters in small watersheds. In particular, short-duration heavy rainfall creates extremely strong hydrodynamic conditions in the valleys of small watersheds, easily triggering shallow landslides, debris flows, and other geological disasters. While these disasters are generally small in scale, their wide distribution, rapid speed, suddenness, and powerful impact make them extremely serious sudden geological disasters. For example, a localized torrential rainstorm of July 4th in a certain area in 2023 resulted in various geological disasters causing 15 deaths, 4 missing persons, and direct economic losses of 227,844,740 yuan.

[0003] The most effective and economical solution to avoid losses from such small watershed geological disasters is to build reliable early warning and forecasting capabilities. Currently, early warning models based on fixed rainfall thresholds or driven by hydrogeophysical models have significant shortcomings. First, the response characteristics of geological disasters differ under different rainfall patterns, making it unrealistic to completely determine the regional response characteristics of geological disasters under different rainfall patterns. Using fixed rainfall thresholds for early warning of small watershed geological disasters carries an extremely high risk of false alarms and missed alarms. Second, early warning driven by hydrogeophysical models requires detailed geological and topographical conditions and accurate soil and rock parameters, making it more suitable for early warning of individual geological disasters. Obtaining geological and topographical conditions and soil and rock parameters for the entire small watershed is extremely costly and impractical.

[0004] Since the main forms of geological disasters in small watersheds are shallow soil landslides and debris flows, rainfall is the decisive triggering factor. Therefore, constructing a predictive model of rainfall patterns, rainfall amounts, and other rainfall characteristics in relation to the occurrence of geological disasters in small watersheds is crucial for accurate early warning of such disasters. To this end, a rainfall monitoring data-driven intelligent early warning method for geological disasters in small watersheds is proposed. Summary of the Invention

[0005] The present invention aims to provide a method for intelligent early warning of geological disasters in small watersheds driven by rainfall monitoring data, so as to overcome the problems of unreliable early warning results using fixed rainfall thresholds and the high cost or difficulty in obtaining and timely updating geological and topographical conditions and soil parameters driven by hydrogeological and physical models, and realize accurate and intelligent early warning of geological disasters in small watersheds based entirely on rainfall monitoring data.

[0006] This scheme's rainfall monitoring data-driven intelligent early warning method for geological disasters in small watersheds includes:

[0007] Step S1: Obtain historical rainfall data and geological disaster historical data of the target watershed or other watersheds with similar geology, geomorphology and meteorological climate, and perform preprocessing, dividing the historical rainfall data into m rainfall events and the geological disaster historical data into n geological disaster events;

[0008] Step S2: Match the geological disaster event described in step S1 to the corresponding rainfall event to obtain the m event in which the geological disaster occurred. Y A set of rainfall events Y and m where no geological disasters occurred. N A set of N rainfall events;

[0009] Step S3, the m that caused the geological disaster mentioned in step S2 Y The set of rainfall events Y is used as a positive sample, and the m events that did not experience geological disasters as described in step S2 are included. N A set of N rainfall events is used as negative samples, and a dataset DS is constructed using positive and negative samples for training, validating, and testing machine learning models for intelligent early warning of geological disasters in small watersheds.

[0010] Step S4: Use the dataset DS described in step S3 to train, validate, and test the intelligent early warning machine learning model for geological disasters in small watersheds, and obtain the intelligent early warning machine learning model and array for geological disasters in small watersheds with the best prediction performance;

[0011] Step S5, for any rainfall event R composed of rainfall monitoring data... f Using the output of the machine learning model array for intelligent early warning of geological disasters in small watersheds described in step S4 as the data source, a probability calculation model and uncertainty analysis model for geological disasters in the target small watershed under the influence of rainfall events are constructed to calculate the rainfall event R. f Probability and uncertainty of geological disaster occurrence in the target small watershed;

[0012] Step S6, for any rainfall event R composed of rainfall monitoring data... f Based on the rainfall event R obtained in step S5 f The probability and uncertainty of geological disasters in the target small watershed, and the impact of rainfall event R. f Early warning of geological disasters in the target small watershed.

