Multi-model fusion debris flow early warning method, device, electronic equipment and system

Abstract

This invention provides a multi-model fusion method, device, electronic device, and system for debris flow early warning. The method includes: collecting geographical and geological data, real-time rainfall, runoff, and auxiliary monitoring data; fusing a shallow flow numerical model, a moving least squares machine learning model, and a generalized regression neural network model, and obtaining a comprehensive runoff prediction through weighted least squares calculation; constructing critical runoff indicators based on the safety requirements of disaster-bearing bodies such as roadbeds, retaining walls, and culverts, and inversely calculating critical rainfall thresholds, then making differentiated adjustments based on the average gradient of the gully; generating a debris flow early warning report by comparing real-time predictions with thresholds, achieving final graded and zoned early warning for the warning area. This invention improves the accuracy and engineering relevance of early warnings by integrating physical mechanisms and data patterns, achieving refined early warning from "whether it will occur" to "when, where, and what the impact will be," while reducing the error of a single model and adapting to different terrains and disaster-bearing bodies along highways.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, and in particular to a multi-model fusion method, device, electronic equipment, and system for debris flow early warning. It is suitable for refined early warning and prevention decision-making in high-risk debris flow areas such as along highways and small watersheds in mountainous areas. Background Technology

[0002] Debris flows are a typical sudden geological disaster in mountainous areas. Especially under meteorological conditions such as typhoons and heavy rainfall, they can easily form high-water floods in small watersheds along highways, eroding loose deposits in the gully bed and thus inducing clusters of debris flows. This poses a serious threat to the safety of infrastructure such as highways and railways, as well as construction camps, construction access roads, and the lives and property of residents along the route.

[0003] Currently, debris flow early warning mainly relies on simulations using a single hydrological model (such as the NAM model) or empirical rainfall thresholds for judgment. However, while traditional hydrological models can simulate runoff physical processes, they struggle to capture the "short-duration, highly fluctuating" nonlinear relationship between rainfall and runoff under typhoon rainfall. Furthermore, machine learning models, when applied alone, are easily limited by the sample size of small watersheds in mountainous areas. Therefore, their accuracy in responding to sudden runoff changes below the coarsening layer of the gully bed is unstable and easily affected by environmental disturbances. Summary of the Invention

[0004] This invention provides a debris flow early warning method, device, electronic equipment, and system based on multi-model fusion, which addresses the shortcomings of existing technologies in terms of unstable early warning accuracy and susceptibility to environmental interference, thereby achieving the goal of effectively improving early warning accuracy and engineering relevance while reducing the impact of environmental interference.

[0005] This invention provides a multi-model fusion method for debris flow early warning, comprising: Obtain basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the early warning area; Based on the aforementioned geographical and geological data and the aforementioned real-time rainfall monitoring data, a shallow water flow numerical model is driven to simulate and obtain the first runoff prediction result. Based on the aforementioned real-time rainfall monitoring data and the aforementioned real-time runoff monitoring data, a second runoff prediction result and a third runoff prediction result are obtained respectively through a machine learning model and a neural network model. The least squares method is used to weight and fuse the first runoff prediction result, the second runoff prediction result, and the third runoff prediction result to obtain a comprehensive runoff prediction result; Based on the safety requirements of disaster-bearing bodies within the warning area, a dual critical threshold is constructed, and based on the dual critical threshold and the comprehensive runoff prediction results, the corresponding critical rainfall threshold is obtained by reverse calculation. By combining the average gradient of gullies in different zones within the warning area, the critical rainfall threshold is adjusted differentially to form a hierarchical and zoned warning threshold library; Based on the hierarchical and zonal early warning threshold library and the comprehensive runoff prediction results, a debris flow early warning report for the current zone within the early warning area is generated.

[0006] According to the present invention, a debris flow early warning method based on multi-model fusion is provided, wherein the basic geographic geological data includes digital elevation model (DEM) topographic data, gully bottom elevation, loose deposit particle size and density, watershed area, average gully gradient, and road disaster-bearing body safety tolerance parameters. The real-time rainfall monitoring data includes rainfall intensity, rainfall duration, cumulative rainfall, maximum short-duration rainfall intensity, and previous rainfall. The real-time runoff monitoring data includes runoff depth, flow velocity, peak flow, and initial runoff. The auxiliary monitoring data includes pore water pressure and temperature and humidity data of the trench bed.

[0007] According to the present invention, a debris flow early warning method based on multi-model fusion is provided, wherein the shallow water flow numerical model is constructed based on the two-dimensional shallow water wave equation; The machine learning model uses the moving least squares (MLS) model, with rainfall, average rainfall intensity, short-duration maximum rainfall intensity, and initial runoff as input factors, to predict the runoff variation coefficient. The neural network model includes a generalized regression neural network (GRNN) for monthly trend forecasting and a backpropagation (BP) neural network for nonlinear fitting. The GRNN uses the forecasted pre-monthly and inter-monthly rainfall and runoff data as training samples, and outputs monthly runoff forecasts after optimizing the smoothing factor. The weighting coefficients of the weighted fusion are determined by least squares optimization to minimize the expected sum of squared errors between the fused predicted values ​​and the measured values.

[0008] According to the debris flow early warning method provided by the present invention, the construction of dual critical thresholds includes constructing a critical runoff index, a critical rainfall threshold, and a model and control logic for differentiated adjustment; Constructing the critical runoff index includes: determining one or more safety thresholds among roadbed flow depth, retaining wall impact resistance strength, and culvert flow capacity based on the disaster-bearing body's safety tolerance parameters; Constructing the critical rainfall threshold includes: combining the Takahashi critical flow depth model and the Tognacca critical flow model to back-calculate the safe rainfall threshold corresponding to different safety levels; The differential adjustment includes: for steep sections where the average gradient of the gullies is greater than a first threshold, lowering the basic critical rainfall threshold by a first percentage; for gentle sections where the average gradient of the gullies is less than a second threshold, raising the basic critical rainfall threshold by a second percentage; and for roadbed sections, correcting the velocity or depth threshold based on their roughness characteristics.

[0009] The debris flow early warning method based on multi-model fusion provided by the present invention further includes: The pore water pressure data monitored by the ditch bed piezometer is obtained, and the comprehensive runoff prediction result is corrected by combining the temperature and humidity parameters with the pore water pressure data. And / or, based on the degree to which the comprehensive runoff forecast exceeds the critical runoff index and the status of real-time rainfall parameters exceeding the critical rainfall threshold, an early warning level system is constructed.

[0010] According to a multi-model fusion debris flow early warning method provided by the present invention, the step of generating a debris flow early warning report for the current partition within the early warning area includes: By comparing the comprehensive runoff prediction results with the hierarchical and regional early warning threshold database, the probability of disaster occurrence, disaster type, disaster intensity, risk level, and the impact range of the highway disaster-bearing body are determined, and the debris flow early warning report is generated. Information on the location, risk level, and impact range of the disaster is extracted from the debris flow early warning report and rendered into a 3D BIM model of a small watershed along the highway to generate a 3D visualized risk map.

[0011] The debris flow early warning method based on multi-model fusion provided by the present invention further includes: The parameters of the shallow water flow numerical model, the machine learning model, and the neural network model are regularly calibrated and updated using the latest rainfall-runoff monitoring data, and the graded and zoned early warning threshold library is supplemented and corrected using newly added disaster cases or environmental change data.

