Mountain torrent debris flow early warning system based on multi-source data comprehensive analysis

Abstract

The invention discloses a mountain torrent debris flow early warning method and system based on multi-source data comprehensive analysis. And the monitoring system carries out multi-source monitoring on rainfall, material source and mud level change characteristics of the ditch domain. The method is based on double-threshold rainfall intensity parameters, a quantitative analysis model of channel mud level and flow, and a blocking and collapsing mechanism of object source deformation. The threshold value of each element is determined according to the disaster history, a disaster outbreak probability formula is established, and key factor weights are screened through a random forest algorithm. And data preprocessing is carried out through cross validation and dynamic adjustment. And substituting monitoring data into the analysis model, and comprehensively judging the early warning level of the mountain torrent debris flow. And issuing an early warning signal of a corresponding level to the object in the threatened area through the early warning system. After operation, an incremental random forest method is adopted to update the model through online learning, data redisk is carried out, and model threshold correction and parameter inversion analysis are carried out. According to the pre-warning system, a multi-index combined action comprehensive monitoring and pre-warning evaluation system for quantitative analysis is established.

Description

Technical Field

[0001] The present invention relates to the technical field of flash flood and debris flow warning, and particularly to a flash flood and debris flow warning method and system based on comprehensive analysis of multi-source data. Background Art

[0002] At present, the meteorological risk warning of flash flood and debris flow disasters in China still remains at the regional, trend-based, and warning stages. Due to the differences in the geological landforms, source material compositions and distributions, water catchment conditions, threatened objects, and disaster formation mechanisms and evolution processes of individual flash flood and debris flow gullies, the current regional warning models are difficult to accurately judge the time, space, and disaster formation risks of flash flood and debris flow gullies.

[0003] In the western Sichuan Plateau, the Panxi region, and the mountainous areas around the Sichuan Basin, etc., the mountains are high and the slopes are steep, with criss-crossing ravines. The terrain elevation differences in small mountainous watersheds are huge, and the rainfall varies greatly in the vertical direction, with an obvious three-dimensional climate. Rainfall is the main inducement of flash flood and debris flow disasters. However, due to objective conditions, it is difficult to monitor rainfall in the upper reaches of flash flood and debris flow gullies, and there are many existing blind spots. The main manifestation is that it is difficult to obtain rainfall information in uninhabited areas in the upper reaches, the "space-air-ground" collaborative monitoring ability is insufficient, and the accuracy of the surface rainfall currently mastered is not enough, and it cannot provide sufficient decision-making support for active disaster prevention and avoidance in threatened areas. Among the current disaster prevention measures, the "precursor signals" and "evolution processes" of flash flood and debris flow disasters such as sudden changes or exceedances of water level and flow rate, and the activation of source materials are not monitored in place, and the modern means of capturing disaster signs are insufficient and the integration degree is not high. Summary of the Invention

[0004] The purpose of the present invention is to provide a flash flood and debris flow warning method and system based on comprehensive analysis of multi-source data to overcome the technical problems existing in the prior art such as insufficient monitoring and inability to provide sufficient decision-making support for active disaster prevention and avoidance in threatened areas.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] On the one hand, the present invention proposes a flash flood and debris flow early warning method based on comprehensive analysis of multi-source data. The key lies in: utilizing deformation monitoring of major material sources, rainfall monitoring in the watershed, and mud level monitoring in the flow area channel to obtain monitoring data on influencing factors and characteristics related to the outbreak of flash floods and debris flows, and utilizing historical and surrounding similar geological disasters to construct a "geological-meteorological-hydrological" fully coupled flash flood and debris flow early warning model. Through data processing, noise and outliers are eliminated, and the mutual verification of multi-source data is used to improve the accuracy of early warning and reduce false alarms and missed reports. During the operation of the system, the model is corrected and tested using monitoring data and disaster occurrence situations to continuously improve the accuracy of early warning. Transform the prediction of flash flood and debris flows from "experience-single threshold" to "data-model-decision-making" intelligent closed loop. The core lies in the deep integration of dynamic thresholds, multi-model collaboration, and full-chain perception. The following steps are included:

[0007] Step 1: Set up monitoring equipment for target objects and build a ditch monitoring system.

[0008] Through investigation of the specific monitoring objects, the development conditions of debris flow, disaster history, basic characteristics of debris flow, distribution range of threatened objects, emergency evacuation routes and shelters are found out. For the existence of major landslide sources in the debris flow, the source reserves and the degree of ditch blockage that may be caused after the start are found out. The investigation divides the debris flow formation area, flow area, accumulation area and danger area. Arrange rainfall monitoring equipment in the debris flow formation area, displacement and deformation monitoring equipment at potential major landslide sources, and mud level monitoring equipment in the formation area ditch. The specific number depends on the length of the ditch, the basin area, and the development of the source. Use multi-location and multi-type monitoring equipment to obtain monitoring data on the formation and movement of debris flows.

[0009] Step 2: Establish a threshold model framework for specific monitoring targets based on the rainfall intensity, debris flow rate, outflow volume and solid material source distribution of the region and the disaster history of the monitored object.

[0010] At present, the monitoring and early warning of debris flow is only based on mud level or rainfall, which is prone to distortion of individual monitoring data and is difficult to accurately reflect the relationship between rainfall, material source deformation, channel mud level change and debris flow outbreak. This system determines the possibility of debris flow by building a model that includes rainfall, material source and mud level, and taking into account the impact of various factors on debris flow.

[0011] According to the disaster history and the disasters in similar channels in the adjacent areas, the thresholds of each factor are preliminarily determined, and the probability formula of disaster outbreak is calculated by using the disaster history data. This realizes the comprehensive monitoring, early warning and evaluation system from the traditional single indicator threshold to quantitative analysis of multiple indicators.

[0012] To solve the problems of rainfall concentration in debris flow, which induces debris flow, and the time difference when the debris flow moves to the threatened area. In view of the characteristics of debris flow, according to the calculation method of debris flow characteristic values in relevant specifications, a model among mud level height, rainfall and debris flow discharge is constructed, and the thresholds of rainfall and mud level in the monitored target gully are inversely calculated according to the possible hazards caused by different discharges in the threatened area.

[0013] Solid material sources have a significant impact on the discharge and specific gravity of debris flow. In many debris flow occurrences, major source deformation and instability lead to blockage and breach, resulting in an increase in the debris flow blockage coefficient Dc, enhanced debris flow discharge and impact force, and greater hazards. To solve this problem, a blockage and breach correction module is added to the early warning model. The displacement and deformation monitoring of major landslide and avalanche sources is used to timely control the stability of the sources and their participation in debris flow activities. When the source deforms to instability and blocks the gully, the debris flow blockage and breach correction module is activated to raise the debris flow early warning level.

[0014] Step 3: Calibrate and verify the early warning threshold using historical data

[0015] Using historical data, the random forest algorithm is adopted to screen the weights of key factors. According to the determined weights, substitute them into the disaster occurrence probability model, cross-validate the historical data, and determine the false negative rate and false positive rate of the model. If the requirements are not met, adjust the weights of key factors until the prediction accuracy is achieved. Due to the continuous change of the geological conditions of debris flow gullies and the influence of limited early data, dynamic adjustment is carried out after operation to continuously correct the threshold parameters of the model.

[0016] Step 4: Conduct multi-source data fusion based on the monitored data, judge the data credibility of the monitoring data, and eliminate abnormal monitoring data.

[0017] To reduce the influence of local deformation of monitoring equipment, flow changes in individual gullies, and abnormal rainfall in individual gully segments within the basin on monitoring and early warning, it is necessary to judge the data credibility of the data in the real-time monitoring dataset. Through multi-source data fusion analysis, preprocess the data by spatio-temporal filling, noise filtering, and anomaly detection, and screen out the monitoring credible dataset; use multi-source data cross-validation to establish an intelligent decision-making mechanism for conflict data and verify the reliability of the data.

[0018] Step 5: According to the monitoring data and meteorological forecasts, through early warning model analysis, conduct early warning grading and issue early warning signals to prompt corresponding emergency response measures.