[0013] The beneficial effects of this plan are:

[0014] This solution targets geological disaster monitoring in small watersheds. Based on rainfall monitoring data in small watersheds, it introduces a geological disaster triggering time factor to accurately characterize the impact of rainfall. Furthermore, by using an array of intelligent early warning machine learning models for geological disasters in small watersheds, it overcomes the shortcomings of using a single machine learning model that cannot integrate and consider the interaction of different geological disaster triggering time factors under rainfall events, thereby improving the accuracy of early warning results.

[0015] Furthermore, in step S1, historical rainfall data is segmented into m rainfall events by the continuous no-rainfall time interval IT. The continuous no-rainfall time interval IT is determined based on the local climate conditions, rainfall patterns, and whether the time is the rainy or dry season in the target watershed, and is expressed as:

[0016]

[0017] Furthermore, in step S2, matching geological disaster events to corresponding rainfall events includes:

[0018] Step S2.1: For any rainfall event R among the m rainfall events in step S1, represent it as a time series in hours to obtain the start time R of that rainfall event. s and end time R e ;

[0019] S2.2, For any one of the n geological disaster events GD in step S1, obtain the occurrence time R of geological disaster event GD. o ;

[0020] S2.3, if R is satisfied s <R o ≤R e The geological disaster event GD described in step S2.2 is matched with the rainfall event R described in step S2.1.

[0021] The beneficial effect is that matching geological disaster events with rainfall events based on time information can accurately quantify and correlate geological disasters with rainfall, thereby improving the accuracy of subsequent early warnings.

[0022] Furthermore, in step S3, the step of constructing the dataset DS includes:

[0023] Step S3.1: Obtain the start time R of any rainfall event R from the rainfall event set Y obtained in step S2. s The time R during this rainfall event when the geological disaster occurred. o The duration of rainfall (DY) and the cumulative rainfall (CY) are calculated using the following formulas:

[0024]

[0025] Step S3.2, divide the rainfall duration DY into the preceding rainfall duration DY. a Duration of triggering rainfall DY o Duration of previous rainfall DY a Duration of triggering rainfall DY o The calculation formula is:

[0026] DY o =θ, DY a =DY-DY o Where θ represents the geological disaster triggering time factor, with the unit being h, and θ = 1, 2, ..., 24;

[0027] Step 3.3: Divide the rainfall duration CY into the cumulative rainfall amount CY of the preceding period. a Cumulative rainfall CY triggered by rainfall o Previous rainfall cumulative rainfall CY a Cumulative rainfall CY triggered by rainfall o The calculation formula is:

[0028]

[0029] Step 3.4, for any geological disaster triggering time factor θ in any rainfall event R within the rainfall event set Y in step S2, using [DY] a ,CY a ,CY o Using ] as the feature vector and “[1]” as the label, construct the positive sample dataset DS-Y of dataset DS;

[0030] Step 3.5: For any rainfall event R in the set N of rainfall events where no geological disaster has occurred, based on steps S3.1 to S3.3, calculate the rainfall duration DY, cumulative rainfall CY, and previous rainfall duration DY. a Triggering rainfall duration DY o Previous rainfall cumulative rainfall CY a Cumulative rainfall CY triggered by rainfall o The calculation formula is:

[0031]

[0032] Step 3.6, for any geological disaster triggering time factor θ in the set N of rainfall events where no geological disaster has occurred, using [DY a ,CY a ,CY o Using ] as the feature vector and “[0]” as the label, construct the negative sample dataset DS-N of dataset DS;

[0033] Step 3.7: Merge the positive sample dataset DS-Y and the negative sample dataset DS-N to obtain the dataset DS.