[0012] The present invention also provides a debris flow early warning device based on multi-model fusion, comprising: The data acquisition module is used to acquire basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the warning area; The calculation module is used to drive a shallow water flow numerical model to simulate and obtain a first runoff prediction result based on the geographical and geological basic data and the real-time rainfall monitoring data, and to predict and obtain a second runoff prediction result and a third runoff prediction result respectively through a machine learning model and a neural network model based on the real-time rainfall monitoring data and the real-time runoff monitoring data. The fusion module is used to perform weighted fusion of the first runoff prediction result, the second runoff prediction result, and the third runoff prediction result using the least squares method to obtain a comprehensive runoff prediction result; The threshold module is used to construct a dual critical threshold based on the safety requirements of the disaster-bearing bodies within the warning area, and to back-calculate the corresponding critical rainfall threshold based on the dual critical threshold and the comprehensive runoff prediction results. The optimization module is used to differentiate the critical rainfall threshold by combining the average gradient of the gullies in different zones within the warning area, thereby forming a hierarchical and zoned warning threshold library. The report generation module is used to generate a debris flow early warning report for the current zone within the early warning area based on the hierarchical and zonal early warning threshold library and the comprehensive runoff prediction results.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a program or instructions stored in the memory and executable on the processor, wherein when the processor executes the program or instructions, it implements the steps of the multi-model fusion debris flow early warning method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a program or instructions stored thereon, wherein when the program or instructions are executed by a computer, the steps of the debris flow early warning method of multi-model fusion as described above are implemented.

[0015] The present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the computer is able to execute the multi-model fusion debris flow early warning method as described above.

[0016] This invention also provides a multi-model fusion debris flow early warning system, comprising: The data acquisition and processing unit includes various geographic and geological data sensors, rainfall monitoring equipment, runoff monitoring equipment, and auxiliary data monitoring equipment, used to acquire and process basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the early warning area; The multi-model fusion computing unit integrates a shallow flow numerical model, a machine learning model, and a neural network model, and is configured to perform weighted fusion calculations to output comprehensive runoff prediction results. The intelligent threshold management unit is used to store and manage a hierarchical and zonal early warning threshold library determined based on the disaster-bearing body's safety parameters and the channel gradient zoning. The intelligent early warning analysis unit includes a data comparison module and a judgment module, which are used to generate a debris flow early warning report for the current zone within the early warning area based on the hierarchical and zoned early warning threshold library and the comprehensive runoff prediction results.

[0017] The debris flow early warning method, device, electronic equipment and system provided by this invention improves the accuracy and engineering relevance of early warning by integrating physical mechanisms and data patterns. It realizes refined early warning from "whether it will happen" to "when, where and what kind of impact", while reducing the error of a single model and adapting to different terrains and disaster-bearing bodies in small watersheds along highways. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments of this invention or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 One of the flowcharts for the multi-model fusion debris flow early warning method provided by the present invention; Figure 2 This is a flowchart illustrating the multi-model weight coupling process in the debris flow early warning method based on multi-model fusion provided by the present invention. Figure 3 The second schematic diagram of the multi-model fusion debris flow early warning method provided by the present invention; Figure 4 This is a schematic diagram showing the verification results of the debris flow early warning method for small watersheds based on the multi-model fusion method provided by the present invention. Figure 5 A schematic diagram of the structure of the debris flow early warning device with multi-model fusion provided by the present invention; Figure 6 A schematic diagram of the physical structure of the electronic device provided by the present invention; Figure 7 This is a schematic diagram of the structure of the debris flow early warning system based on multi-model fusion provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] This invention addresses the problems of unstable early warning accuracy and susceptibility to environmental interference in existing technologies. It acquires basic geographical and geological data of small watersheds along highways (including average gradient of channels and parameters of loose deposits), and real-time rainfall and runoff monitoring data (including pore water pressure in the gully bed). It integrates shallow water flow numerical models, machine learning models (moving least squares models), and neural network models (generalized regression neural networks) to construct a multi-dimensional calculation system. Combining critical flow depth (adapted to roadbed / retaining wall safety) and critical flow rate (adapted to culvert flow) as dual judgment criteria, it generates response feature vectors that fit the terrain differences (steep sections / gentle sections / roadbed sections), achieving accurate early warning of debris flow disaster types (such as medium-scale debris flows), intensity (flow depth / flow velocity level), and the impact range of the highway disaster-bearing body (roadbed / retaining wall / culvert area). By employing multi-model weighted coupling (with weights determined using the least squares method) and differentiated threshold adjustment (combined with channel gradient and gully bed coarsening layer characteristics for correction), the errors of single models and interference from gully bed hydrodynamic factors are eliminated. This solves the problems of poor generalization and insufficient accuracy of traditional early warning technologies. It can be widely applied in areas prone to debris flows, such as highways along typhoon-prone areas in South China and small watersheds in mountainous regions, providing scientific and efficient technical support for disaster prevention and control during highway construction and operation. The invention will be further described and illustrated below with reference to the accompanying drawings and through several specific embodiments.

[0022] Figure 1 This is one of the flowcharts illustrating the multi-model fusion debris flow early warning method provided by the present invention, such as... Figure 1 As shown, the method includes: S101 acquires basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the warning area.

[0023] This invention can be understood as follows: First, it acquires basic data related to debris flow early warning to provide data support for subsequent early warning calculations. Specifically, it can acquire basic geographical and geological data of small watersheds along highways (including average gradient of gullies, particle size and density of loose deposits, etc.), real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary data such as pore water pressure in gully beds, and comprehensively utilize these data to provide a complete input for the model.

[0024] Optionally, the basic geographic and geological data includes digital elevation model (DEM) topographic data, gully bottom elevation, loose sediment particle size and density, watershed area, average gully gradient, and road disaster-bearing body safety tolerance parameters; the real-time rainfall monitoring data includes rainfall intensity, rainfall duration, cumulative rainfall, short-duration maximum rainfall intensity, and previous rainfall; the real-time runoff monitoring data includes runoff depth, flow velocity, peak flow, and initial runoff; and the auxiliary monitoring data includes gully bed pore water pressure and temperature and humidity data.

[0025] This invention can be understood to include two main components in its data acquisition: basic data acquisition and real-time monitoring deployment and acquisition. Basic data acquisition primarily involves the collection of geographic and geological foundational data, including DEM topographic data and channel bottom elevation (z). ), loose sediment particle size (d m ) and density (ρ) m ), watershed area (A) and average gradient of gullies, and safety tolerance parameters of highway disaster-bearing bodies (limits of impact resistance of retaining walls, limits of flow depth of roadbed, and limits of flow capacity of bridges and culverts).

[0026] Specifically, LiDAR technology is used to acquire DEM data with a resolution of 1-5 meters, and the elevation of the bottom of the channel (z) is extracted. ); Conduct on-site investigation of loose sediments and determine particle size distribution (d). 50 d 16 d 88 ) and density (ρ) m =1.8~2.0t / m³, adapted to the characteristics of colluvial deposits); calculate the watershed area (A) and average gradient of gullies based on GIS; collect safety parameters of highway disaster-bearing bodies, such as the limit of roadbed flow depth, the limit of retaining wall impact resistance strength, and the limit of culvert flow capacity.

[0027] Real-time monitoring deployment is mainly for the real-time collection of rainfall and runoff monitoring data. Real-time rainfall data includes rainfall intensity (I), duration (D), cumulative rainfall (P), and maximum 30-minute rainfall intensity (I). 30 ), maximum 10-minute rainfall intensity (I 10 ), and the preceding 24-hour rainfall. Real-time runoff data includes runoff depth (h), flow velocity (u / v), peak flow (Qc), and initial runoff (Q). 初 Specifically, tipping bucket rain gauges (accuracy 0.2mm, measuring range 127cm / h) will be deployed within the watershed to record parameters such as rainfall intensity, duration, and cumulative rainfall. A triangular weir will be installed at the ditch outlet (corresponding to the highway section location), equipped with a water level recorder (placed inside a PVC pipe in a flow stabilization well) to collect runoff depth and velocity data. The flow rate will be calculated using the following formula: .

[0028] Optionally, in addition to the above deployment, temperature and humidity sensors and trench bed piezometers can also be deployed simultaneously to record the ambient temperature and humidity and the pore water pressure of the trench bed.

[0029] S102, based on the geographical and geological basic data and the real-time rainfall monitoring data, drive the shallow water flow numerical model to simulate and obtain the first runoff prediction result, and based on the real-time rainfall monitoring data and the real-time runoff monitoring data, respectively use the machine learning model and the neural network model to predict and obtain the second runoff prediction result and the third runoff prediction result.