[0019] This part of the work is completed by the early warning subsystem, which conducts reliable early warning based on the risk coefficient of mountain flood and debris flow disasters. The early warning system determines the corresponding early warning level and issues early warning signals according to the indicators of occurrence probability, rainfall intensity, mud level change, and source stability in the decision-making system. To ensure rapid early warning, Jetson Xavier is deployed at the monitoring station to calculate the P value in real time, reduce dependence on the cloud, and upload the early warning records to the blockchain (Hyperledger Fabric) to ensure the immutability of the decision-making process.

[0020] Step 6: Conduct post-evaluation and optimization of the model through the monitoring data recorded after the monitoring and early warning system is enabled and the disaster occurrence situation.

[0021] In view of the lack of existing disaster data for the current monitoring of mountain floods and debris flows, there may be deviations in the initial early warning model and threshold setting. Use the multi-source monitoring data and disaster occurrence situation during operation to review the data, and use the incremental random forest method for model iteration through online learning to invert some parameters of the original model, continuously upgrade the system, and improve the pertinence of the system and the accuracy of early warning.

[0022] Step 7: Use multiple mountain flood and debris flow monitoring systems within the region to achieve cross-basin collaborative early warning and build a swarm intelligence early warning.

[0023] Due to the regional nature of mountain flood and debris flow disasters, although the disaster occurrence conditions in different channels are different, the rainfall area is relatively wide, and the surrounding channels have certain reference value. By sharing the monitoring data (such as rainfall station networks) in adjacent regions through blockchain technology, cross-basin collaborative early warning can be achieved. Currently, this technology and working idea have been widely applied in the dual control of geological disaster points and areas.

[0024] The specific work content and technical methods of each step are as follows

[0025] Step 1: Set up monitoring devices for the target object and build a gully monitoring system.

[0026] The multi-source monitoring data of mountain flood and debris flow includes rainfall monitoring data within the mountain flood and debris flow gully area, mud level monitoring data within the mountain flood and debris flow channel, and displacement monitoring data of major landslide and avalanche sources beside the mountain flood and debris flow channel.

[0027] Set up monitoring devices for the target object and build a gully monitoring system. Obtain the monitoring data of the formation and movement of debris flows using multi-location and multi-type monitoring devices. The specific process and technical requirements are as follows:

[0028] 1. Deployment of monitoring devices

[0029] Table 1 Monitoring device table

[0030]

[0031] 2. Technical Requirements

[0032] Data Transmission: Adopt LoRaWAN or 5G narrowband IoT to ensure real-time performance (delay ≤ 1 minute).

[0033] Disaster Resistance Design: The equipment has an IP68 protection level, and the power supply system has dual redundancy of solar energy + super capacitor.

[0034] Time and Space Synchronization: All devices are built-in with GPS modules, the time stamp error ≤ 1 second, and the spatial coordinates are unified to the WGS2000 coordinate system.

[0035] Step 2: According to the rainfall intensity, debris flow discharge, scouring volume and solid material source distribution of the debris flow in the disaster history of the region and the monitoring object, establish a threshold model framework for specific monitoring objectives.

[0036] According to the rainfall intensity, debris flow discharge, scouring volume and solid material source distribution of the debris flow in the disaster history of the region and the monitoring object, establish a threshold model framework for specific monitoring objectives. The specific content is as follows:

[0037] 1. Model Input

[0038] Rainfall Parameters: Short-duration rainfall intensity (Imax), effective cumulative rainfall E (attenuation coefficient k = 0.1 according to local stratum conditions).

[0039] Short-duration rainfall intensity (Imax), the maximum rainfall intensity in a short duration, unit: millimeter per hour (mm / h).

[0040] Effective cumulative rainfall E, the cumulative rainfall in a certain period before the disaster (unit: millimeter, mm), usually considering the time decay effect. The effective cumulative time is generally considered as 24 hours. It reflects the cumulative impact of previous rainfall on soil moisture content. The closer the soil is to saturation, the lower the instantaneous rainfall intensity required to trigger debris flow.

[0041]

[0042] According to the disaster history and adjacent engineering experience, when constructing a bivariate threshold model, draw an Imax-E scatter plot and fit a power-law curve I = a·E b , obtaining the relationship between E and I values. The rainfall threshold Ic for debris flow outbreak in this gully can be obtained. Using the model of I = a·E b to obtain the threshold of rainfall intensity, and considering the synergistic effect of instantaneous rainfall intensity and previous rainfall at the same time, it is more in line with the debris flow triggering mechanism than the univariate threshold.

[0043] Source Deformation Parameters: Displacement rate Vd (mm / d), acceleration ad (mm / h2), volume change rate ΔV (m3 / d).

[0044] Sludge level parameters: sludge level value H (m), sludge level rising rate Vm (m / h), number of sludge level mutations Nsurge.

[0045] 2. Model structure

[0046] 1) Core formula for disaster outbreak probability:

[0047]

[0048] Where:

[0049] Id = f(Ecum) rainfall intensity;

[0050] Hd, Vm are the monitored sludge level height and the source displacement rate;

[0051] Ic, Hc, Vmc are the rainfall intensity, sludge level height and source displacement rate thresholds;

[0052] σ is the Sigmoid function, outputting the disaster probability P.

[0053] 2) According to the characteristics of debris flow, establish the relationship between the flow in the active section and the rainfall

[0054] The flow of debris flow in the sludge level monitoring section: Qc = Wc·Vc ………………(1)

[0055] In the formula: Qc—the debris flow flow at the survey section (m 3 / s);

[0056] Wc—the cross-sectional area (m 2 );

[0057] Vc—the debris flow velocity at the cross-section (m / s);

[0058] Muddy debris flow is considered as dilute debris flow, and the velocity empirical formula is used for calculation:

[0059]

[0060] In the formula:

[0061] V c —the average velocity of the debris flow cross-section (m / s);

[0062] —the velocity correction coefficient caused by the change in sediment concentration in the debris flow, α = (γ H Φ + 1) 1 / 2 ;

[0063] H—hydraulic radius (m), generally can be replaced by the average sludge level depth H (m);

[0064] I - Hydraulic gradient of debris flow (‰), generally the longitudinal slope of the gully bed can be used instead;

[0065] - Roughness coefficient of clear - water riverbed, the value of n is determined according to the specific gully morphology in accordance with the specifications.

[0066] The relationship between flow rate and mud level can be obtained from the comprehensive flow rate and flow velocity calculation formula, as shown in Equation (3).

[0067]

[0068] In the formula:

[0069] B - Average width of the debris - flow cross - section at the mud - level monitoring point (m);

[0070] H - Mud - level depth of debris flow at the mud - level monitoring point (m);

[0071] Thus, the relationship between mud level and flow rate can be obtained.

[0072] According to the rain - flood method, the flow rate of debris flow and rainfall can be fitted into the following formula:

[0073] Q C =(1 + Φ)Q P ·D C =αI·D C …………(4)

[0074] In the formula:

[0075] α - Comprehensive coefficient of debris - flow discharge, catchment area and catchment conditions, calculated and integrated by the rain - flood method;

[0076] I - Average hourly rainfall intensity (mm / h) of rainfall in the debris - flow basin;

[0077] DC - Channel blockage coefficient, initially determined by channel characteristics, adjusted according to the stability and scale of the major landslide and avalanche sources.

[0078] Integrating Equation (3) and Equation (4) can initially establish the relationship between rainfall and mud level. This can initially determine the flow rate and scale of debris flow before the formation of debris flow and before the mud level shows debris - flow danger signs.

[0079] If Q c ≥Q critical (Critical flow rate), it indicates that the flow rate predicted based on rainfall intensity and mud level reaches the critical value, and the downstream debris - flow drainage capacity is insufficient or may cause scouring and burial in the threatened area.

[0080] 3. Blockage - breach correction module:

[0081] When the volume change of the source material ΔV > 1000m 3 a breach model is triggered:

[0082] When the displacement and deformation of a major landslide source reach the threshold, it indicates debris flow activity of the source material parameters. There may even be channel blockage, and thus the risk of a breach. It is necessary to consider the blocking and breach effects on the debris flow discharge, increase the Dc value, and conduct early warning analysis based on the increased discharge.