[0034] The beneficial effect is that by dividing a rainfall event into the effects of preceding rainfall and the effects of triggered rainfall, the effects of preceding rainfall reflect the influence of preceding rainfall on the parameters of the geological body itself during the evolution of geological disasters, while the effects of triggered rainfall reflect the influence of triggered rainfall on the internal shear deformation or instability of the geological body, thereby achieving an accurate characterization of the impact of rainfall on the geological disaster evolution process.

[0035] Furthermore, step 4, in constructing the intelligent early warning machine learning model array for geological disasters in small watersheds, includes the following steps:

[0036] Step 4.1: For any geological disaster triggering time factor θ, randomly divide the positive sample dataset DS-Y from step S3.4 into a positive sample set TV for the training-validation dataset according to an 80%:20% ratio. θ -Y and the positive sample set TD of the test dataset θ -Y;

[0037] Step 4.2, for any geological disaster triggering time factor θ, according to the positive sample set TV θ The negative sample dataset DS-N is randomly divided into a training and validation negative sample set TV with an equal amount of sample data for Y. θ -N, and divide the remaining data into the negative sample set TD of the test dataset. θ -N;

[0038] Step 4.3, select the correct sample set TV. θ -Y and negative sample set TV θ -N merging forms the dataset TV for training and validating machine learning models under the geological disaster triggering time factor θ. θ And further divide the dataset TV according to an 80%:20% ratio. θ Divided into training subset TV θ -T and verification subset TV θ -V is used for training and validating machine learning models;

[0039] Step 4.4, convert the positive sample set TD θ -Y and negative sample set TD θ -N is merged to form the test dataset TD for testing machine learning models under the geological disaster triggering time factor θ. θ Used for testing machine learning models;

[0040] Step 4.5: For any geological disaster triggering time factor θ, construct a machine learning model M. θ As a machine learning model for intelligent early warning of geological disasters in small watersheds, it utilizes a training subset TV θ -T and verification subset TV θ -V for machine learning model M θ Perform training and validation to optimize the machine learning model M. θ The parameters in the model are used to obtain the trained machine learning model M. θ and using the test dataset TD θ For machine learning model M θ Continue testing and evaluation;

[0041] Step S4.6: For any geological hazard triggering time factor θ, repeat steps S4.1 to S4.5 100 times to construct a 24×100 small watershed geological hazard intelligent early warning machine learning model array. Among them, θ=1,2,...,24, i=1,2,...,100.

[0042] The beneficial effect is that by constructing and optimizing machine learning model arrays, the accuracy of geological disaster prediction under rainfall events can be improved.

[0043] Furthermore, in step S5, any rainfall monitoring data constituting a rainfall event R f Its rainfall monitoring data includes real-time rainfall monitoring data or forecast rainfall monitoring data for the target small watershed.

[0044] The beneficial effects are: the rainfall monitoring data is more comprehensive and richer, improving the accuracy and timeliness of subsequent early warnings.

[0045] Furthermore, in step 5, a probability calculation model and uncertainty analysis model for geological disasters in the target small watershed under the influence of rainfall events are constructed to calculate the rainfall event R. f The probability and uncertainty of geological disasters in the target small watershed are determined, including the following steps:

[0046] Step S5.1, based on rainfall monitoring data, construct rainfall event R. f For any geological disaster triggering time factor θ, calculate the rainfall event R based on step S3.5. f Rainfall duration DY, cumulative rainfall CY, and previous rainfall duration DY a Triggering rainfall duration DY o Previous rainfall cumulative rainfall CY a Cumulative rainfall CY triggered by rainfall o ;

[0047] Step S5.2, using the geological disaster triggering time factor θ obtained in step S5.1, [DY] a ,CY a ,CY o [ ] is the input feature vector, based on a machine learning model array for intelligent early warning of geological disasters in small watersheds. Geological hazard prediction was performed, resulting in a 24×100 geological hazard prediction matrix P driven by rainfall monitoring data. Each element p in matrix P... θi The value is 1 or 0, where θ = 1, 2, ..., 24, i = 1, 2, ..., 100, 1 indicates that a geological disaster has occurred, and 0 indicates that no geological disaster has occurred;