[0030] This invention can be understood as follows: After acquiring basic data, it first uses geographical and geological data and real-time rainfall data to simulate runoff dynamic parameters through a shallow water flow numerical model, obtaining the calculation results referred to as the first runoff prediction result. Simultaneously, based on real-time rainfall and runoff data, it uses machine learning models and neural network models to predict key runoff parameters. That is, the prediction results obtained through machine learning models serve as the second runoff prediction result, and the prediction results obtained through neural network model calculations serve as the third runoff prediction result.

[0031] S103, using the least squares method, the first runoff prediction result, the second runoff prediction result, and the third runoff prediction result are weighted and fused to obtain a comprehensive runoff prediction result.

[0032] This can be understood as follows: based on the first, second, and third runoff prediction results obtained through the above steps, the results of multiple models are weighted, coupled, and fused using the least squares method. The resulting overall budget result is called the comprehensive runoff prediction result.

[0033] S104. Based on the safety requirements of the disaster-bearing bodies within the warning area, a dual critical threshold is constructed, and based on the dual critical threshold and the comprehensive runoff prediction results, the corresponding critical rainfall threshold is obtained by reverse calculation.

[0034] This invention can be understood as follows: In order to provide quantitative early warning of debris flows in the warning area according to safety, this invention constructs a dual critical threshold system including real-time rainfall and actual runoff based on the safety requirements of the carrier in the warning area. Based on this pre-constructed system, and combined with the comprehensive runoff prediction results obtained from the above steps, a back-calculation is performed to obtain the corresponding safe critical rainfall threshold.

[0035] S105, Based on the average gradient of the channels in different zones within the warning area, the critical rainfall threshold is adjusted differentially to form a graded and zoned warning threshold library.

[0036] This invention can be understood as follows: In order to make the prediction results closer to the actual situation of different zones within the warning area and improve the prediction accuracy, the present invention calculates the average gradient of the channel within the warning area for different zones, and makes differentiated adjustments to the threshold of each zone based on the average gradient. That is, the threshold is adjusted according to the average gradient of the channel (steep section, gentle section) to form a graded warning threshold library for each zone.

[0037] S106, Based on the hierarchical and zonal early warning threshold library and the comprehensive runoff prediction results, generate a debris flow early warning report for the current zone within the early warning area.

[0038] It can be understood that, in the final stage of this invention, the comprehensive runoff prediction results are compared with the early warning threshold to generate a debris flow early warning report that is adapted to the disaster-bearing bodies of highways (roadbed, retaining wall, culvert) and includes the disaster type, intensity, impact range and risk level.

[0039] Optionally, generating a debris flow early warning report for the current zone within the early warning area includes: comparing the comprehensive runoff prediction results with the graded zone early warning threshold library to determine the probability of disaster occurrence, disaster type, disaster intensity, risk level, and the impact range of the highway disaster-bearing body, and generating the debris flow early warning report; extracting disaster location, risk level, and impact range information from the debris flow early warning report, rendering it onto a three-dimensional BIM model of a small watershed along the highway, and generating a three-dimensional visualized risk map.

[0040] In essence, when generating a debris flow early warning report, this invention first compares the comprehensive runoff prediction results with dual critical thresholds to determine the probability of disaster occurrence, disaster type (e.g., medium-scale debris flow), intensity (flow depth, velocity level), and the impact range of the road-bearing bodies (roadbed, retaining walls, culvert areas). Based on this, information on the disaster location, risk level, and impact range can be extracted from the early warning report and rendered into a 3D BIM model of the small watershed along the highway to generate a 3D visualized risk map. Optionally, this invention can also simultaneously include the basis for the early warning threshold, the model calculation process, and historical verification records in the early warning report, providing complete decision support for highway debris flow prevention and control.

[0041] Optionally, when generating an early warning report, the present invention can also generate a visualized early warning report based on the early warning level, disaster intensity, and impact range information, and render it into a three-dimensional geographic information model of the early warning area to form a three-dimensional risk map.

[0042] The debris flow early warning method provided by this invention, through the construction of a multi-model fusion and multi-indicator linkage, and adaptability to the scenario of "small watersheds along highways and typhoon-prone areas", can effectively improve the accuracy of early warning and the engineering targeting, and realize refined early warning from "whether it will occur" to "when, where and what kind of impact", while reducing the error of a single model, adapting to different terrains and highway disaster-bearing bodies in small watersheds along highways, and improving the early warning accuracy and applicability to highway engineering for low-frequency, small-scale, cluster debris flows.

[0043] In the debris flow early warning method provided by the above embodiments, the shallow water flow numerical model is optionally constructed based on the two-dimensional shallow water wave equation; the machine learning model adopts the moving least squares (MLS) model, using rainfall, average rainfall intensity, short-duration maximum rainfall intensity, and initial runoff as input factors to predict the runoff variation coefficient; the neural network model includes a generalized regression neural network (GRNN) for monthly trend forecasting and a backpropagation backpropagation (BP) neural network for nonlinear fitting, wherein the GRNN uses the predicted pre-monthly and inter-monthly rainfall and runoff data as training samples, and outputs the monthly runoff forecast after optimizing the smoothing factor; the weight coefficients of the weighted fusion are determined by least squares optimization to minimize the expected sum of squared errors between the fused predicted value and the measured value.

[0044] This can be understood as meaning that the prediction models used in this invention can be pre-built based on historical data, or they can be temporarily built and trained using collected data. As described in the above embodiments, the construction of the main calculation and prediction models used in this invention is as follows: First, construct a numerical model of shallow water flow: Set the input parameters: DEM topographic data, rainfall intensity (e.g., 100-year return period rainfall I=0.83mm / min), and loose deposit friction parameters (Manning coefficient ntd: 0.04~0.06 for gully sections, considering the erosion resistance of the gully bed coarsening layer; 0.07~0.08 for roadbed sections; yield stress τ=45~120Pa).

[0045] Based on the two-dimensional shallow flow equation (SWE), DEM topographic data and real-time rainfall data are input to simulate the spatiotemporal distribution of runoff depth and velocity, and output parameters such as flow depth range, peak velocity, and peak flow rate. The governing equations are as follows: ; Momentum equation (x / y direction, momentum conservation): ; ; In the formula: h is the depth of the debris flow (m), which is directly related to the siltation depth of the highway and the inundation range of the roadbed; for The direction of debris flow velocity (m / s) determines the impact strength of the roadbed and the flow pressure of bridges and culverts, reflecting the spatial expansion and temporal evolution of runoff within a small watershed. The elevation of the bottom of the ditch (m) is obtained based on LiDAR data and affects the runoff confluence path; g is the acceleration due to gravity (9.8 m / s²). for The directional friction term (m / s2) reflects the resistance effect of the small watershed topography (such as gully slope and roughness) on runoff movement.

[0046] The simulation outputs include runoff depth distribution (e.g., runoff depth of 0.15~0.8m for a 200-year return period rainfall), velocity range (2.8~5.5m / s), and peak flow (Qc) under different rainfall scenarios, providing physical process support for early warning.

[0047] Secondly, construct a machine learning model (MLS): Set input factors: Select rainfall (P), average rainfall intensity (Iave), and maximum 30-minute rainfall intensity (I). 30 ), maximum 10-minute rainfall intensity (I 10 ) and initial runoff (Q 初 There are a total of 5 impact factors.

[0048] A moving least squares (MLS) model was used, with rainfall (P), average rainfall intensity (Iave), and maximum 30-minute rainfall intensity (I) as the parameters. 30 ), maximum 10-minute rainfall intensity (I 10 ) and initial runoff (Q 初 Using Q as the input factor, the predicted runoff variation coefficient (μ=Q) is calculated. 峰 / P), introducing the concept of tight support to capture minute changes in parameters.

[0049] Model training: Using 25 rainfall events around the Guangle Expressway as samples (18 training events and 7 validation events, including data during Typhoon Utor), the model was implemented through MATLAB programming, and the runoff variation coefficient (μ) was output. The results showed that the maximum relative error was 36%, which is better than the traditional model.

