Mountain torrent disaster intelligent monitoring and early warning system based on Internet of Things

By introducing a dynamic risk factor calculation module and an IoT sensor network, and combining it with an AI model to dynamically adjust the warning threshold, the problem of existing systems being unable to respond to dynamic risk changes has been solved, achieving efficient and accurate flash flood warnings and self-optimization capabilities.

CN121034054APending Publication Date: 2025-11-28SICHUAN UNIV
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202511219078.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-08-28
Publication Date
2025-11-28

AI Technical Summary

Technical Problem

Existing flash flood disaster early warning systems are unable to effectively detect and respond to dynamic risk changes caused by human engineering activities and fluid characteristics, resulting in biased early warning results and insufficient timeliness and accuracy.

Method used

A dynamic risk factor calculation module is introduced to access external engineering activity data and water and sediment content data from the monitoring layer in real time. These factors are quantified as correction parameters through a predefined rule base and physical dynamics model. Combined with IoT sensor network and AI model, flash flood risk is predicted, and early warning thresholds and decisions are dynamically adjusted.

Benefits of technology

It significantly improves the timeliness and accuracy of early warning, enhances the system's adaptability to complex environments and the robustness of prediction results, and realizes a closed-loop evolution from monitoring to self-diagnosis and self-learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121034054A_ABST
    Figure CN121034054A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of mountain torrent disaster early warning, in particular to an intelligent mountain torrent disaster monitoring and early warning system based on the Internet of Things, which comprises a sensing layer, a network layer, a platform layer and an application layer, the platform layer specifically comprises a data processing module, a mountain torrent risk prediction module, an early warning decision module, a dynamic risk factor calculation module and a disaster simulation and optimization module; compared with the prior art that a risk prediction model mostly depends on static historical data and is difficult to respond to dynamic risks caused by human activities and fluid characteristic changes, the scheme innovatively introduces a dynamic risk factor calculation module to quantify the human engineering activities and the water and sand content into fusible correction parameters in real time; therefore, the prediction model has the capability of sensing and adapting to the dynamic change of the risk, and the timeliness and accuracy of early warning are remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mountain flood disaster early warning, and in particular to a mountain flood disaster intelligent monitoring and early warning system based on the Internet of Things. BACKGROUND

[0002] Mountain flood disaster early warning is an important research direction in the field of disaster prevention and reduction, and its core lies in accurately predicting the possibility of disaster occurrence through real-time monitoring data. Traditional early warning systems mainly rely on basic hydrological and meteorological data such as rainfall and water level, and combine historical statistical rules or mathematical models to make risk judgments.

[0003] However, most existing risk prediction models are based on static historical data, and their warning logic and thresholds are usually fixed. This static model has significant defects, as it cannot effectively perceive and respond to dynamic risk changes caused by human engineering activities (such as road construction, house building, and water conservancy projects) and fluid characteristics (such as the rapid change of sediment content in water bodies). For example, a rainfall of the same intensity will have a completely different risk probability of inducing a mountain flood in an area with good vegetation coverage compared to an area that has just experienced excavation and dumping of earth, or river blockage. Due to the lack of quantitative and integration mechanisms for these dynamic factors, existing technologies are prone to deviations in their warning results when faced with complex and changing external environments, and their timeliness and accuracy are constrained.

[0004] To solve the above problems, the present application innovatively introduces a dynamic risk factor calculation module. This module can access external engineering activity data and real-time water and sediment content data obtained from the monitoring layer in real time, and according to a predefined rule base and physical dynamics model, it quantifies human activities and water and sediment effects into standardized correction parameters. These dynamic parameters are transmitted in real time to the mountain flood risk prediction engine, allowing it to dynamically adjust the prediction calculation, thereby enabling the entire early warning system to perceive and adapt to dynamic changes in risk, significantly improving the timeliness and accuracy of the warning. SUMMARY

[0005] To overcome the problems presented in the background art, the present application proposes a mountain flood disaster intelligent monitoring and early warning system based on the Internet of Things.