[0048] Step S5.3: For any geological hazard triggering time factor θ, calculate the geological hazard prediction result P under that triggering time factor. θ and the uncertainty of the result U θ The calculation formula is:

[0049]

[0050] Step S5.4: During rainfall event R f The formulas for calculating the probability p and uncertainty u of geological disasters in the target small watershed are as follows:

[0051]

[0052] The beneficial effect is that by calculating the geological disaster prediction results and uncertainty of the target small watershed under any geological disaster triggering time factor, the interaction of different geological disaster triggering time factors under rainfall events can be taken into account, thereby improving the accuracy of early warning.

[0053] Furthermore, in step S6, the rainfall event R... f The steps for issuing early warnings of geological disasters in the target small watershed include:

[0054] Given rainfall event R f Geological disaster early warning information for small watersheds, including warning level WL, probability of geological disaster occurrence p, and uncertainty u, and calculation of rainfall event R. f The formula for calculating the geological disaster early warning level WL in a small watershed is:

[0055]

[0056] The beneficial effects are: the early warning includes both qualitative and quantitative information and its uncertainty, which helps decision-makers and information recipients to accurately grasp the geological disaster risks of the target small watershed under the influence of rainfall events. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of an embodiment of the intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data of the present invention. Detailed Implementation

[0058] The following detailed description illustrates the specific implementation method:

[0059] Example

[0060] Intelligent early warning methods for geological disasters in small watersheds driven by rainfall monitoring data, such as Figure 1 As shown, it includes:

[0061] Step S1: Obtain historical rainfall data and geological disaster historical data for the target watershed, or other watersheds with similar geology, geomorphology, and meteorological climate. Preprocess the historical rainfall data and geological disaster historical data, including data denoising, interpolation, and information completion. Each preprocessing operation uses existing technologies and will not be described in detail here.

[0062] Historical rainfall data is divided into m rainfall events, and historical geological disaster data is divided into n geological disaster events. Specifically, the historical rainfall data is divided into m rainfall events by the continuous no-rainfall time interval IT. The continuous no-rainfall time interval IT is determined based on the local climate conditions, rainfall patterns, and whether the current time is the rainy or dry season in the target watershed. The dry and rainy seasons are determined according to relevant industry specifications or local standards, which will not be elaborated here, but are expressed as follows:

[0063]

[0064] Step S2: Match the geological disaster event described in step S1 to the corresponding rainfall event to obtain the m event in which the geological disaster occurred. Y A set of rainfall events Y and m where no geological disasters occurred. N Given a set N rainfall events, matching geological disaster events to corresponding rainfall events involves the following steps:

[0065] Step S2.1: For any rainfall event R among the m rainfall events in step S1, represent it as a time series in hours to obtain the start time R of that rainfall event. s and end time R e ;

[0066] S2.2, For any one of the n geological disaster events GD in step S1, obtain the occurrence time R of geological disaster event GD. o ;

[0067] S2.3, if R is satisfied s <Ro ≤R e The geological disaster event GD described in step S2.2 is matched with the rainfall event R described in step S2.1.

[0068] Step S3, the m that caused the geological disaster mentioned in step S2 Y The set of rainfall events Y is used as a positive sample, and the m events that did not experience geological disasters as described in step S2 are included. N A set of N rainfall events is used as negative samples. A dataset DS is constructed using both positive and negative samples for training, validating, and testing a machine learning model for intelligent early warning of geological disasters in small watersheds. The steps for constructing the dataset DS include:

[0069] Step S3.1: Obtain the start time R of any rainfall event R from the rainfall event set Y obtained in step S2. s The time R during this rainfall event when the geological disaster occurred. o The duration of rainfall (DY) and the cumulative rainfall (CY) are calculated as follows: The unit for duration (DY) is hours (h), and the unit for cumulative rainfall (CY) is mm (mm).