[0083] Step 3: Calibrate and validate the early warning threshold using historical data

[0084] Calibrate and validate the early warning threshold using historical data. Use the random forest algorithm to screen the weights of key factors. According to the determined weights, substitute them into the disaster occurrence probability model, and perform cross-validation on the historical data to determine the false negative rate and false positive rate of the model. If the requirements are not met, adjust the weights of the key factors until the prediction accuracy is achieved. The specific process and technical requirements are as follows:

[0085] 1. Historical data training:

[0086] Use the random forest algorithm to screen the weights of key factors wi, and the AUC should be > 0.85.

[0087] The specific steps are as follows:

[0088] 1. Data cleaning and feature engineering → 2. Model training and importance calculation → 3. Weight normalization and dynamic correction → 4. Multi-model collaborative verification.

[0089] · Input data:

[0090] Feature variables (X): rainfall parameters (I max , E cum ), source material deformation parameters (displacement rate, volume change), mud level parameters (mud level rising rate), terrain and geological parameters (slope, lithology code), etc.

[0091] Target variable (y): binary label (1 = debris flow occurred, 0 = debris flow did not occur).

[0092] Data preprocessing:

[0093] Normalization: Standardize (Z-score) or normalize (Min-Max) continuous variables (such as rainfall intensity, displacement rate).

[0094] Encoding: Perform one-hot encoding on categorical variables (such as lithology type).

[0095] Balance data: If the positive and negative samples are unbalanced (few disaster samples), use SMOTE oversampling or random undersampling.

[0096] This method can effectively identify the core parameters triggering the dominant debris flow (such as short-term rainfall intensity and displacement rate), provide a quantitative basis for the threshold model, and ultimately reduce the false alarm rate and missed alarm rate.

[0097] 2. Cross-validation:

[0098] Divide 70% of the data for training and 30% for testing, requiring that the missed alarm rate (FN) < 5% and the false alarm rate (FP) < 10%.

[0099] 3. Dynamic adjustment:

[0100] · Based on Bayesian update, regularly and after a disaster occurs, correct the threshold parameters (such as Ic decreases by 10% when θ > 0.3m 3 / m 3 ).

[0101] Step 4: Perform multi-source data fusion based on the data obtained from monitoring, judge the data credibility of the monitoring data, and eliminate abnormal monitoring data.

[0102] Perform multi-source data fusion based on the data obtained from monitoring, judge the data credibility of the monitoring data, and eliminate abnormal monitoring data. The specific process and technical requirements are as follows:

[0103] 1. Data preprocessing process

[0104] 1) Temporal and spatial alignment:

[0105] Use a sliding time window (window size 5 minutes) to align data with different frequencies (such as rainfall at the minute level and displacement at the hour level).

[0106] Use GIS spatial interpolation (Kriging method) to fill the monitoring blind area.

[0107] 2) Noise filtering:

[0108] For mud level data: Use wavelet transform (Daubechies 4) to remove high-frequency vibration noise.

[0109] For displacement data: Use the CUSUM algorithm to detect mutation points and eliminate sensor drift.

[0110] 3) Anomaly detection:

[0111] Based on the Isolation Forest, identify outliers (such as a sudden increase in single-point rainfall but no response from the surrounding area).

[0112] 2. Multi-source cross-validation rules

[0113] 1) Consistency check:

[0114] If the rainfall exceeds the threshold but there is no displacement of the source material, start an unmanned aerial vehicle inspection to confirm the authenticity of the data.

[0115] If the mud level suddenly rises but the rainfall does not reach the standard, check the data of the pressure sensor at the upstream breach point.

[0116] 2) Conflict decision-making mechanism:

[0117] 2 / 3 principle: At least two monitoring means need to trigger an anomaly to issue a warning (such as rainfall + displacement, or displacement + mud level).

[0118] Step 5: Based on the monitoring data and meteorological forecast, through the analysis of the early warning model, conduct early warning grading, and issue early warning signals to prompt corresponding emergency response measures.

[0119] Based on the monitoring data and meteorological forecast, through the analysis of the early warning model, conduct early warning grading, and issue early warning signals to prompt corresponding emergency response measures. This part of the work is completed by the early warning subsystem, which conducts credible early warnings according to the risk coefficient of mountain flood and debris flow disasters. The specific content and technical requirements are as follows:

[0120] The early warning is issued based on the following indicators:

[0121] 1) Probability of occurrence value: Calculate the probability of disaster occurrence under the current conditions (0 - 1) through a machine learning model (such as random forest) or a statistical model.

[0122] 2) Rainfall threshold: Critical rainfall (such as hourly rainfall intensity ≥ 30mm or cumulative rainfall ≥ 100mm).

[0123] The rainfall-duration curve (I-D curve) determines the risk of short-term heavy rainfall.

[0124] 3) Mud level change: The mud level increases rapidly (the mud level at the monitoring point rises ≥ 0.5m within 1 hour).

[0125] Dry-up phenomenon (the river channel suddenly dries up, which may indicate upstream blockage).

[0126] 4) Instability of the material source: The reserve of loose accumulations (evaluate the material source quantity through remote sensing or drone inspection).

[0127] Risk of breach (the monitoring subsystem combines the slope stability model to judge the risk of breach and whether to trigger the breach mechanism).

[0128] 1. Early warning level classification

[0129] According to the combination of indicators, it is divided into four levels of early warnings (blue, yellow, orange, red):

[0130] Table 2 Early warning level classification table

[0131]

[0132]

[0133] 2. Automated Response Technology

[0134] Edge Computing Node: Deploy Jetson Xavier at the monitoring station to calculate the P value in real time and reduce the dependence on the cloud.

[0135] Blockchain Evidence Storage: The early warning records are uploaded to the blockchain (Hyperledger Fabric) to ensure the immutability of the decision-making process.

[0136] Step 6: Conduct post-evaluation and optimization of the model based on the monitoring data recorded after the monitoring and early warning system is enabled and the occurrence of disasters.

[0137] After the monitoring and early warning system is enabled, conduct post-evaluation and optimization of the model based on the monitoring data recorded during the operation (including rainfall and source deformation) and the occurrence of mountain flood and debris flow disasters (specific to flow rate, mud level, and threat area range). The key to this step is to adjust the model using the data during the operation period to solve the problems of the accuracy of the pre-model and thresholds, and to achieve the intelligent learning and improvement functions of the model. The specific content and steps are as follows.

[0138] 1. Post-disaster Data Review

[0139] 1) Precision Evaluation Index:

[0140] Early Warning Lead Time T lead (Target ≥ 1 hour);

[0141] Spatial Error (predicted range vs. actual impact area, with the requirement of overlap rate ≥ 80%).

[0142] 2) Root Cause Analysis:

[0143] Establish a false alarm / missed alarm case library and label the reasons (such as sensor failure, model parameter deviation).

[0144] 2. Model Iteration Method

[0145] Online Learning:

[0146] Adopt Online Random Forest, and update the model every time 10 new disaster data are added.

[0147] Parameter Inversion:

[0148] Based on the rainfall list, inversely deduce the comprehensive coefficient α of the debris flow discharge, area, and catchment conditions according to the post-disaster mud mark height, and correct the discharge formula:

[0149] α = Q C / (I·D C )

[0150] Q CThe debris flow discharge of the sudden disaster is calculated using the mud level and channel characteristics of the debris flow in the flow section;

[0151] I is the effective rainfall intensity when flash floods and mudslides occur;

[0152] Dc is the blockage coefficient of the debris flow channel. If the source of the material changes, it should be adjusted based on the original value.

[0153] 3. System upgrade cycle

[0154] Short term: Update threshold parameters monthly and optimize algorithm weights quarterly.

[0155] Long term: Reconstruct the channel 3D model and update the digital twin every year based on LiDAR scanning data.

[0156] Step 7: Utilize multiple mountain torrent and mudslide monitoring systems in the region to achieve cross-basin coordinated early warning and build group intelligent early warning.