[0006] The technical solution of the present application is: a mountain flood disaster intelligent monitoring and early warning system based on the Internet of Things, comprising: a perception layer composed of multiple groups of intelligent sensing nodes deployed in the monitoring area, for collecting environmental data; a network layer in communication connection with the perception layer, for transmitting the environmental data to the platform layer; a platform layer for processing the environmental data transmitted by the network layer in combination with stored historical data to obtain a mountain flood probability, and outputting a warning instruction according to the mountain flood probability; The application layer is used to receive early warning instructions issued by the platform layer and to publish information.

[0007] As a preferred option, the platform layer specifically includes: A11: Data processing module, used to preprocess the raw environmental data collected by the perception layer, including cleaning, fusion and formatting. A12: Flash flood risk prediction module, which uses a built-in AI model to calculate and output the probability value of flash floods occurring in the future based on preprocessed data and dynamic risk factors. A13: Early warning decision module, used to receive probability values ​​output by the risk prediction engine, automatically judge and generate early warning information of different levels according to preset threshold rules, and trigger the release process; A14: Dynamic risk factor calculation module, used to quantify the dynamic impact of human activities and fluid characteristics on risk, and transform them into corrected parameters that can be understood and integrated by the prediction engine; A15: Disaster Simulation and Optimization Module, used to review the entire disaster process after a disaster using digital twin technology, conduct comparative analysis to locate system defects, and automatically generate targeted engineering or monitoring optimization solutions.

[0008] As a preferred embodiment, the intelligent sensing nodes in the perception layer include the following devices: A21: Rain gauge sensor for real-time monitoring of rainfall intensity and cumulative rainfall; A22: Water level sensor, used to monitor water level changes in water bodies including rivers and reservoirs in real time; A23: Video sensor used to provide real-time visual footage of the site; A24: Soil moisture sensor used to monitor soil moisture content; A25: Turbidity meter and laser sediment concentration sensor, used to monitor the sediment content in water bodies.

[0009] Preferably, when the data processing module performs preprocessing on the raw environmental data collected by the perception layer, including cleaning, fusion, and formatting, it specifically includes: S11: Receive raw environmental data packets uploaded by each sensor node in the perception layer; S12: Parse and decode the original data packet to extract the physical measurement values ​​and the corresponding timestamps and device ID information; S13: Perform data cleaning on the extracted physical measurement values, including removing outliers that are clearly outside the reasonable range and filtering out impulse noise caused by momentary sensor failures; S14: Perform time alignment and formatting on the cleaned data, and convert it into a preset standard data format; S15: Fusion processing of data from different sensor types but at the same spatiotemporal reference to generate multidimensional environmental data records with unified timestamps and location identifiers; S16: Perform quality assessment on the fused data records and mark them with quality labels, then output the standardized high-quality data stream to the flash flood risk prediction module and the dynamic risk factor calculation module.

[0010] As a preferred embodiment, the dynamic risk factor calculation configuration is as follows: External data representing the type and scale of human engineering activities are accessed, and these are quantified into benign or malignant impact factors according to a predefined rule base; real-time water and sediment content data collected by the perception layer are accessed, and the enhancement effect on water flow impact force is calculated according to the water and sediment dynamics model. The rule base for quantifying human engineering activities into impact factors is defined as follows: benign impact factors with negative values ​​correspond to slope stabilization, afforestation, and diversion channel construction projects, used to reduce the probability of flash flood risk; malignant impact factors with positive values ​​correspond to slope cutting for housing construction, roadbed excavation, and dumping soil to block river channels, used to increase the probability of flash flood risk.

[0011] Preferably, the dynamic risk factor calculation module, in quantifying the dynamic impact of human activities and fluid characteristics on risk and transforming it into corrective parameters that can be understood and integrated by the prediction engine, specifically includes: S21: Data access step, receiving in real time structured engineering activity data from an external government engineering approval database, as well as real-time water and sediment content monitoring data streams from the turbidity meter and laser sediment concentration sensor in the sensing layer; S22: Engineering impact quantification steps, analyze engineering activity data, extract key information such as engineering type, geographical coordinates, construction scale and impact radius; based on a predefined engineering impact rule library, classify engineering types as benign or malignant, and calculate impact factor values ​​according to construction scale and impact radius, where benign projects generate negative impact factors and malignant projects generate positive impact factors. S23: Steps for calculating the water and sediment effect: analyze the water and sediment content monitoring data, and use the water and sediment dynamic function model to calculate the water and sediment enhancement effect coefficient; S24: Parameter standardization step, normalizes and formats the human activity impact factor and water and sediment enhancement effect coefficient to generate a standardized dynamic correction parameter set; S25: Parameter output and fusion steps, which transmit the standardized dynamic correction parameter set to the flash flood risk prediction engine in real time.