[0070]

[0071] Step S3.2, divide the rainfall duration DY into the preceding rainfall duration DY. a Duration of triggering rainfall DY o Duration of previous rainfall DY a Duration of triggering rainfall DY o The calculation formula is:

[0072] DY o =θ, DY a =DY-DY o Where θ represents the geological disaster triggering time factor, with the unit being h, and θ = 1, 2, ..., 24;

[0073] Step 3.3: Divide the rainfall duration CY into the cumulative rainfall amount CY of the preceding period. a Cumulative rainfall CY triggered by rainfall o Previous rainfall cumulative rainfall CY a Cumulative rainfall CY triggered by rainfall o The calculation formula is:

[0074]

[0075] Step 3.4, for any geological disaster triggering time factor θ in any rainfall event R within the rainfall event set Y in step S2, using [DY] a,CY a ,CY o ] is the feature vector, and “[1]” is the label to indicate that a geological disaster has occurred. The positive sample dataset DS-Y of dataset DS is constructed.

[0076] Step 3.5: For any rainfall event R in the set N of rainfall events where no geological disaster has occurred, based on steps S3.1 to S3.3, calculate the rainfall duration DY, cumulative rainfall CY, and previous rainfall duration DY. a Triggering rainfall duration DY o Previous rainfall cumulative rainfall CY a Cumulative rainfall CY triggered by rainfall o The calculation formula is:

[0077]

[0078] Step 3.6, for any geological disaster triggering time factor θ in the set N of rainfall events where no geological disaster has occurred, using [DY a ,CY a ,CY o [] is the feature vector, and "[0]" is the label to indicate that no geological disaster has occurred. The negative sample dataset DS-N of dataset DS is constructed.

[0079] Step 3.7: Merge the positive sample dataset DS-Y and the negative sample dataset DS-N to obtain the dataset DS.

[0080] Step S4: Using the dataset DS described in Step S3, train, validate, and test the intelligent early warning machine learning model for small watershed geological disasters to obtain the intelligent early warning machine learning model for small watershed geological disasters with the best prediction performance and the array of intelligent early warning machine learning models for small watershed geological disasters. The steps for constructing the intelligent early warning machine learning model array for small watershed geological disasters are as follows:

[0081] Step 4.1: For any geological disaster triggering time factor θ, randomly divide the positive sample dataset DS-Y from step S3.4 into a positive sample set TV for the training-validation dataset according to an 80%:20% ratio. θ -Y and the positive sample set TD of the test dataset θ -Y;

[0082] Step 4.2, for any geological disaster triggering time factor θ, according to the positive sample set TV θ The negative sample dataset DS-N is randomly divided into a training and validation negative sample set TV with an equal amount of sample data for Y. θ -N, and divide the remaining data into the negative sample set TD of the test dataset. θ -N;

[0083] Step 4.3, select the correct sample set TV. θ -Y and negative sample set TV θ -N merging forms the dataset TV for training and validating machine learning models under the geological disaster triggering time factor θ. θ And further divide the dataset TV according to an 80%:20% ratio. θ Divided into training subset TV θ -T and verification subset TV θ -V is used for training and validating machine learning models;

[0084] Step 4.4, convert the positive sample set TD θ -Y and negative sample set TD θ -N is merged to form the test dataset TD for testing machine learning models under the geological disaster triggering time factor θ. θ Used for testing machine learning models;

[0085] Step 4.5: For any geological disaster triggering time factor θ, construct a machine learning model M. θ As a machine learning model for intelligent early warning of geological disasters in small watersheds, it utilizes a training subset TV θ -T and verification subset TV θ -V for machine learning model M θ Perform training and validation to optimize the machine learning model M. θ The parameters in the model are used to obtain the trained machine learning model M. θ and using the test dataset TD θ For machine learning model M θ Continue testing and evaluation;

[0086] Step S4.6: For any geological hazard triggering time factor θ, repeat steps S4.1 to S4.5 100 times to construct a 24×100 small watershed geological hazard intelligent early warning machine learning model array. Among them, θ=1,2,...,24, i=1,2,...,100.