[0157] Utilize multiple flash flood and debris flow monitoring systems in the region to achieve cross-basin coordinated early warning and build group intelligent early warning. Flash flood and debris flow is not a single geological disaster problem. Often, multiple channels will erupt simultaneously in one region. Therefore, sharing and analyzing regional data can not only improve the monitoring and early warning capabilities of a single object, but also achieve "point-surface dual control" of geological disasters and achieve cross-basin coordinated early warning.

[0158] System composition

[0159] On the other hand, the present invention proposes a flash flood and mud-rock flow early warning system based on comprehensive analysis of multi-source data, comprising:

[0160] The monitoring data acquisition subsystem includes rainfall monitoring equipment, mud level monitoring equipment, and displacement monitoring equipment. It is used to collect multi-source monitoring data of mountain torrents and mud-rock flows in the target area in real time by monitoring rainfall in the gully area, major material sources, and mud level that characterizes the flow characteristics of debris flows.

[0161] The comprehensive analysis and decision-making subsystem includes three parts: data preprocessing, early warning analysis model, and early warning level determination. By constructing a model that includes rainfall, source materials, and mud levels, and considering the impacts of various factors on debris flows, the possibility of debris flow occurrence is determined; an integrated monitoring and early warning evaluation system is realized, which changes from the threshold determination of the disaster level based on traditional single indicators to quantitative analysis with the combined action of multiple indicators; the random forest algorithm is used to screen the weights of key factors; according to the determined weights, they are brought into the disaster occurrence probability model, and historical data is cross-validated to preprocess the multi-source monitoring data of mountain flood and debris flows, obtaining a real-time monitoring data set; it is also used to judge the data credibility of the data in the real-time monitoring data set and screen out the reliable monitoring data set; a blockage and breach correction module is added to consider the impact of solid source materials on the outbreak of mountain flood and debris flows; cross-validation is carried out using multi-source data; after operation, the data is used for review and correction, and the model is iteratively calculated to improve the accuracy of the model; the preprocessed monitoring data is brought into the early warning model for analysis to determine the early warning level.

[0162] The early warning subsystem mainly includes emergency broadcast equipment, wireless communication transmission equipment, and data storage and regional sharing equipment. According to the early warning level determined in the comprehensive analysis and decision-making subsystem, the corresponding early warning broadcasts are sent to the threatened objects in the mountain flood and debris flow threat area by using the emergency broadcast equipment, and early warning text messages and early warning alarms are sent to the mobile communication devices in the threat area by using the wireless communication transmission equipment; the relevant data is stored by using the data storage equipment so that the system can be upgraded through post-disaster review and online learning, and the monitoring data is sent to the region for real-time sharing, and an intelligent early warning system combining points and areas in the region can also be established.

[0163] Furthermore, the monitoring data acquisition subsystem includes rainfall monitoring equipment arranged in the mountain flood and debris flow gully area, mud level monitoring equipment in the mountain flood and debris flow gully, and displacement monitoring equipment for major landslide and avalanche source materials on the side of the mountain flood and debris flow gully.

[0164] The early warning subsystem includes a wireless emergency broadcast device arranged in the threatened area and mobile terminals held by relevant personnel. Among them, the comprehensive analysis and decision-making subsystem is communicatively connected to the mobile terminals through communication equipment.

[0165] It should be noted that the monitoring data acquisition subsystem and the early warning subsystem may also include other equipment, which are not limited here.

[0166] In addition, in this system, an application provides a computer device, which can be a server. The computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device 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 computer device is used to store forest fire warning data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it realizes the mountain flood and debris flow warning method based on multi-source data comprehensive analysis described in Embodiment 1.

[0167] In this warning system, a computer device is also provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, it realizes the steps in the mountain flood and debris flow warning method based on multi-source data comprehensive analysis described in steps 2, 3, and 4.

[0168] In this warning system, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by the processor, it realizes the steps in the above-mentioned embodiment of the mountain flood and debris flow warning method based on multi-source data comprehensive analysis.

[0169] In this warning system, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the above-mentioned embodiment of the mountain flood and debris flow warning method based on multi-source data comprehensive analysis.

[0170] The remarkable effect of the present invention is:

[0171] 1. In the method and system of the present invention, monitoring data is collected by rainfall monitoring devices arranged in the gully area, mud level monitoring devices in the gully, and displacement monitoring devices for major landslide and debris sources on the gully side. By submitting the data of the rainfall monitoring devices, mud level monitoring devices, and displacement monitoring devices to the comprehensive decision-making system for analysis, when there is short-term heavy rainfall or continuous rainfall in the gully area, sudden increase in the gully mud level, or increase in the displacement of major landslide and debris sources, the data is fed back to the comprehensive analysis and decision-making system. According to the rainfall value, mud level value, and displacement value, the degree of mountain flood and debris flow outbreak is analyzed, and different-level warning signals and emergency evacuation instructions are issued by the wireless emergency broadcast system and / or mobile terminals in the threatened area. The accurate early warning and forecasting of mountain flood and debris flow disasters are realized, and the problems of false alarms and concealment reports of single warning devices are solved.

[0172] 2. In the method and system described in the present invention, the multi-source monitoring data of rainfall, material source and mud level are comprehensively utilized to analyze and process the data, remove abnormal values, compare and analyze the water source conditions, material source conditions and flow characteristic data characterizing the debris flow, and use the disaster history and the disaster occurrence of similar channels in the adjacent area to preliminarily determine the thresholds of each factor, and use the disaster history data to calculate the probability formula of disaster outbreak. The probability and danger of debris flow outbreak are comprehensively determined by combining the probability of debris flow, rainfall characteristics, material source displacement deformation blocking effect and debris flow mud level change data, and the early warning system issues a warning signal of the corresponding level to the threatened area object, and takes corresponding risk avoidance measures in time to reduce the hazards of debris flow. The early warning system realizes the comprehensive monitoring, early warning and evaluation system of quantitative analysis of the joint action of multiple indicators from the threshold determination of the disaster level of the traditional single indicator. The early warning of mountain torrents and debris flow is transformed from "experience-driven" to "data-model-decision" intelligent closed loop. The core of the system lies in the deep integration of dynamic threshold, multi-model collaboration and full-chain perception.

[0173] 3. In the method and system described in the present invention, according to the characteristics of debris flow, the thresholds of rainfall and mud level of the monitoring target ditch are inverted according to the relationship between ditch characteristics, rainfall, characteristic parameters of debris flow and mud level of the monitoring ditch section. The flow rate and the scope of the danger zone that may cause mountain torrents and debris flow can be more accurately inferred based on the rainfall in the ditch area that will occur. On the one hand, the data of meteorological forecasts can be used to release the probability and danger level of possible debris flow in advance, so that the protected objects can be prepared in advance and evacuated quickly after the monitoring data is released. On the other hand, by introducing the effective rainfall parameter into the early warning model, the impact of recent rainfall on the outbreak of mountain torrents and debris flow can be comprehensively reflected. The monitoring data of rainfall in the debris flow formation area can predict the possibility and scale level of debris flow outbreak. After the debris flow outbreak, before it reaches the mud level monitoring point in the circulation area, an early warning can be issued, providing more time for risk avoidance.

[0174] 4. In the method and system described in the present invention, the multi-source monitoring data after the operation of the monitoring system and the degree of debris flow outbreak and the flow data of the debris flow are analyzed, the false alarm and missed alarm rates are determined through intelligent analysis, the monitoring data and the probability of occurrence are brought in for post-disaster review and summary, the existing monitoring and early warning models are iteratively analyzed, the characteristic parameters of the debris flow are inverted and analyzed, and the system is continuously upgraded through the intelligent learning system. This measure can continuously improve the accuracy of the monitoring system.

[0175] 5. In the method and system described in the present invention, the monitoring data of adjacent areas (such as rainfall station network) are shared through blockchain technology to achieve cross-basin collaborative early warning, providing basic data and real-time data for management departments to establish group intelligent early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0176] Figure 1 is the flowchart of the early warning method of the present invention;

[0177] Figure 2 is the structural schematic diagram of the early warning system of the present invention. Detailed implementation manners

[0178] The following further elaborates in detail on the specific implementation manners and working principles of the present invention in conjunction with the accompanying drawings.