[0012] As a preferred option, the hydrodynamic function model, when in operation, specifically includes: S31: Call the baseline risk curve, which retrieves the baseline water level-risk relationship curve based on historical clean water data from the built-in database; S32: Based on the current water and sediment content data, the sediment influence coefficient is calculated using the Lorenz curve model; S33: The basic risk value corresponding to each water level in the benchmark water level-risk relationship curve is weighted and superimposed with the sediment influence coefficient to generate a dynamic water level-risk relationship curve. The principle formula is as follows: ; in, This represents the dynamic risk value of the dynamic water level-risk relationship curve. The base risk value for the benchmark water level-risk relationship curve. For predefined weighting coefficients, The sediment influence coefficient; S34: Input the current real-time water level data into the dynamic water level-risk relationship curve to calculate the comprehensive risk value at the current moment after correction for water and sediment content; S35: Automatically adjust the system's early warning trigger threshold based on the dynamic water level-risk relationship curve.

[0013] As a preferred embodiment, the flash flood risk prediction module, based on preprocessed data and dynamic risk factors, uses a built-in AI model to calculate and output the probability value of flash floods occurring within a future period. Specifically, this includes: S41: Synchronously receive the standardized environmental data stream from the data processing module and the standardized dynamic correction parameter set from the dynamic risk factor calculation module; S42: The standardized environmental data stream and the dynamically corrected parameter set are spatiotemporally aligned and feature-stitched, and a unified input feature vector containing static environmental features and dynamic influence features is generated through a weighted fusion algorithm; S43: Input the unified input feature vector in parallel into the physical information neural network model and the hydrological mechanism model. The physical information neural network model outputs data-driven probability predictions, and the hydrological mechanism model outputs calculated probability values ​​based on physical equations. S44: Based on the current data quality indicators and model confidence, dynamically assign weights to the probability values ​​output by the two models, and use an adaptive weighted average algorithm to generate preliminary ensemble probability values; S45: Based on the dynamic influence factor in the unified input feature vector, the initial integration probability value is nonlinearly corrected; S46: Repeat steps S43-S45 at multiple future time points to generate time-series prediction results containing the probability values ​​of flash floods at different forecast periods, and transmit the time-series prediction results to the early warning decision module.

[0014] As a preferred approach, when performing nonlinear correction on the initial integrated probability value based on the dynamic influencing factors in the unified input feature vector, the final flash flood probability value is generated by introducing human activity influencing factors as probability multiplier factors and water and sediment enhancement effect values ​​as probability exponent correction terms. The principle formula is as follows: ; in, This represents the final probability value of a flash flood. To initially integrate probability values, To correct the weight parameters, As a factor influencing human activities, This is the scaling factor. This represents the water and sediment enhancement effect value.

[0015] As a preferred embodiment, the disaster simulation and optimization module, when in operation, specifically includes: S51: After a disaster occurs, the system automatically retrieves and integrates high-precision pre-disaster terrain data, multi-source monitoring data throughout the disaster process, and external environmental data to construct a complete data chain for the disaster process. S52: Based on the complete data chain, the hydrodynamic model and the geological disaster model are driven to perform parallel calculations in the digital twin engine to simulate and reproduce the evolution process of flash floods, the inundation range, the change of impact force and the occurrence process of secondary disasters; S53: Spatially and temporally overlay and compare the simulated disaster evolution process, inundation range, peak water level and disaster intensity with the actual recorded disaster information to identify the areas of difference and key points of divergence between the simulation results and the actual disaster. S54: Based on the aforementioned discrepancy regions and key divergence points, analyze the root causes of the discrepancies and identify the types of system defects, including: perception layer defects, model layer defects, engineering defense defects, and emergency response defects. S55: Based on the type of system defect located, access the pre-set optimization measures knowledge base, automatically match and generate targeted reinforcement solutions, including sensor network optimization deployment suggestions, model parameter calibration strategies, engineering structure reinforcement measures and emergency plan revision guidelines; S56: Output the generated enhancement plan to the system management platform in the form of a structured report, and automatically create optimization task work orders.