[0087] Step S5, for any rainfall event R composed of rainfall monitoring data... f Using the output of the machine learning model array for intelligent early warning of geological disasters in small watersheds described in step S4 as the data source, a probability calculation model and uncertainty analysis model for geological disasters in the target small watershed under the influence of rainfall events are constructed to calculate the rainfall event R. f The probability and uncertainty of geological disasters occurring in the target small watershed are determined by the following steps:

[0088] Step S5.1, based on rainfall monitoring data, construct rainfall event R. f For any geological disaster triggering time factor θ, calculate the rainfall event R based on step S3.5. f Rainfall duration DY, cumulative rainfall CY, and previous rainfall duration DY a Triggering rainfall duration DY o Previous rainfall cumulative rainfall CY a Cumulative rainfall CY triggered by rainfall o ;

[0089] Step S5.2, using the geological disaster triggering time factor θ obtained in step S5.1, [DY] a ,CY a ,CY o [ ] is the input feature vector, based on a machine learning model array for intelligent early warning of geological disasters in small watersheds. Geological hazard prediction was performed, resulting in a 24×100 geological hazard prediction matrix P driven by rainfall monitoring data. Each element p in matrix P... θi The value is 1 or 0, where θ = 1, 2, ..., 24, i = 1, 2, ..., 100, 1 indicates that a geological disaster has occurred, and 0 indicates that no geological disaster has occurred;

[0090] Step S5.3: For any geological hazard triggering time factor θ, calculate the geological hazard prediction result P under that triggering time factor. θ and the uncertainty of the result U θ The calculation formula is:

[0091]

[0092] Step S5.4: During rainfall event R f The formulas for calculating the probability p and uncertainty u of geological disasters in the target small watershed are as follows:

[0093]

[0094] Step S6, for any rainfall event R composed of rainfall monitoring data... f Based on the rainfall event R obtained in step S5 f The probability and uncertainty of geological disasters in the target small watershed, and the impact of rainfall event R. f Early warning of geological disasters in the target small watershed is provided, specifically: a rainfall event R is given. f Geological disaster early warning information for small watersheds, including warning level WL, probability of geological disaster occurrence p, and uncertainty u, and calculation of rainfall event R. f The formula for calculating the geological disaster early warning level WL in a small watershed is:

[0095]

[0096] This embodiment acquires historical rainfall data and geological disaster historical data from other small watersheds with similar geological geomorphology and meteorological climate to the target small watershed. It expands the dataset DS by adding historical data on geological disaster occurrence patterns similar to those in the target small watershed under rainfall-induced conditions. Based on the rainfall monitoring data of the small watershed, a geological disaster triggering time factor is introduced to accurately characterize the impact of rainfall. Furthermore, by using an array of intelligent early warning machine learning models for small watershed geological disasters, it overcomes the limitation of using a single machine learning model that cannot integrate and consider the interaction of different geological disaster triggering time factors under rainfall events, thus improving the accuracy of early warning results. Simultaneously, it provides early warning information that includes both qualitative and quantitative information and its uncertainties, which helps decision-makers and information recipients accurately grasp the geological disaster risk of the target small watershed under the influence of rainfall events. Compared to the currently used fixed rainfall threshold early warning and small watershed geological disaster early warning methods driven by hydrogeological and physical models, the method in this embodiment does not require the pre-design of a precise early warning model or the acquisition of precise data, resulting in lower costs, more comprehensive consideration of various factors, no false alarms or missed alarms, higher reliability, and more complete early warning information, thereby improving the predictive performance of the intelligent early warning machine learning model and array for small watershed geological disasters.