[0179] Example 1:

[0180] Taking the application in a certain gully in Wenchuan County, Sichuan as an example:

[0181] Step 1: Collect and preprocess multi-source monitoring data of mountain torrents and debris flows in the target area in real time to obtain a real-time monitoring data set;

[0182] During specific implementation, 6 rain gauges are arranged in the gully area, 4 mud level gauges are arranged in the flow area of the main gully and at the confluence of the main gully and the main tributaries, and 12 displacement gauges are arranged for major material sources in the gully area, with the fiber optic length being 2 km.

[0183] Step 2: Construct a debris flow early warning model

[0184] 1) Dual-variable rainfall threshold

[0185] According to the historical debris flow outbreak situation of this gully, an effective rainfall parameter is introduced, and a rainfall threshold model for this gully is fitted using dual-variable rainfall data.

[0186] 2) Construct the core formula for the probability of disaster outbreak:

[0187]

[0188] Where:

[0189] Id = f(Ecum) (dynamic rainfall intensity threshold);

[0190] Hc and Vmc are the monitored mud level height and the displacement rate of the material source;

[0191] Ic, Hc, and Vmc are the rainfall intensity, mud level height, and displacement rate thresholds of the material source;

[0192] 3) According to the rain flood method, the flow rate of debris flow and rainfall can be fitted into the following formula:

[0193] Q C = (1 + Φ)Q P ·D C = αI·D C

[0194] Step 3: Calibrate and verify the warning thresholds using historical data.

[0195] Use historical data to determine the rainfall parameters (E cum ), source displacement (Vm), and mud level value (Hc) in the probability model, etc.

[0196] Determine the thresholds of each monitoring data in the model according to historical data. (The following thresholds are for red alerts. The case is only for illustrating the system working process, and the thresholds and parameters for other levels are not elaborated here.)

[0197] Adopt double rainfall intensity parameters, I max = 100 mm.

[0198] The displacement threshold of the source is determined according to engineering experience, Vmc = 0.05 m / d.

[0199] The mud level threshold of the debris flow is comprehensively determined according to the channel characteristics and the flow capacity downstream to determine the critical flow Q critical = 180 (m 3 / s), and the corresponding mud level threshold Hc = 2.15 m.

[0200] Determine the comprehensive coefficient α = 2.35 of the debris flow discharge, area, and catchment conditions through channel characteristic analysis.

[0201] The blockage coefficient Dc of the debris flow channel = 2.2.

[0202] Step 4: Data processing and cross-validation;

[0203] It should be noted that the rainfall monitoring data, mud level monitoring data, and displacement monitoring data are all data sequences formed by continuous monitoring data over a period of time in the past.

[0204] 1. The steps for preprocessing the multi-source monitoring data of mountain torrents and debris flows are as follows:

[0205] Step 4.1: Align different frequency data (such as rainfall in minutes and displacement in hours) using a sliding time window (window size 5 minutes). Perform data cleaning on the multi-source monitoring data of mountain torrents and debris flows;

[0206] Step 4.2: Calculate the data similarity between any two monitoring data after cleaning;

[0207] Step 4.3: Aggregate all monitoring data according to the data similarity and similarity threshold to obtain a real-time monitoring dataset.

[0208] It should be noted that there are inevitably obvious errors in the original monitoring data, so this part of the data is removed through cleaning. In addition, since the amount of multi-source monitoring data of mountain torrents and debris flows is relatively large, the dimension of the data can be effectively reduced through aggregation processing. In other embodiments, other methods can also be used to achieve this, which will not be elaborated here.

[0209] The statistical results based on the processed data are as follows:

[0210] Rainfall intensity value: short-term rainfall I = 25.3 mm / h, 24-hour effective rainfall E = 36.5 mm / h.

[0211] The displacement value of the main material source through spatio-temporal processing V = 0.035 m / d.

[0212] For the main gully, especially the gully in the flow area adjacent to the protected object, the mud level H = 2.2 m is monitored.

[0213] 2. Multi-source cross-validation rules

[0214] According to the monitoring data, the rainfall, material source displacement, and mud level in the gully all change, and the change trends are consistent and can be mutually verified, indicating that rainfall has an adverse impact on both the gully mud level and the stability of the material source. Therefore, the monitoring and early warning level can be determined by substituting the monitoring data into the model and comparing it with the threshold.

[0215] Step five, determination and release of the early warning level.

[0216] Substitute the monitoring data into the core formula for disaster outbreak, and the calculated probability of mountain torrents and debris flows is 0.83.

[0217] Substitute the rainfall intensity and blockage coefficient into the flow calculation formula, and the predicted flow Q = 198.5 m 3 / s.

[0218] The comparison of each index with the threshold is as follows:

[0219] Rainfall intensity I = 45E0.3 = 45 × 36.5 0.3 = 132.403 > Ic = 100 mm

[0220] The displacement of the material source V = 0.035 < Vmc = 0.05 m / d.

[0221] Flow Q = 198.5 m 3 / s > Q critical = 180 (m 3 / s).

[0222] Mud level H = 2.2 m > threshold Hc = 2.15 m.

[0223] According to the calculation results, it shows that the rainfall intensity, flow rate, and mud level in this gully area all exceed the thresholds, and the probability value of mountain flood outbreak reaches 0.83, exceeding the red warning standard of 0.80. Although the deformation amount of the material source does not reach the threshold, it is analyzed that it may be caused by the differences in geological bodies. Refer to the following table to determine the warning level as red. The warning subsystem issues an alarm to the entire threatened area, requiring the monitoring personnel to organize the evacuation of people in the dangerous area, notify the warning results to the management department and the emergency department, keep the emergency rescue team on standby, and activate the post-disaster rescue plan.

[0224] Table 3 Warning Level Table

[0225]

[0226] Step 6, conduct post-evaluation and optimization of the model after the disaster occurs.

[0227] The key to this step lies in using the data during operation to adjust the model, solve the problem of the accuracy of the pre-set model and thresholds, and realize the intelligent learning and improvement functions of the model. It is found that the instability of the material source participates in the debris flow activity. The monitoring data shows that when the slope body continues at a speed of 0.035 m / d for less than 30 minutes, it enters the accelerated deformation stage and quickly becomes unstable to participate in the debris flow activity. This indicates that the originally set threshold is too large. Adjust the threshold according to the deformation monitoring data of each slope body, and the threshold setting is adjusted to 0.035 m / d. Due to the debris flow scouring the gully, the gully blockage coefficient is adjusted to DC = 2.5.

[0228] Step 7, summarize the data of this monitoring and warning into the regional database to achieve data sharing.

[0229] In this case, the accuracy of the warning is improved, and the warning is issued when the rainfall intensity reaches the threshold and the mud level just rises to the threshold. This method can reduce the false alarm rate and missed alarm rate when applied in the region, and the warning time is also greatly advanced, providing more evacuation time.

[0230] Example 2:

[0231] For the early meteorological warning and monitoring data, taking a certain gully in Maoxian, Sichuan as an example for the situation where there are abnormalities and data mismatches in the monitoring data, the operation process is described as follows:

[0232] Step 1, collect and preprocess the multi-source monitoring data of mountain flood and debris flow in the target area in real time to obtain the real-time monitoring data set;

[0233] Specifically, when implementing, 4 rain gauges are arranged in the gully area, 2 mud level gauges are arranged in the flow-through area of the main gully and the confluence of the main branch gully and the main gully, 4 displacement gauges are arranged for the major material sources existing in the gully, and the optical fiber length is 1.8 km.

[0234] Step 2, construct a debris flow warning model

[0235] 1) Bivariate threshold of rainfall

[0236] According to the historical debris flow outbreaks in this gully, an effective rainfall parameter is introduced, and a rainfall threshold model for this gully is fitted using bivariate rainfall data. The rainfall intensity I = 30E 0.25 .