[0050] Next, construct the neural network model: This includes Generalized Regressive Neural Networks (GRNNs) and Backpropagation Neural Networks (BPNNs). GRNNs use pre-monthly and inter-monthly rainfall and runoff data as training samples, and output monthly runoff forecasts after optimizing the smoothing factor (σ). BPNNs optimize model weights to improve nonlinear fitting capabilities. Specifically: The GRNN model uses monthly rainfall and runoff from the 12 months preceding the forecast month as training samples (adapted to the rainfall patterns of the typhoon season in South China). It employs cross-validation to optimize the smoothing factor (σ=0.25~0.30, improving the accuracy of extreme rainfall response) and quickly outputs monthly runoff forecasts. The training speed is 30% faster than that of the BP neural network.

[0051] A backpropagation (BP) neural network optimizes the input layer (5 neurons), hidden layer (12 neurons, adapted to the nonlinear characteristics of runoff in small watersheds), and output layer (1 neuron: Q). 峰 The gradient descent method is used to adjust the weights, thereby improving the adaptability to typhoon-induced heavy rainfall events.

[0052] Finally, establish model coupling: With the objective of minimizing the expected value of the forecast error variance, the weights of each model are solved by weighted fusion of the results from each model using the least squares method. For example... Figure 2 The diagram shown is a flowchart illustrating the multi-model weight coupling in the debris flow early warning method based on multi-model fusion provided by the present invention, and is based on the following equation: .

[0053] The weights of each model are determined by the principle of prioritization. (Shallow water flow model) (MLS model) and (GRNN model) to obtain comprehensive runoff prediction values. .

[0054] For example, in the case of the G2 ditch on the Guangle Expressway, ω a =0.45、ω m =0.35, ω9=0.20, combined runoff prediction value The prediction error is reduced by 15% to 20% after coupling compared to the single model.

[0055] This invention integrates shallow water flow numerical calculation, moving least squares (MLS) machine learning, and generalized regression neural network (GRNN) to conduct runoff-rainfall coupled debris flow early warning. It is applicable to disaster prediction and risk prevention in debris flow-prone areas such as highways in typhoon-prone areas and small watersheds in mountainous areas. It can specifically solve the problem of early warning of clustered, low-frequency, small-scale debris flows during highway construction and operation.

[0056] Optionally, the debris flow early warning method based on multi-model fusion provided in the above embodiments may include the following: The construction of dual critical thresholds includes constructing a critical runoff index, a critical rainfall threshold, and a model and control logic for differentiated adjustment. Constructing the critical runoff index includes determining one or more safety thresholds for roadbed flow depth, retaining wall impact resistance strength, and culvert flow capacity based on the safety tolerance parameters of the disaster-bearing body. Constructing the critical rainfall threshold includes combining the Takahashi critical flow depth model and the Tognacca critical flow model to inversely deduce the safe rainfall thresholds corresponding to different safety levels. Constructing the differentiated adjustment includes lowering the basic critical rainfall threshold by a first proportion for steep sections where the average gradient of the gully is greater than a first threshold; raising the basic critical rainfall threshold by a second proportion for gentle sections where the average gradient of the gully is less than a second threshold; and correcting the velocity or flow depth thresholds for highway roadbed sections based on their roughness characteristics.

[0057] It can be understood that, based on the above embodiments, the dual critical threshold of the present invention includes a critical runoff index and a critical rainfall threshold. Therefore, when constructing the threshold system, the construction of the dual critical threshold may include the following steps: Critical runoff indices: Based on highway structural safety and disaster-bearing body safety tolerance parameters, and verified through shallow water flow model simulation, the following criteria were determined: roadbed flow depth ≤ 1.0m (corresponding to flow velocity ≤ 5.5m / s), and retaining wall impact resistance strength ≤ 1.2×10⁻⁶. 4 N / m², culvert flow capacity ≥15m³ / s (1.2 times the design value).

[0058] Back-calculation of critical rainfall threshold: combining the Takahashi critical flow depth model , And, the Tognacca critical flow model ( ), By reverse calculation, the rainfall thresholds corresponding to different levels can be determined.

[0059] Taking the G2 ditch of the Guangle Expressway as an example, the shallow water flow model shows that the runoff Q = 0.65 m³ / s corresponds to a roadbed flow depth of 1.0 m. Combined with measured rainfall data, the rainfall conditions are inferred, such as: the orange warning threshold is 1 hour rainfall intensity ≥ 40 mm and 24 hours ≥ 200 mm (50-year return period rainfall standard), and the red warning threshold is 1 hour rainfall intensity ≥ 50 mm and 24 hours ≥ 250 mm (100-year return period rainfall standard).

[0060] Based on the established critical runoff index and critical rainfall threshold, the following differentiated threshold adjustments are made to further improve the prediction accuracy for different regions: ① Steep sections (average gradient of the gully > 300‰, corresponding slope ≈ 16.7°): Shallow flow simulation shows that the flow depth is 20% higher than that of the gentle section under the same rainfall, so the warning threshold is lowered by 15% (orange warning with 1-hour rainfall intensity ≥ 34mm). ② Gentle section (average gradient of the gully <300‰, corresponding slope <15°): lower flow depth, threshold increased by 10% (1-hour rainfall intensity ≥44mm under orange alert). ③ Highway subgrade section: Considering the high roughness of the subgrade (ntd = 0.07~0.08) and the characteristics of the ditch bed coarsening layer, the flow velocity threshold (≤5.0m / s) of the highway subgrade section is modified in combination with the characteristics of the ditch bed coarsening layer to ensure that it is compatible with the actual project.

[0061] Furthermore, the debris flow early warning method based on multi-model fusion provided in the above embodiments may optionally include: acquiring pore water pressure data monitored by a ditch bed piezometer, and correcting the comprehensive runoff prediction result by combining temperature and humidity parameters with the pore water pressure data; and / or, constructing an early warning level system based on the degree to which the comprehensive runoff prediction value exceeds the critical runoff index and the condition that the real-time rainfall parameters exceed the critical rainfall threshold.

[0062] In order to further reduce errors caused by pore water pressure and improve accuracy, this invention can also obtain pore water pressure data monitored by the ditch bed piezometer, and combine it with temperature and humidity parameters to correct the comprehensive runoff prediction results and eliminate interference from ditch bed hydrodynamic factors.

[0063] Furthermore, this invention can also construct an early warning level system based on the comparison of real-time runoff and real-time rainfall with critical runoff and critical rainfall thresholds: divided into four levels: no risk (flow depth < 0.05m), low risk (0.05m ≤ flow depth < 0.10m), medium risk (0.10m ≤ flow depth < 0.15m), and high risk (flow depth ≥ 0.15m), specifically: No risk: Flow depth < 0.05m (runoff parameters < 80% of critical value), no possibility of debris flow; Low risk: 0.05m ≤ flow depth < 0.10m (80% ≤ runoff parameter < critical value), small-scale debris flow may occur, with the impact limited to the gully; Medium risk: 0.10m ≤ flow depth < 0.15m (critical value ≤ runoff parameter < 120% critical value), medium-scale debris flow is highly likely, and roadbed and culvert protection needs to be activated; High risk: Flow depth ≥ 0.15m (runoff parameter ≥ 120% critical value), large-scale debris flow is very likely to occur, and construction personnel and equipment need to be evacuated urgently.

[0064] The disaster location (e.g., section K35+200 of the Guangle Expressway), risk level, and impact range are extracted from the early warning report and rendered into a 3D BIM model of a small watershed along the highway, generating an interactive 3D visualized risk map. Users can view disaster details by rotating and zooming, and clicking on a risk area will link to the original monitoring data (rainfall, runoff) and the model calculation process.