[0016] The beneficial effects of this invention are: 1. Compared with existing risk prediction models that rely heavily on static historical data and are difficult to respond to dynamic risks brought about by human activities and changes in fluid properties, this solution innovatively introduces a dynamic risk factor calculation module, which quantifies human engineering activities and water and sediment content in real time into fusionable correction parameters, enabling the prediction model to perceive and adapt to dynamic changes in risks, and significantly improving the timeliness and accuracy of early warning. 2. Compared with existing technologies that mostly use a single data-driven or mechanism model for prediction, which have insufficient generalization ability or poor physical interpretability in complex scenarios, this solution uses a physical information neural network and a hydrological mechanism model for parallel operation, and performs fusion correction through an adaptive weighting algorithm. It combines the flexibility of the data-driven model with the reliability of the mechanism model, thereby comprehensively improving the robustness and practical value of the prediction results. 3. Compared with existing technologies where the warning thresholds are mostly fixed values ​​and cannot be dynamically adjusted according to fluid characteristics such as sediment content, which may lead to false alarms or missed alarms, this solution dynamically adjusts the water level-risk relationship curve and warning thresholds through a hydrodynamic model, enabling the system to respond sensitively to changes in water sediment content, greatly enhancing the adaptability and scientific nature of early warning decisions under different geological and hydrological conditions. 4. Compared with the existing system which lacks an effective review and self-optimization mechanism after a disaster, this solution innovatively introduces digital twin technology, which can automatically review the entire disaster process, compare simulation with the actual situation, locate defects and generate optimization solutions, and promote the system to achieve closed-loop evolution from monitoring and early warning to self-diagnosis and self-learning, significantly improving the system's long-term vitality and intelligence level. 5. Compared with existing technologies that mostly use single or a small number of sensors for monitoring, resulting in data with limited dimensions and difficulty in fully reflecting the causes of disasters, this solution adopts a three-dimensional sensing network that integrates sensors for rainfall, water level, video, soil moisture, and sediment concentration. This enables the collaborative collection and fusion of multi-dimensional information such as meteorological, hydrological, geological, and fluid characteristics, greatly improving the comprehensiveness and reliability of disaster monitoring. Attached Figure Description

[0017] Figure 1 The diagram shown is a schematic representation of the structure of the Internet of Things-based intelligent monitoring and early warning system for flash floods according to the present invention. Figure 2 The diagram shown is a schematic representation of the platform layer structure in the Internet of Things-based intelligent monitoring and early warning system for flash floods of the present invention. Figure 3 The diagram illustrates the workflow of the dynamic risk factor calculation module in the IoT-based intelligent monitoring and early warning system for flash floods of this invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Please see Figure 1 - Figure 3 The present invention provides an embodiment of an intelligent monitoring and early warning system for flash floods based on the Internet of Things, comprising: The perception layer consists of multiple sets of intelligent sensor nodes deployed in the monitoring area, used to collect environmental data; The network layer, which is communicatively connected to the perception layer, is used to transmit the environmental data to the platform layer; The platform layer is used to process the environmental data transmitted by the network layer and combine it with the stored historical data to obtain the probability of flash floods, and output early warning commands based on the probability of flash floods. The application layer is used to receive early warning instructions issued by the platform layer and to publish information.

[0020] In this embodiment, the present invention constructs a four-layer Internet of Things architecture consisting of a perception layer, a network layer, a platform layer, and an application layer, thereby achieving closed-loop management of the entire chain from environmental data collection to early warning information dissemination. This significantly improves the systematicness, real-time performance, and automation level of flash flood monitoring, and enhances the overall effectiveness and response speed of disaster early warning.

[0021] As a preferred option, the platform layer specifically includes: A11: Data processing module, used to preprocess the raw environmental data collected by the perception layer, including cleaning, fusion and formatting. A12: Flash flood risk prediction module, which uses a built-in AI model to calculate and output the probability value of flash floods occurring in the future based on preprocessed data and dynamic risk factors. A13: Early warning decision module, used to receive probability values ​​output by the risk prediction engine, automatically judge and generate early warning information of different levels according to preset threshold rules, and trigger the release process; A14: Dynamic risk factor calculation module, used to quantify the dynamic impact of human activities and fluid characteristics on risk, and transform them into corrected parameters that can be understood and integrated by the prediction engine; A15: Disaster Simulation and Optimization Module, used to review the entire disaster process after a disaster using digital twin technology, conduct comparative analysis to locate system defects, and automatically generate targeted engineering or monitoring optimization solutions.