[0237] 2) Construct the core formula for the probability of disaster outbreak:

[0238]

[0239] Where:

[0240] Id = f(Ecum) (dynamic rainfall intensity threshold);

[0241] Hc and Vmc are the monitored mud level height and the displacement rate of the source material;

[0242] Ic, Hc, and Vmc are the rainfall intensity, mud level height, and displacement rate thresholds of the source material;

[0243] 3) According to the rain - flood method, the flow rate of debris flow and rainfall can be fitted into the following formula:

[0244] Q C = (1 + Φ)Q P ·D C = αI·D C

[0245] Step 3: Use historical data to calibrate and verify the warning threshold.

[0246] Use historical data to determine the rainfall parameter (E cum ) in the probability model, the displacement of the source material (Vm), and the mud level value (Hc), etc.

[0247] Determine the thresholds of each monitoring data in the model according to historical data. (The following thresholds are for red - alert. The case is only for illustrating the working process of the system, and the thresholds and parameters for other levels are not elaborated here.)

[0248] Adopt a double - rainfall - intensity parameter, I max = 85mm, and the short - term rainfall intensity threshold I = 25mm.

[0249] The displacement threshold of the source material is determined according to engineering experience, Vmc = 0.05m / d.

[0250] The mud level threshold of debris flow is comprehensively determined according to the gully characteristics and the downstream flow - passing capacity, and the critical flow rate Q critical = 140 (m 3 / s), and the corresponding mud level threshold Hc = 1.85m.

[0251] Through the analysis of channel characteristics, the comprehensive coefficient α of debris flow discharge, area, and catchment conditions is determined to be 1.95.

[0252] The blockage coefficient Dc of the debris flow channel is 2.3.

[0253] Step 4: Data processing and cross-validation;

[0254] Before a certain rainfall, the meteorological forecast indicated that there would be moderate to heavy rain in this area, and the predicted rainfall would reach 30 mm. An orange geological disaster warning was issued. Considering the threshold of mountain flood and debris flow outbreak in this channel, since the rainfall was greater than the threshold, meteorological warning information was released in advance, and the attention process was entered ahead of time.

[0255] 1. After the rainfall occurred, the steps for preprocessing the multi-source monitoring data of mountain flood and debris flow are as follows:

[0256] Step 4.1: Align data with different frequencies (such as rainfall in minutes and displacement in hours) using a sliding time window (window size 5 minutes). Clean and process the multi-source monitoring data of mountain flood and debris flow;

[0257] Step 4.2: Calculate the data similarity between any two monitoring data after cleaning and processing; conduct a similarity analysis of rainfall in different regions. It is found that rainfall occurred at all 4 rainfall monitoring points in the channel, with the rainfall ranging from 15 to 30 mm. Among them, the rainfall at three monitoring points was less than 20 mm, and only one point reached 30 mm.

[0258] Step 4.3: Conduct cross-validation on the monitoring data. It is found that the displacement of the source points in the gully area is less than 3 cm. The change in the mud level in the channel is mostly less than 1 m, and the mud level increment is 1.5 m near the point with large rainfall, with the monitored mud level value being 1.95 m and the mud level value in the main channel flow area being 1.25 m.

[0259] According to the above monitoring data, the channel flow in the channel has increased under rainfall conditions, but there is no ability to cause disasters.

[0260] The statistical results based on the processed data are as follows:

[0261] For the rainfall intensity values: the short-term rainfall I = 18.5 mm / h, and the 24-hour effective rainfall E = 26.8 mm.

[0262] The displacement value of the main source through spatio-temporal processing V = 0.008 m / d.

[0263] For the main channel, especially the channel in the flow area adjacent to the protected object, the monitored mud level H = 1.25 m.

[0264] 2. Multi-source cross-validation rules

[0265] According to the monitoring data, the rainfall, source displacement, and mud level in the gully have all changed, and their changing trends are basically the same. However, the rainfall and mud level at individual points exceed the average value by a large margin, indicating that there is a large amount of regional rainfall in the basin.

[0266] Although the rainfall at individual points reaches the threshold, the overall situation shows that the gully area has not reached the conditions for a flash flood and debris flow outbreak.

[0267] Therefore, the monitoring and early warning level can be determined by substituting the monitoring data into the model and comparing it with the threshold. Step 5: Determination and release of the early warning level.

[0268] Substitute the monitoring data into the core formula for disaster outbreak, and the calculated probability of a flash flood and debris flow is 0.48.

[0269] Substitute the rainfall intensity and blockage coefficient into the flow calculation formula, and the predicted flow Q = 65m 3 / s.

[0270] The comparison of each index with the threshold is as follows:

[0271] Rainfall intensity I = 45E 0.3 = 30×26.8 0.25 = 68.26 < Ic = 85mm

[0272] The displacement of the source V = 0.008 < Vmc = 0.05m / d.

[0273] Flow Q = 65m 3 / s < Q critical = 140(m 3 / s).

[0274] Mud level H = 1.25m < threshold Hc = 1.85m.

[0275] According to the calculation results, it shows that the rainfall intensity, flow, and mud level in this gully area have not exceeded the threshold, and the probability value of a flash flood reaching 0.65 reaches the yellow warning level. Although the rainfall at individual rain gauge points reaches the threshold, the overall situation does not reach the conditions for emergency evacuation. Refer to the following table to determine the warning level as yellow. The warning subsystem issues an alarm to the townships in the threatened area, restricts the activities of people in high-risk areas, notifies the management department and the emergency department of the warning results, keeps the emergency rescue team on standby, and prepares emergency supplies.

[0276] Table 4 Warning Level Table

[0277]

[0278] Step 6: Conduct post-evaluation and optimization of the model after the disaster occurs.

[0279] The key to this step lies in using the data during operation to adjust the model, solve the problems of the accuracy of the early model and threshold, and realize the intelligent learning and improvement functions of the model. Through later investigation, it is found that in the local area of the gully area, due to the relatively large rainfall on the sunny slope, the rainfall monitoring data at individual rainfall monitoring points is greater than the overall. Although multiple source points in the gully are affected by rainfall, no large deformation has occurred, and solid substances have not participated in debris flow activities, so no large-scale mountain flood debris flow has been formed. At the same time, there has been no continuous rainfall in the gully area before rainfall, the effective rainfall value is small, and the rainfall is converted into a small channel runoff flow. The original warning model basically accurately reflects the characteristics of mountain flood debris flow.

[0280] Step 7: Summarize the data of this monitoring and warning into the regional database to achieve data sharing.

[0281] In this case, through the comprehensive judgment of multi-source data, it is possible to avoid issuing high-risk warnings due to abnormal data at individual points, accurately reflect the characteristics of mountain flood debris flow in the entire gully area, and improve the accuracy of warnings to avoid false alarms of issuing orange warnings simply based on meteorological warnings.

[0282] In the above case, through multi-source monitoring means, this system can comprehensively master the water source conditions, source conditions affecting mountain flood debris flow, and the mud level characteristics representing the debris flow flow.

[0283] In the case, through the preprocessing and cross-validation of the monitoring data, false alarms and missed alarms caused by abnormal individual monitoring data can be avoided.

[0284] In the above case, the monitoring system uses the incremental random forest method to update the model through online learning, can continuously update the data, and according to the specific monitoring object, can establish a more targeted and accurate monitoring and warning model, and regularly adjust the threshold to adapt to the changes in the gully.

[0285] The above case summarizes the monitoring and warning data to form a database that is updated and expanded in real time, which is convenient for the risk warning and analysis of geological disasters in the region.

[0286] To sum up, first preprocess the multi-source monitoring data of mountain flood debris flow collected, which can effectively reduce obvious incorrect data and the dimension of the data; then judge the data credibility of the data in the real-time monitoring data set, and the use of cross-validation and dynamic adjustment can help screen out the monitoring data with high credibility in the monitoring data and retain it, so as to obtain a monitoring credible data set and predict and obtain preliminary prediction data. According to the comprehensive analysis of each warning data, accurately evaluate and warn the danger level of mountain flood debris flow disasters. To sum up, through comprehensive analysis, scientific warning model construction, and real-time warning threshold and parameter correction, this scheme improves the credibility of mountain flood debris flow disaster warnings.