[0065] Furthermore, the debris flow early warning method based on the multi-model fusion provided in the above embodiments may optionally include: periodically calibrating and updating the parameters of the shallow flow numerical model, the machine learning model, and the neural network model using the latest rainfall-runoff monitoring data, and supplementing and correcting the graded and zoned early warning threshold library using newly added disaster cases or environmental change data.

[0066] This can be understood as follows: to further improve the model's generalization ability and accuracy, this invention continuously updates the model by constantly expanding the sample size during operation. Specifically, this includes the following processing steps: Real-time calibration: Quarterly on-site monitoring data is collected (e.g., the measured peak flow of 0.85 m³ / s at the triangular weir corresponds to a 1-hour rainfall intensity of 52 mm), and the threshold is fine-tuned to the appropriate value (e.g., the 1-hour rainfall intensity for an orange alert is fine-tuned to 38 mm). That is, the model parameters (Manning coefficient for shallow flow model, GRNN smoothing factor, and MLS basis function) are updated by combining the measured data from the mud level gauge and rain gauge.

[0067] Model Updates: When new monitoring sections or changes in terrain conditions are added, the shallow flow model is iteratively simulated again to supplement the threshold library and ensure long-term stability of early warning accuracy. For example, when new monitoring channels or changes in channel material sources are added along highways, the shallow flow model is iteratively simulated again to supplement the differentiated threshold library. Environmental correction: The comprehensive runoff prediction results are corrected by using temperature and humidity parameters and pore water pressure data of the ditch bed to eliminate interference from hydrodynamic factors of the ditch bed and improve the stability of the early warning.

[0068] Based on the above embodiments, this invention integrates the physical simulation capabilities of shallow flow models, the nonlinear fitting advantages of machine learning, and the rapid forecasting characteristics of neural networks through five major stages: data acquisition, multi-model calculation, threshold construction, early warning output, and dynamic calibration. It combines dual critical thresholds (critical runoff + critical rainfall) and differentiated adjustment strategies to achieve a full-chain early warning of debris flows, from "whether it occurs" to "intensity, range, and risk level." The core lies in the weighted coupling of multiple models and dynamic calibration, which reduces the error of a single model and adapts to the different terrains and disaster-bearing structures (roadbed, retaining walls, culverts) along highways.

[0069] To further illustrate the technical solution of the present invention, the following description takes the typical debris flow gully G2 of the T5 section of the Guangle Expressway in Lechang City, Guangdong Province as an example, but does not limit the scope of protection of the present invention.

[0070] The G2 debris flow gully has a drainage area of ​​1.18 km², a channel length of 1.975 km, an elevation range of 320.0–972.5 m, a height difference of 652.5 m, an average gradient of 330‰, and a corresponding slope of approximately 18.2°. This gully is a typical example of a debris flow induced by Typhoon Utor (August 15–17, 2013), with a runoff volume of 3.5 × 10⁻⁶ m. 4 m³, causing structural damage to the roadbed retaining wall and construction access road. For example Figure 3 The diagram shown is a second flowchart of the multi-model fusion debris flow early warning method provided by the present invention, including the following specific implementation steps for debris flow early warning of debris flow gully G2: Step 1: Data collection, including basic data collection and real-time data collection.

[0071] (1) Basic data collection ①Topographic and geological data: 5m resolution DEM data was acquired using LiDAR technology, and the elevation z at the bottom of the G2 ditch was extracted. =320.0~972.5m, watershed shape coefficient 30.3% (elongated confluence characteristics). Field investigation showed that the loose deposits were mainly Quaternary colluvial deposits (Q4col+del), with a grain size distribution of d 16 =15mm, d 50 =32mm, d 88 =68mm, calculate the average particle size d m =(15+32+68) / 3=38.3mm, density of the aggregate ρ m =1.9t / m³. The parameters of the disaster-bearing body are based on the design standards of the Guangle Expressway: roadbed flow depth limit ≤1.0m, retaining wall impact resistance strength ≤1.2×10⁻⁶. 4 N / m², culvert flow capacity ≥15m³ / s.

[0072] ② Monitoring equipment deployment: A tipping bucket rain gauge (American Onset RGM-3 model, accuracy 0.2mm) is deployed in the upstream catchment area of ​​G2 ditch to record rainfall intensity, duration and cumulative rainfall; a 90° triangular weir is set up at the ditch outlet (corresponding to K35+200 of the highway), equipped with a HOBO automatic water level recorder (placed in the PVC pipe of the stabilizing well, recording at 10-minute intervals, accuracy 0.8mm), and temperature and humidity sensors (monitoring range -20~60℃, 0~100% RH) and piezometers (monitoring pore water pressure in the ditch bed) are deployed simultaneously.

[0073] (2) Real-time data acquisition ①Rainfall data: From 20:00 on August 15 to 11:00 on August 17, 2013, the rain gauge station upstream of G2 ditch recorded a cumulative rainfall of 306.2 mm, with a maximum 1-hour rainfall intensity of 52 mm (14:00 on August 16), a maximum 30-minute rainfall intensity of 28 mm, a maximum 10-minute rainfall intensity of 15 mm, and a previous 24-hour rainfall of 220 mm, which meets the "once-in-a-century rainfall standard" (the maximum 3-day rainfall in the past century is 300 mm) recorded in the literature.

[0074] ② Runoff Data: Water level recorder monitoring showed that the peak runoff occurred at 14:30 on August 16th, with a water height of h=280mm at the triangular weir. The peak flow rate Qc=0.85m³ / s was calculated using the formula Q=1.343×(h / 1000)2.47. The runoff process exhibited a "sharp rise and fall" characteristic, with the rise lasting approximately 45 minutes and the receding approximately 3 hours. The initial runoff rate Qc was... 初 =0.05m³ / s.

[0075] Step 2 involves training and calibrating the models, including training and calibrating the shallow water flow numerical model, machine learning model, and neural network model.

[0076] (1) Construction of training samples Historical rainfall-runoff data from 2010 to 2012 for G2 gully and surrounding gullies G1 and G3-G5 (a total of 25 rainfall events, including 3 rainstorm events) were selected. Of these, 18 events were used for model training and 7 for validation. The training samples included rainfall parameters (P, Iave, I...). 30 I 10 ), runoff parameters (Q) 峰 Q 初 The data includes the measured outflow volume, sourced from the Guangdong Provincial Hydrology and Water Resources Bureau and the Guangle Expressway construction monitoring archives.

[0077] (2) Multi-model parameter optimization ① Shallow Flow Numerical Model: Input DEM topographic data and the rainfall time series of Typhoon Utor, set Manning coefficient n=0.05 (channel section, considering the scour resistance characteristics of the coarsening layer of the channel bed) and roadbed section n=0.07, time step Δt=1s (satisfying CFL conditions), grid size 5m×5m; simulated output of G2 ditch flow depth distribution: upstream catchment area flow depth 0.05~0.10m, midstream main channel flow depth 0.10~0.15m, downstream roadbed section flow depth 0.15~0.20m, simulated peak flow rate 0.82m³ / s, relative error of 3.5% with measured value 0.85m³ / s.

[0078] ②MLS model: with P (306.2mm), Iave (2.1mm / min), I 30 (0.93 mm / min), I10 (1.5mm / min), Q 初 (0.05 m³ / s) is used as the input factor, and a tight support radius of 0.3 is introduced. The predicted runoff variation coefficient is μ=Q. 峰 / P=0.85 / 306.2≈0.0028, with a relative error of 3.7% compared to the measured value μ=0.0027.

[0079] ③ GRNN Model: Using monthly rainfall (mean 153.7 mm) and monthly runoff (mean 0.23 m³ / s) from 2010 to 2012 as training samples, the smoothing factor σ=0.30 was optimized using cross-validation. The NSE efficiency coefficient between the monthly runoff forecast and the measured values ​​reached 0.78. A BP neural network was used to optimize the model weights, with 5 neurons in the input layer, 12 neurons in the hidden layer, and 1 neuron in the output layer. After training, the runoff peak prediction error was ≤8%.