[0097] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics is not elaborated upon here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for intelligent early warning of geological disasters in small watersheds driven by rainfall monitoring data, characterized in that, include: Step S1: Obtain historical rainfall data and geological disaster historical data of the target watershed or other watersheds with similar geology, geomorphology and meteorological climate, and perform preprocessing, dividing the historical rainfall data into m rainfall events and the geological disaster historical data into n geological disaster events; Step S2: Match the geological disaster events described in Step S1 to corresponding rainfall events to obtain the geological disaster occurrence events. A set of rainfall events Y and no geological disasters occurred. A set of N rainfall events; Step S3, the geological disaster mentioned in step S2 Using a set Y of rainfall events as positive samples, the events that did not experience geological disasters as described in step S2... A set of N rainfall events is used as negative samples, and a dataset DS is constructed using positive and negative samples for training, validating, and testing machine learning models for intelligent early warning of geological disasters in small watersheds. Step S4 involves training, validating, and testing the intelligent early warning machine learning model for small watershed geological disasters using the dataset DS described in step S3, to obtain the intelligent early warning machine learning model and array for small watershed geological disasters with the best prediction performance. The construction of the intelligent early warning machine learning model array for small watershed geological disasters includes the following steps: Step 4.1, for any geological disaster triggering time factor The positive sample dataset DS-Y from step S3.4 is randomly divided into a training-validation dataset positive sample set according to an 80%:20% ratio. and the positive sample set of the test dataset ; Step 4.2, for any geological disaster triggering time factor According to the positive sample set The negative sample dataset DS-N is randomly divided into training and validation negative sample sets with an equal amount of sample data. The remaining data was then divided into a negative sample set for the test dataset. ; Step 4.3, convert the sample set to the correct value. With negative sample set Combined, forming the geological disaster triggering time factor Datasets for training and validating machine learning models And further divide the dataset according to an 80%:20% ratio. Divided into training subsets and verification subset Used for training and validating machine learning models; Step 4.4, convert the positive sample set With negative sample set Combined, forming the geological disaster triggering time factor Test dataset for testing machine learning models Used for testing machine learning models; Step 4.5, for any geological disaster triggering time factor Build a machine learning model As a machine learning model for intelligent early warning of geological disasters in small watersheds, it utilizes a training subset and verification subset For machine learning models Perform training and validation to optimize the machine learning model. The parameters in the model are used to obtain the trained machine learning model. and using the test dataset For machine learning models Continue testing and evaluation; Step S4.6, for any geological disaster triggering time factor Repeat steps S4.1 to S4.5 100 times to achieve the construction. Small watershed geological disaster intelligent early warning machine learning model array ,in, , ; Step S5, for any rainfall monitoring data constituting a rainfall event Using the output of the machine learning model array for intelligent early warning of geological disasters in small watersheds described in step S4 as the data source, a probability calculation model and uncertainty analysis model for geological disasters in the target small watershed under the influence of rainfall events are constructed to calculate the rainfall events. Probability and uncertainty of geological disaster occurrence in the target small watershed; Step S6, for any rainfall monitoring data constituting a rainfall event Based on the rainfall events obtained in step S5 The probability and uncertainty of geological disasters in the target small watershed, in relation to rainfall events. Early warning of geological disasters in the target small watershed.

2. The method for intelligent early warning of geological disasters in small watersheds driven by rainfall monitoring data according to claim 1, characterized in that: In step S1, historical rainfall data is segmented into m rainfall events by the continuous no-rainfall time interval IT. The continuous no-rainfall time interval IT is determined based on the local climate conditions, rainfall patterns, and whether the time is the rainy or dry season of the target watershed, and is expressed as follows: 。 3. The intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data according to claim 1, characterized in that: In step S2, matching geological disaster events to corresponding rainfall events includes: Step S2.1: For any rainfall event R among the m rainfall events in step S1, represent it as a time series in hours to obtain the start time of the rainfall event. and end time ; S2.2, For any one of the n geological disaster events GD in step S1, obtain the occurrence time of geological disaster event GD. ; S2.3, if satisfied The geological disaster event GD described in step S2.2 is matched with the rainfall event R described in step S2.