[0287] The above has introduced the technical solution provided by the present invention in detail. Specific examples are used herein to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only for helping to understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A flash flood and debris flow warning method based on comprehensive analysis of multi-source data, characterized in that, It includes the following steps: Step 1: Set up monitoring equipment for the target object, construct a gully monitoring system, and the monitoring data acquisition subsystem collects and preprocesses multi-source monitoring data of mountain torrents and debris flows in the target area in real time to obtain a real-time monitoring data set; Step 2: The comprehensive analysis and decision-making subsystem establishes a threshold model framework for specific monitoring targets based on the rainfall intensity, debris flow discharge, scouring volume, and solid source distribution of debris flows in the disaster history of the region and the monitoring object; Step 3: Calibrate and verify the warning threshold using historical data; Step 4: Conduct multi-source data fusion based on the monitored data, judge the data credibility of the monitoring data, and eliminate abnormal monitoring data; Step 5: According to the monitoring data and meteorological forecasts, through the analysis of the warning model, conduct warning classification, and the warning subsystem issues warning signals to prompt corresponding emergency response measures; Step 6: Conduct post-evaluation and optimization of the model through the monitoring data recorded after the monitoring and warning system is enabled and the disaster occurrence situation; Step 7: Use multiple mountain torrent and debris flow monitoring systems within the region to achieve cross-basin collaborative warning and construct a group intelligent warning.

2. The method for flash flood and debris flow warning based on comprehensive analysis of multi-source data according to claim 1, wherein: The multi-source monitoring data of mountain torrents and debris flows include rainfall monitoring data in the mountain torrent and debris flow gully area, mud level monitoring data in the mountain torrent and debris flow gully channel, and displacement monitoring data of major landslide and debris sources on the side of the mountain torrent and debris flow gully channel. Thus, a monitoring data acquisition subsystem is established.

3. The flash flood and debris flow warning system based on comprehensive analysis of multi-source data according to claim 1, wherein: Through Step 2, the comprehensive analysis and decision-making subsystem can comprehensively analyze the relationship between each monitoring data and the occurrence of debris flows, and recommend corresponding threshold models, specifically including the following content: According to the disaster history and the situation of disasters occurring in similar channels in the adjacent area, change the traditional mode of only using a single rainfall intensity, fit the short-term rainfall intensity to the effective cumulative rainfall E, construct a double-threshold model of rainfall, initially determine the thresholds of each element respectively, and use the disaster history data to statistically obtain the probability formula for the occurrence of disasters, realizing the comprehensive monitoring and warning evaluation system from the threshold determination of the traditional single index to the quantitative analysis of the joint action of multiple indicators; For the characteristics of mountain torrents and debris flows, construct a model between the mud level height, rainfall, and debris flow discharge, and invert the thresholds of rainfall and mud level in the monitored target gully according to the possible hazards caused by different flow rates in the threatened area; Add a blockage and breach correction module to the warning model, use the displacement deformation monitoring of major landslide and debris sources to timely control the stability of the source and its participation in debris flow activities. When the source deforms and becomes unstable, causing the gully to be blocked, start the debris flow blockage and breach correction module to improve the debris flow warning level.

4. The flash flood and debris flow warning method based on comprehensive analysis of multi-source data according to claim 1, characterized in that: In Step 3, the comprehensive analysis and decision-making subsystem is used to calibrate and verify the warning threshold using historical data, specifically including: Use the random forest method to screen the weights of key factors through historical data; Use the data to cross-validate the weights of each monitoring data determined in the model, and limit the false negative rate and false positive rate; Based on Bayesian update, correct the threshold parameters regularly and according to the mutation of monitoring data, and conduct dynamic adjustment.

5. The flash flood and debris flow warning method based on comprehensive analysis of multi-source data according to claim 1, characterized in that: In step 4, the subsystem for obtaining data based on monitoring data performs multi-source data fusion, and the comprehensive analysis and decision-making subsystem judges the data credibility of the monitoring data and eliminates abnormal monitoring data; The specific process is as follows: 1) The data preprocessing process includes: (1) Spatial and temporal alignment processing of data A. Use a sliding time window to align different data; B. Use GIS spatial interpolation (Kriging method) to fill the monitoring blind area; (2) Filter the noise of the monitoring data: A. Use wavelet transform (Daubechies 4) to remove the high-frequency vibration noise of the mud level data; B. Use the CUSUM algorithm to detect the mutation points of the displacement data and eliminate sensor drift; (3) Anomaly detection: Identify outliers based on the Isolation Forest; 2) Perform cross-validation on multi-source monitoring data (1) Perform consistency verification on the data collected by the monitoring system; (2) Build a conflict decision-making mechanism to determine data conflicts and avoid false alarms caused by individual data anomalies.

6. The method for flash flood and debris flow warning based on comprehensive analysis of multi-source data according to claim 1, characterized in that: In step 5, the data analyzed and verified in step 4 and the meteorological forecast data of the region are brought into the warning model determined in steps 2 and 3 to determine the reached warning level. The warning subsystem issues corresponding warning signals and prompts corresponding measures according to the corresponding warning level. The specific process is as follows: The warning system determines the corresponding warning level based on the indicators of occurrence probability, rainfall intensity, mud level change, and source stability in the decision-making system; According to the indicator combination, it is divided into four levels of warnings, corresponding to four situations: attention, warning, high risk, and emergency; Deploy Jetson Xavier at the monitoring station to perform calculations on the edge nodes, calculate the P value in real time, and reduce the dependence on the cloud; Through blockchain evidence storage, upload the warning records to the blockchain to ensure that the decision-making process cannot be tampered with.

7. The flash flood and debris flow warning method based on comprehensive analysis of multi-source data according to claim 1, characterized in that: Regarding the lack of existing disaster data for the current monitoring of mountain torrent debris flow, there may be deviations in the initial warning model and threshold setting. This mountain torrent debris flow warning method based on comprehensive analysis of multi-source data uses multi-source monitoring data and disaster occurrence situations during the operation process to conduct data review in step 6. Through online learning, the incremental random forest method is used for model iteration, and the parameters of the original model are inversely analyzed to continuously upgrade the system and improve the pertinence of the system and the accuracy of the warning. The specific process is as follows: 1) Conduct data review after the disaster, evaluate the accuracy of the warning, establish a false alarm / missed alarm case library, and mark the reasons; 2) Use incremental random forest for online learning, iteratively update the model, and inversely analyze the parameters initially set in the model to improve the accuracy of the model; 3) Every year, in combination with the integrated aerial, ground, and space three-inspection work, use LiDAR scanning data to reconstruct the three-dimensional model of the gully and update the digital twin.

8. The method for early warning of mountain flood and debris flow based on comprehensive analysis of multi-source data according to claim 1, characterized in that: Using regional big data and the system's intelligent learning function, in step 7, use multiple mountain torrent debris flow monitoring systems in the region to achieve cross-basin collaborative warning, build a group intelligent warning, improve the warning accuracy, and realize data sharing. The specific process is as follows: Share the monitoring data of adjacent regions through blockchain technology to achieve cross-basin collaborative warning; Use the conditions for the occurrence of surrounding geological disasters and the model to correct the data of the monitoring object; When geological disasters such as mountain torrents and debris flows occur in adjacent peripheral channels, this channel can enter the attention level, reducing the risks caused by the untimely manifestation of the movement characteristics of debris flow disasters due to regional rainfall differences, overly large channel ranges, and lag in the deformation of the material source.

9. The flash flood and debris flow warning method based on comprehensive analysis of multi-source data according to claim 1, characterized in that: By arranging monitoring projects according to the characteristics of the channel and selecting monitoring equipment with stronger anti-interference ability, an efficient and accurate monitoring subsystem has been formed. The specific methods are as follows: In the case of the existence of large material sources that may block the channel, fiber Bragg grating displacement meters and material source monitoring equipment can reflect the activity and stability of the material source, and soil moisture sensors can feedback the saturated condition of the slope material source, providing a basis for considering the role of the material source; Tipping bucket rain gauges or microwave radar rain gauges are respectively arranged in the upper reaches, middle reaches, and catchment areas of the channel. They have the advantages of high precision and strong anti-interference ability and can conduct rainfall monitoring effectively; when the channel basin area is large and there are many tributaries, monitoring can also be carried out by dividing into sub-hydrological unit areas to improve the overall reflection ability; Non-contact microwave radar mud level gauges and ultrasonic mud level gauges are used and installed in stable channels in the flow-through and protected areas, which can accurately feedback the debris flow mud level and flow characteristics of each channel section.