[0080] (3) Determination of model coupling weights The weights are solved using the least squares method, and the objective function is... Minimize and calculate the shallow water flow model weights ω a =0.45, MLS model weights ω m =0.35, GRNN model weights ω9=0.20, comprehensive runoff prediction value =0.45×0.82+0.35×0.86+0.20×0.83=0.84m³ / s, with a relative error of 1.2% compared to the measured value of 0.85m³ / s.

[0081] Step 3: Determine and verify the early warning threshold.

[0082] (1) Construction of dual critical thresholds ① Critical runoff index: Based on the measured damage data of G2 ditch, when the flow depth is ≥0.15m (corresponding to the flow velocity ≥2.8m / s), the roadbed retaining wall will be damaged by lateral scouring; combined with shallow water flow simulation, the critical flow depth h0=0.15m is determined, and the corresponding critical flow rate Qcr=0.65m³ / s.

[0083] ② Critical rainfall threshold: Based on the Takahashi critical flow depth model: , Substitution =0.812, σ=2.65g / cm³, ρ=1.0g / cm³, φ=32° (internal friction angle of loose aggregate), θ=18.2°, d m =38.3mm, calculated as follows =0.145m, and the rainfall thresholds are calculated as follows: 1 hour rainfall intensity ≥ 50mm and 24 hours ≥ 250mm (orange alert), 1 hour rainfall intensity ≥ 60mm and 24 hours ≥ 300mm (red alert).

[0084] ③ Differentiated adjustment: The average gradient of G2 ditch is 330‰ (>300‰), which is a steep section. The warning threshold is lowered by 15%. After the adjustment, the orange warning threshold is ≥42.5mm of rainfall in 1 hour and ≥212.5mm in 24 hours. Due to the high roughness of the roadbed section (n=0.07), the threshold is raised by 30%. The orange warning threshold is ≥65mm of rainfall in 1 hour and ≥325mm in 24 hours.

[0085] (2) Early warning verification At 2:00 PM on August 16, 2013, real-time monitoring data for G2 Gully reached the orange alert threshold (1-hour rainfall intensity 52 mm, 24-hour rainfall 280 mm), with a comprehensive runoff forecast of 0.84 m³ / s (>critical flow 0.65 m³ / s). The system issued an "orange alert," indicating that "a medium-scale debris flow will occur in G2 Gully, with a depth of 0.15~0.20 m, affecting the roadbed and construction access road from K35+150 to K35+250." The actual debris flow occurred at 2:30 PM, with a depth of 0.18 m and a runoff volume of 3.5 × 10⁻⁶ m³ / s. 4 m³, 15m of roadbed retaining wall was damaged and 1 culvert was blocked, which is 100% consistent with the early warning results.

[0086] Step 4: Visualize the output and apply it after the disaster.

[0087] The location of the G2 ditch disaster (corresponding to the K35+200 section of the Guangle Expressway), risk level (orange warning), runoff depth distribution (upstream 0.05~0.10m, midstream 0.10~0.15m, downstream roadbed section 0.15~0.20m), and potential impact range (within 100m along the ditch, including roadbed retaining walls, culverts, and temporary construction areas) are rendered onto the 3D digital model of the northern section of the Guangle Expressway (built based on BIM technology, including precise geometric parameters of engineering structures such as roadbed, retaining walls, and culverts) using GIS spatial mapping technology, generating an interactive 3D visualized risk map.

[0088] This map supports multi-view operation (rotation, zoom, and sectioning). Clicking on any risk area will display a pop-up window showing the core early warning basis, including: ① real-time monitoring data (1-hour rainfall intensity 52mm, 24-hour rainfall 280mm); ② model calculation results (comprehensive runoff prediction value 0.84m³ / s, critical flow 0.65m³ / s); ③ disaster impact prediction (retaining wall erosion risk, culvert blockage probability). Simultaneously, the map marks "key protection areas" (roadbed retaining wall section K35+180-K35+220, culvert at K35+200) and "emergency response suggestions" (close culvert inlet in advance, relocate construction materials around the retaining wall), providing precise prevention and control decision support for on-site personnel.

[0089] like Figure 4 The diagram shown illustrates the verification results of the debris flow early warning method for small watersheds based on the multi-model fusion method provided by the present invention. Experiments demonstrate that the present invention has at least the following beneficial effects and advantages: Firstly, accuracy is improved. After multi-model fusion, the runoff prediction error is reduced by 15% to 20%. For scenarios similar to the Guangle Expressway, such as "coarsened gully bed + typhoon heavy rainfall", the matching degree between the early warning threshold and the actual disaster is over 94% (16 out of 17 gullies were backtested and found to be consistent), effectively reducing false alarms and missed alarms of low-frequency small-scale debris flows.

[0090] Secondly, it is highly practical. Based on the safety requirements of highway disaster-bearing structures (roadbed, retaining walls, culverts), it outputs disaster types (such as medium-sized debris flows), intensity (flow depth 0.10~0.15m, flow velocity 2.8~5.5m / s), highway impact range (such as the K35+150-K35+250 section), and risk levels, providing specific guidance for highway construction emergency rescue and equipment relocation. Third, it has wide adaptability. Based on the differential threshold adjustment of gully gradient, it is suitable for different scenarios such as steep sections (>300‰), gentle sections (<300‰) and roadbed sections. It can be extended to small watersheds along highways in typhoon-prone areas of South China and mountainous areas of Southwest China. It is especially suitable for low-frequency debris flow early warning in gully beds with coarsening layers. Fourth, it can achieve dynamic optimization. By combining quarterly measured data from triangular weirs and piezometers for calibration, along with corrections for gully bed hydrodynamic factors, the model ensures long-term stability under scenarios of "changes in material sources and minor adjustments to topography," thus addressing the impact of typhoon rainfall intensity changes and gully erosion. Fifth, it offers convenient visualization. The 3D BIM visualization risk map of small watersheds along the highway can intuitively mark the roadbed mileage and culvert location affected by disasters, making it easier for highway management and technical personnel to quickly formulate prevention and control plans based on "key protection areas + evacuation routes".

[0091] This invention is not only applicable to the highway scenario described in the embodiment, but can also be applied to early warning of other debris flow risk areas such as railway lines, mining areas, and areas surrounding villages and towns by replacing the safety parameters of the disaster-bearing body (such as the scour resistance of bridge piers and the safety elevation of residential areas) and adjusting the model weights and threshold system.

[0092] It should be noted that the scope of protection of this invention is not limited to the aforementioned embodiment of the small watershed in the northern section of the Guangle Expressway. Any modifications, equivalent substitutions, or scenario expansions based on the core technical solution of this invention—"runoff-rainfall monitoring + multi-model fusion + differentiated thresholds"—should be included within the scope of protection of this invention. For example, within the Guangle Expressway, model weights and threshold parameters can be adjusted according to the characteristics of different sections of the gully (e.g., the gradient of G4 gully is 422‰, and the watershed area of ​​G2 gully is 1.18 km²) to achieve section-specific early warnings. When expanding outwards, this method can be applied to other debris flow-prone areas such as along railway lines and around mining areas. By replacing the critical parameters of the disaster-bearing bodies (e.g., the flow capacity of railway bridges and the impact resistance of mining area drainage ditches) and optimizing the model weight allocation, it can adapt to the early warning needs of different scenarios. Furthermore, based on the technical framework of this invention, replacing the monitoring equipment with higher-precision equipment such as radar rain gauges and radar water level gauges, or introducing new algorithms such as random forests into model coupling, all fall within the scope of protection of this invention.

[0093] Based on the same inventive concept, this invention also provides a multi-model fusion debris flow early warning device according to the above embodiments. This device is used to detect the safety risks of containers in the above embodiments. Therefore, the descriptions and definitions in the multi-model fusion debris flow early warning method of the above embodiments can be used to understand the various execution modules in this invention. For details, please refer to the above method embodiments, which will not be repeated here.