1.

4. The intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data according to claim 3, characterized in that: In step S3, the steps for constructing the dataset DS include: Step S3.1: Obtain the start time of any rainfall event R from the rainfall event set Y obtained in step S2. Time of geological disasters during this rainfall event The duration of rainfall (DY) and the cumulative rainfall (CY) are calculated using the following formulas: ， ； Step S3.2, divide the rainfall duration DY into the preceding rainfall duration. Duration of triggering rainfall Duration of previous rainfall Duration of triggering rainfall The calculation formula is: , ,in, This represents the time factor for geological disaster triggering, expressed in hours (h). ; Step 3.3: Divide the rainfall duration CY into the cumulative rainfall amount of the preceding period. Cumulative rainfall triggered by rainfall Previous rainfall cumulative rainfall Cumulative rainfall triggered by rainfall The calculation formula is: ， ； Step 3.4, for any rainfall event R in the rainfall event set Y from step S2, for any geological disaster triggering time factor... ,by Using the feature vector and "[1]" as the label, construct the positive sample dataset DS-Y of dataset DS; Step 3.5: For any rainfall event R in the set N of rainfall events where no geological disaster has occurred, based on steps S3.1 to S3.3, calculate the rainfall duration DY, cumulative rainfall CY, and duration of previous rainfall. Duration of triggering rainfall Previous rainfall cumulative rainfall Cumulative rainfall triggered by rainfall The calculation formula is: ； Step 3.6: For any geological disaster triggering time factor in the set N of rainfall events where no geological disaster has occurred... ,by Using the feature vector and "[0]" as the label, construct the negative sample dataset DS-N of dataset DS; Step 3.7: Merge the positive sample dataset DS-Y and the negative sample dataset DS-N to obtain the dataset DS.

5. The intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data according to claim 4, characterized in that: In step 4, ...

6. The intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data according to claim 5, characterized in that: In step S5, any rainfall monitoring data constitutes a rainfall event. Its rainfall monitoring data includes real-time rainfall monitoring data or forecast rainfall monitoring data for the target small watershed.

7. The intelligent early warning method for geological disasters in small watersheds driven by rainfall monitoring data according to claim 5, characterized in that: In step 5, a probability calculation model and an uncertainty analysis model for geological disasters in the target small watershed under the influence of rainfall events are constructed to calculate the rainfall events. The probability and uncertainty of geological disasters in the target small watershed are determined, including the following steps: Step S5.1, rainfall events based on rainfall monitoring data For any geological disaster triggering time factor Based on step S3.5, calculate the rainfall event. Rainfall duration DY, cumulative rainfall CY, and duration of previous rainfall Duration of triggering rainfall Previous rainfall cumulative rainfall Cumulative rainfall triggered by rainfall ; Step S5.2, using the geological disaster triggering time factor obtained in step S5.

1. Down Using the input feature vector, a machine learning model array for intelligent early warning of geological disasters in small watersheds is used. To conduct geological disaster prediction, we obtain data driven by rainfall monitoring. Geological disaster prediction result matrix ,matrix Elements The value is 1 or 0, where, , 1 indicates that a geological disaster has occurred, and 0 indicates that no geological disaster has occurred; Step S5.3, for any geological disaster triggering time factor Calculate the geological disaster prediction results under this triggering time factor. and result uncertainty The calculation formula is: ； Step S5.4: During a rainfall event The formulas for calculating the probability p and uncertainty u of geological disasters in the target small watershed are as follows: 。 8. The method for intelligent early warning of geological disasters in small watersheds driven by rainfall monitoring data according to claim 1, characterized in that: In step S6, the rainfall event The steps for issuing early warnings of geological disasters in the target small watershed include: Give rainfall events Geological disaster early warning information for small watersheds, including early warning level WL, probability of geological disaster occurrence p, and uncertainty u, and calculation of rainfall events. The formula for calculating the geological disaster early warning level WL in a small watershed is: 。

Images