10. The method for early warning of mountain torrents and debris flows based on comprehensive analysis of multi-source data according to claim 1, characterized in that: Using rainfall monitoring, material source monitoring, and mud level changes to comprehensively determine the threshold model for debris flow occurrence, reducing the probability of false alarms and missed alarms of traditional single indicators. The specific methods are as follows: For the rainfall index, the traditional early warning relying only on the short-term rainfall intensity index is changed to jointly determining the rainfall intensity by the short-term rainfall intensity and the longer effective rainfall amount, comprehensively reflecting the gradual accumulation of long-term rainfall and the saturation effect on the material source, and superimposing the effect of short-term heavy rainfall near the disaster, more comprehensively considering the impact of rainfall on mountain torrents and debris flows; By monitoring the material source through displacement, moisture content, etc., the possibility of solid material sources participating in debris flows can be reflected; Using the monitoring of the mud level in the stable channel section can not only reflect the situation of the channel discharge, but also combine with the monitoring data of water source conditions and material source conditions to establish a connection between the debris flow discharge and rainfall, and the blocking and collapse of the material source; The monitoring data obtained by multi-source monitoring means not only complement each other, but also can be verified and cross-checked with each other to improve the prediction accuracy; Using multiple mountain torrent and debris flow monitoring systems in the region to achieve cross-basin collaborative early warning and construct a group intelligent early warning can solve the problems of insufficient data and untimely response of a single monitoring object. It can not only improve the early warning accuracy of a single monitoring object, but also provide a basis for the monitoring of the entire region, realizing the "point and area dual control" of geological disasters.

11. The method for early warning of mountain torrents and debris flows based on comprehensive analysis of multi-source data according to claim 1, characterized in that: Through comprehensive analysis in the decision-making subsystem, not only the susceptibility model of mountain torrents and debris flows is constructed, but also the relationship between the flow, rainfall, and material source is constructed, which can quantitatively evaluate the development degree of debris flows, so as to more accurately determine the basis for different monitoring objects and different early warning levels, specifically including: Since the disaster-causing impact of debris flow is mainly reflected in the debris flow discharge and the drainage capacity of the threatened area, different debris flow discharges correspond to different scouring and silting ranges, and their threatened object areas are also different. Traditional methods only make evaluations and judgments through a single index without considering the scope of the threatened area. By using the relationship between the discharge and rainfall in Step 2, the discharges and dangerous areas that may be formed by different rainfall amounts can be analyzed, and then their different danger levels can be determined. Based on the deformation, moisture content, and stability of the debris source, the dam-break effect formed when a large amount of debris source participates in debris flow activities can be considered and superimposed into the discharge calculation model. Even the dam-break correction model can be activated to correct the debris flow discharge for early warning analysis.

12. The method for flash flood and debris flow warning based on comprehensive analysis of multi-source data according to claim 1, wherein: Through the comprehensive analysis and decision-making subsystem for establishing the early warning model, through the analysis of rainfall and debris source stability, an analysis and judgment can be made before the formation of debris flow, and the danger level of debris flow can be predicted. Specifically, it includes: For mountain flood debris flow, due to its large basin area and channel length, it takes a certain amount of time for debris flow to form after rainfall. However, when the debris flow enters the flowing area, its flow velocity is fast, and the flow velocity of some channel debris flows can reach about 10 m / s. If only relying on mud level monitoring, when the monitoring data is sent, it only takes a few minutes for the debris flow to reach the protected object after passing through the mud level gauge in the flowing area. It is difficult to evacuate even if the early warning signal is issued in a timely manner. Through model analysis, by using the rainfall and debris source deformation monitoring data several kilometers away from the protected object, an early warning can be issued before the debris flow is fully formed, which can greatly advance the monitoring and early warning time and provide more time for personnel evacuation. The comprehensive analysis and decision-making subsystem can also issue attention information in advance according to the rainfall amount in the channel, the cumulative rainfall over a long time, and the changes in the debris source in the weather forecast, so that the people in the threatened area can make preparations in advance and provide better conditions for evacuation and risk avoidance.

13. The method for flash flood and debris flow warning based on comprehensive analysis of multi-source data according to claim 1, wherein: By correcting and proofreading the monitoring data in the comprehensive analysis and decision-making subsystem, the accuracy of the monitoring data can be improved. The specific steps for determining the error correction coefficient according to the prediction confidence level include: Determining the error correction amount of the monitored reliable data according to the error correction coefficient; Correcting the error of the monitored reliable data according to the error correction amount and forming secondary prediction data; Calculating the prediction deviation value between the secondary prediction data and the preliminary prediction data. If the prediction deviation value is less than the prediction deviation threshold, the preliminary prediction data is corrected using the error correction amount to obtain reliable prediction data. If it is greater than the prediction deviation threshold, the secondary prediction data is used as the new preliminary prediction data and the above steps are repeated.

14. A flash flood and debris flow warning system based on comprehensive analysis of multi-source data, characterized in that: Including: The monitoring data acquisition subsystem, including rainfall monitoring equipment, mud level monitoring equipment, and displacement monitoring equipment, which is used for multi-source monitoring of the rainfall, major debris sources, and mud levels representing the debris flow discharge characteristics in the gully area to collect multi-source monitoring data of mountain flood debris flow in the target area in real time; The comprehensive analysis and decision-making subsystem, including three parts: data preprocessing, early warning analysis model, and early warning level determination; By constructing a model that includes rainfall, material sources, and mud levels, and considering the impacts of various factors on debris flows to determine the likelihood of debris flow occurrence; realizing a comprehensive monitoring, early warning, and evaluation system that transitions from the threshold determination of disaster levels based on traditional single indicators to quantitative analysis with the combined action of multiple indicators; using the random forest algorithm to screen the weights of key factors; according to the determined weights, substituting them into the disaster occurrence probability model, and using historical data for cross-validation to preprocess multi-source monitoring data of mountain flood debris flows to obtain a real-time monitoring data set; also used to judge the data credibility of the data within the real-time monitoring data set and screen out a monitoring credible data set; adding a blockage and breach correction module to consider the impact of solid material sources on the outbreak of mountain flood debris flows; using multi-source data for cross-validation; After operation, use the data for review and correction, perform iterative calculations on the model to improve the accuracy of the model; substitute the preprocessed monitoring data into the early warning model for analysis to determine the early warning level; The early warning subsystem mainly includes emergency broadcast equipment, wireless communication transmission equipment, data storage and regional sharing equipment. According to the early warning level determined in the comprehensive analysis and decision-making subsystem, use the emergency broadcast equipment to issue corresponding early warning broadcasts to the threatened objects in the mountain flood debris flow threat area, and use the wireless communication transmission equipment to send early warning text messages and early warning alarms to mobile communication devices in the threat area; use the data storage equipment to store relevant data for the system to upgrade through post-disaster review and online learning, and send the monitoring data to the region for real-time sharing. An intelligent early warning system combining point and area can also be established in the region.

15. The flash flood and debris flow warning system based on comprehensive analysis of multi-source data according to claim 14, characterized in that: The monitoring data acquisition subsystem includes rainfall monitoring equipment arranged in the mountain flood debris flow gully area, mud level monitoring equipment in the mountain flood debris flow gully, and displacement monitoring equipment for major landslide and avalanche material sources on the side of the mountain flood debris flow gully; The early warning subsystem includes a wireless emergency broadcast device arranged in the threatened area and mobile terminals in the hands of relevant personnel. Among them, the comprehensive analysis and decision-making subsystem is communicatively connected to the mobile terminals through communication equipment.

Images