[0094] According to an embodiment of the present invention, the structure of the multi-model fusion debris flow early warning device is as follows: Figure 5 The diagram shown is a structural schematic of the multi-model fusion debris flow early warning device provided by the present invention. This device can be used to implement multi-model fusion debris flow early warning in the above-described method embodiments. The device includes: a data acquisition module 501, a calculation module 502, a fusion module 503, a threshold module 504, an optimization module 505, and a report generation module 506. Wherein: The data acquisition module 501 is used to acquire basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the early warning area; The calculation module 502 is used to drive a shallow water flow numerical model to simulate and obtain a first runoff prediction result based on the geographical and geological basic data and the real-time rainfall monitoring data, and to predict and obtain a second runoff prediction result and a third runoff prediction result based on the real-time rainfall monitoring data and the real-time runoff monitoring data, respectively, through a machine learning model and a neural network model. The fusion module 503 is used to perform weighted fusion of the first runoff prediction result, the second runoff prediction result and the third runoff prediction result using the least squares method to obtain a comprehensive runoff prediction result; The threshold module 504 is used to construct a dual critical threshold based on the safety requirements of the disaster-bearing bodies within the warning area, and to back-calculate the corresponding critical rainfall threshold based on the dual critical threshold and the comprehensive runoff prediction results. The optimization module 505 is used to differentiate the critical rainfall threshold by combining the average gradient of the gullies in different zones within the warning area, thereby forming a hierarchical and zoned warning threshold library. The report generation module 506 is used to generate a debris flow early warning report for the current zone within the early warning area based on the hierarchical and zonal early warning threshold library and the comprehensive runoff prediction results.

[0095] The debris flow early warning device provided by this invention integrates physical mechanisms and data patterns, thereby improving the accuracy and engineering relevance of early warning. It achieves refined early warning from "whether it will happen" to "when, where, and what kind of impact," while reducing the error of a single model and adapting to different terrains and disaster-bearing bodies in small watersheds along highways.

[0096] It is understood that the relevant program modules in the apparatus of the above embodiments can be implemented by a hardware processor in this invention. Furthermore, the multi-model fusion debris flow early warning device of this invention, utilizing the above program modules, can implement the multi-model fusion debris flow early warning process of the above method embodiments. When used to implement the multi-model fusion debris flow early warning in the above method embodiments, the beneficial effects produced by the device of this invention are the same as those in the corresponding above method embodiments, and can be referred to the above method embodiments, which will not be repeated here.

[0097] As another aspect of the present invention, the present invention also provides an electronic device according to the above embodiments, the electronic device including a memory, a processor and a program or instructions stored in the memory and executable on the processor, wherein when the processor executes the program or instructions, it implements the steps of the multi-model fusion debris flow early warning method as described in the above embodiments.

[0098] Furthermore, the electronic device of the present invention may also include a communication interface and a bus. (See reference) Figure 6 The present invention provides a schematic diagram of the structure of an electronic device, including: at least one memory 601, at least one processor 602, a communication interface 603, and a bus 604.

[0099] The memory 601, processor 602, and communication interface 603 communicate with each other via bus 604. The communication interface 603 is used for information transmission between the electronic device and the data acquisition device. The memory 601 stores a program or instruction that can be run on the processor 602. When the processor 602 executes the program or instruction, it implements the steps of the multi-model fusion debris flow early warning method as described in the above embodiments.

[0100] This electronic device can be understood to include at least a memory 601, a processor 602, a communication interface 603, and a bus 604. The memory 601, processor 602, and communication interface 603 are interconnected via the bus 604, enabling communication between them. For example, the processor 602 can read program instructions for a multi-model fusion debris flow early warning method from the memory 601. Furthermore, the communication interface 603 can also establish a communication connection between the electronic device and a data acquisition device, facilitating information transmission. For instance, the communication interface 603 can be used to read geographical and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the warning area.

[0101] When the electronic device is running, the processor 602 calls the program instructions in the memory 601 to execute the methods provided in the above-described method embodiments.

[0102] When the program instructions in the aforementioned memory 601 can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Alternatively, all or part of the steps of the above method embodiments can be implemented by hardware related to the program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0103] The present invention also provides a non-transitory computer-readable storage medium according to the above embodiments, wherein a program or instructions are stored thereon, which, when executed by a computer, implement the steps of the multi-model fusion debris flow early warning method as described in the above embodiments.

[0104] As another aspect of the present invention, this embodiment also provides a computer program product according to the above embodiments. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer is able to execute the multi-model fusion debris flow early warning method provided in the above method embodiments.

[0105] The electronic device, non-transitory computer-readable storage medium, and computer program product provided by the present invention integrate the underlying mechanisms and data patterns by executing the steps of the multi-model fusion debris flow early warning method described in the above embodiments, thereby improving the accuracy and engineering relevance of the early warning. It achieves refined early warning from "whether it will occur" to "when, where, and what kind of impact," while reducing the error of a single model and adapting to different terrains and disaster-bearing bodies in small watersheds along highways.

[0106] As another aspect of the present invention, the present invention also provides a multi-model fusion debris flow early warning system according to the above embodiments, such as... Figure 7 The diagram shown is a schematic representation of the structure of the multi-model fusion debris flow early warning system provided by the present invention, including: The data acquisition and processing unit includes various geographic and geological data sensors, rainfall monitoring equipment, runoff monitoring equipment, and auxiliary data monitoring equipment, used to acquire and process basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the early warning area; The multi-model fusion computing unit integrates a shallow flow numerical model, a machine learning model, and a neural network model, and is configured to perform weighted fusion calculations to output comprehensive runoff prediction results. The intelligent threshold management unit is used to store and manage a hierarchical and zonal early warning threshold library determined based on the disaster-bearing body's safety parameters and the channel gradient zoning. The intelligent early warning analysis unit includes a data comparison module and a judgment module, which are used to generate a debris flow early warning report for the current zone within the early warning area based on the hierarchical and zoned early warning threshold library and the comprehensive runoff prediction results.

[0107] This embodiment can be understood as a system structure embodiment with the same inventive concept as the above-described method embodiments. The system structurally includes at least the above-described functional units. The data acquisition and processing unit is mainly used to drive multi-source data, specifically acquiring basic geographic and geological data of small watersheds along the highway (including gully gradient and loose sediment parameters), real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary data such as gully bed pore water pressure. It comprehensively utilizes these basic geographic and geological data, high spatiotemporal resolution real-time rainfall and runoff monitoring data, and auxiliary data such as gully bed pore water pressure to provide comprehensive input for the model.

[0108] The multi-model fusion computing unit is used for multi-model fusion calculations. Specifically, it calculates runoff parameters using a shallow flow numerical model, a machine learning model, and a neural network model respectively, and then combines them with a weighted least squares method to obtain a comprehensive runoff prediction result. That is, by coupling the shallow flow numerical model (physical model), the moving least squares machine learning model (MLS), and the generalized regression neural network model (GRNN), the least squares method is used to determine the optimal weights of each model, achieving complementary advantages and improving the comprehensive prediction accuracy of key runoff parameters (such as peak flow).

[0109] The intelligent threshold management unit is used to construct dual critical thresholds. Specifically, from the perspective of highway engineering safety, it establishes critical runoff indicators based on the safety tolerance parameters of the disaster-bearing body (such as the safe flow depth of the roadbed, the impact resistance strength of the retaining wall, and the minimum flow capacity of the culvert), and then uses these to inversely deduce the corresponding critical rainfall threshold, so that the early warning standard is directly linked to engineering risks.

[0110] The intelligent early warning analysis unit compares the comprehensive runoff prediction results with the early warning thresholds to generate an early warning report adapted to the highway's disaster-bearing body, including the disaster type, intensity, impact range, and risk level. Optionally, differentiated threshold adjustments can also be made. Specifically, based on the average gradient of the gully, the early warning area is divided into steep sections, gentle sections, and roadbed sections, and the basic early warning threshold is differentiated and corrected to make the early warning more consistent with the actual terrain and hydrodynamic conditions.

[0111] Optionally, the system of the present invention may further include a dynamic calibration and visualization output unit, mainly used to establish a dynamic update mechanism for model parameters and early warning thresholds, continuously optimizing them using subsequent monitoring data, that is, updating model parameters and differentiated threshold libraries based on measured data to improve the long-term stability of early warnings. Simultaneously, the early warning results (disaster location, type, intensity, range, risk level, etc.) can be rendered onto a 3D BIM model of a small watershed along the highway to generate an interactive, visualized risk map, supporting efficient decision-making.

[0112] This invention effectively integrates physical mechanisms and data patterns, elevating early warning from "whether it occurs" to a more refined level of "when and where, with what intensity, and which targets are affected," significantly improving the accuracy, relevance, and practicality of early warnings. At the same time, it reduces the error of a single model and is adaptable to different terrains and disaster-bearing bodies in small watersheds along highways.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A debris flow early warning method based on multi-model fusion, characterized in that, include: Obtain basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the early warning area; Based on the aforementioned geographical and geological data and the aforementioned real-time rainfall monitoring data, a shallow water flow numerical model is driven to simulate and obtain the first runoff prediction result. Based on the aforementioned real-time rainfall monitoring data and the aforementioned real-time runoff monitoring data, a second runoff prediction result and a third runoff prediction result are obtained respectively through a machine learning model and a neural network model. The least squares method is used to weight and fuse the first runoff prediction result, the second runoff prediction result, and the third runoff prediction result to obtain a comprehensive runoff prediction result; Based on the safety requirements of disaster-bearing bodies within the warning area, a dual critical threshold is constructed, and based on the dual critical threshold and the comprehensive runoff prediction results, the corresponding critical rainfall threshold is obtained by reverse calculation. By combining the average gradient of gullies in different zones within the warning area, the critical rainfall threshold is adjusted differentially to form a hierarchical and zoned warning threshold library; Based on the hierarchical and zonal early warning threshold library and the comprehensive runoff prediction results, a debris flow early warning report for the current zone within the early warning area is generated.

2. The debris flow early warning method based on multi-model fusion according to claim 1, characterized in that, The basic geographic and geological data include digital elevation model (DEM) topographic data, gully bottom elevation, loose sediment particle size and density, watershed area, average gully gradient, and highway disaster-bearing body safety tolerance parameters. The real-time rainfall monitoring data includes rainfall intensity, rainfall duration, cumulative rainfall, maximum short-duration rainfall intensity, and previous rainfall. The real-time runoff monitoring data includes runoff depth, flow velocity, peak flow, and initial runoff. The auxiliary monitoring data includes pore water pressure and temperature and humidity data of the trench bed.

3. The debris flow early warning method based on multi-model fusion according to claim 1 or 2, characterized in that, The shallow water flow numerical model is constructed based on the two-dimensional shallow water wave equation; The machine learning model uses the moving least squares (MLS) model, with rainfall, average rainfall intensity, short-duration maximum rainfall intensity, and initial runoff as input factors, to predict the runoff variation coefficient. The neural network model includes a generalized regression neural network (GRNN) for monthly trend forecasting and a backpropagation (BP) neural network for nonlinear fitting. The GRNN uses the forecasted pre-monthly and inter-monthly rainfall and runoff data as training samples, and outputs monthly runoff forecasts after optimizing the smoothing factor. The weighting coefficients of the weighted fusion are determined by least squares optimization to minimize the expected sum of squared errors between the fused predicted values ​​and the measured values.

4. The debris flow early warning method based on multi-model fusion according to claim 1 or 2, characterized in that, The construction of dual critical thresholds includes constructing a critical runoff index, a critical rainfall threshold, and a model and control logic for differentiated adjustment; Constructing the critical runoff index includes: determining one or more safety thresholds among roadbed flow depth, retaining wall impact resistance strength, and culvert flow capacity based on the disaster-bearing body's safety tolerance parameters; Constructing the critical rainfall threshold includes: combining the Takahashi critical flow depth model and the Tognacca critical flow model to back-calculate the safe rainfall threshold corresponding to different safety levels; The differential adjustment includes: for steep sections where the average gradient of the gullies is greater than a first threshold, lowering the basic critical rainfall threshold by a first percentage; for gentle sections where the average gradient of the gullies is less than a second threshold, raising the basic critical rainfall threshold by a second percentage; and for roadbed sections, correcting the velocity or depth threshold based on their roughness characteristics.

5. The debris flow early warning method based on multi-model fusion according to claim 4, characterized in that, Also includes: The pore water pressure data monitored by the ditch bed piezometer is obtained, and the comprehensive runoff prediction result is corrected by combining the temperature and humidity parameters with the pore water pressure data. And / or, based on the degree to which the comprehensive runoff forecast exceeds the critical runoff index and the status of real-time rainfall parameters exceeding the critical rainfall threshold, an early warning level system is constructed.

6. The debris flow early warning method based on multi-model fusion according to claim 1, characterized in that, The process of generating a debris flow early warning report for the current zone within the early warning area includes: By comparing the comprehensive runoff prediction results with the hierarchical and regional early warning threshold database, the probability of disaster occurrence, disaster type, disaster intensity, risk level, and the impact range of the highway disaster-bearing body are determined, and the debris flow early warning report is generated. Information on the location, risk level, and impact range of the disaster is extracted from the debris flow early warning report and rendered into a 3D BIM model of a small watershed along the highway to generate a 3D visualized risk map.

7. The debris flow early warning method based on multi-model fusion according to claim 1, characterized in that, Also includes: The parameters of the shallow water flow numerical model, the machine learning model, and the neural network model are regularly calibrated and updated using the latest rainfall-runoff monitoring data, and the graded and zoned early warning threshold library is supplemented and corrected using newly added disaster cases or environmental change data.

8. A debris flow early warning device with multi-model fusion, characterized in that, include: The data acquisition module is used to acquire basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the warning area; The calculation module is used to drive a shallow water flow numerical model to simulate and obtain a first runoff prediction result based on the geographical and geological basic data and the real-time rainfall monitoring data, and to predict and obtain a second runoff prediction result and a third runoff prediction result respectively through a machine learning model and a neural network model based on the real-time rainfall monitoring data and the real-time runoff monitoring data. The fusion module is used to perform weighted fusion of the first runoff prediction result, the second runoff prediction result, and the third runoff prediction result using the least squares method to obtain a comprehensive runoff prediction result; The threshold module is used to construct a dual critical threshold based on the safety requirements of the disaster-bearing bodies within the warning area, and to back-calculate the corresponding critical rainfall threshold based on the dual critical threshold and the comprehensive runoff prediction results. The optimization module is used to differentiate the critical rainfall threshold by combining the average gradient of the gullies in different zones within the warning area, thereby forming a hierarchical and zoned warning threshold library. The report generation module is used to generate a debris flow early warning report for the current zone within the early warning area based on the hierarchical and zonal early warning threshold library and the comprehensive runoff prediction results.

9. An electronic device comprising a memory, a processor, and a program or instructions stored in the memory and executable on the processor, characterized in that, When the processor executes the program or instructions, it implements the steps of the debris flow early warning method based on multi-model fusion as described in any one of claims 1 to 7.

10. A debris flow early warning system based on multi-model fusion, characterized in that, include: The data acquisition and processing unit includes various geographic and geological data sensors, rainfall monitoring equipment, runoff monitoring equipment, and auxiliary data monitoring equipment, used to acquire and process basic geographic and geological data, real-time rainfall monitoring data, real-time runoff monitoring data, and auxiliary monitoring data for the early warning area; The multi-model fusion computing unit integrates a shallow flow numerical model, a machine learning model, and a neural network model, and is configured to perform weighted fusion calculations to output comprehensive runoff prediction results. The intelligent threshold management unit is used to store and manage a hierarchical and zonal early warning threshold library determined based on the disaster-bearing body's safety parameters and the channel gradient zoning. The intelligent early warning analysis unit includes a data comparison module and a judgment module, which are used to generate a debris flow early warning report for the current zone within the early warning area based on the hierarchical and zoned early warning threshold library and the comprehensive runoff prediction results.

Images