[0022] In this embodiment, the present invention constructs a modular platform layer that includes data processing, risk prediction, early warning decision-making, dynamic factor calculation, and disaster simulation. This enables intelligent fusion of multi-source data and AI-driven risk assessment, significantly improving the accuracy of flash flood prediction, the scientific nature of decision-making, and the system's self-optimization capabilities.

[0023] As a preferred embodiment, the intelligent sensing nodes in the perception layer include the following devices: A21: Rain gauge sensor for real-time monitoring of rainfall intensity and cumulative rainfall; A22: Water level sensor, used to monitor water level changes in water bodies including rivers and reservoirs in real time; A23: Video sensor used to provide real-time visual footage of the site; A24: Soil moisture sensor used to monitor soil moisture content; A25: Turbidity meter and laser sediment concentration sensor, used to monitor the sediment content in water bodies.

[0024] In this embodiment, the present invention integrates multiple types of sensors such as rainfall, water level, video, soil moisture and sediment concentration to construct an integrated air-space-ground three-dimensional sensing network, realizing comprehensive and coordinated monitoring of rainfall, water level, soil condition and water sediment content, significantly enhancing the multidimensionality and reliability of data acquisition.

[0025] Preferably, when the data processing module performs preprocessing on the raw environmental data collected by the perception layer, including cleaning, fusion, and formatting, it specifically includes: S11: Receive raw environmental data packets uploaded by each sensor node in the perception layer; S12: Parse and decode the original data packet to extract the physical measurement values ​​and the corresponding timestamps and device ID information; S13: Perform data cleaning on the extracted physical measurement values, including removing outliers that are clearly outside the reasonable range and filtering out impulse noise caused by momentary sensor failures; S14: Perform time alignment and formatting on the cleaned data, and convert it into a preset standard data format; S15: Fusion processing of data from different sensor types but at the same spatiotemporal reference to generate multidimensional environmental data records with unified timestamps and location identifiers; S16: Perform quality assessment on the fused data records and mark them with quality labels, then output the standardized high-quality data stream to the flash flood risk prediction module and the dynamic risk factor calculation module.

[0026] In this embodiment, the present invention effectively solves the problems of noise, heterogeneity, and spatiotemporal inconsistency in multi-source heterogeneous sensor data through a standardized preprocessing process of data parsing, cleaning, alignment, fusion, and quality assessment, providing high-quality, standardized, and highly reliable data input for subsequent risk prediction.

[0027] As a preferred embodiment, the dynamic risk factor calculation configuration is as follows: External data representing the type and scale of human engineering activities are accessed, and these are quantified into benign or malignant impact factors according to a predefined rule base; real-time water and sediment content data collected by the perception layer are accessed, and the enhancement effect on water flow impact force is calculated according to the water and sediment dynamics model. The rule base for quantifying human engineering activities into impact factors is defined as follows: benign impact factors with negative values ​​correspond to slope stabilization, afforestation, and diversion channel construction projects, used to reduce the probability of flash flood risk; malignant impact factors with positive values ​​correspond to slope cutting for housing construction, roadbed excavation, and dumping soil to block river channels, used to increase the probability of flash flood risk.

[0028] In this embodiment, the present invention quantifies human engineering activities into benign and malignant influencing factors according to their types, and introduces a hydrodynamic model to quantify changes in fluid characteristics, thereby achieving a precise quantification of the dynamic impact of flash flood risk and significantly improving the risk prediction model's comprehensive response capability to complex human and natural factors.

[0029] Preferably, the dynamic risk factor calculation module, in quantifying the dynamic impact of human activities and fluid characteristics on risk and transforming it into corrective parameters that can be understood and integrated by the prediction engine, specifically includes: S21: Data access step, receiving in real time structured engineering activity data from an external government engineering approval database, as well as real-time water and sediment content monitoring data streams from the turbidity meter and laser sediment concentration sensor in the sensing layer; S22: Engineering impact quantification steps, analyze engineering activity data, extract key information such as engineering type, geographical coordinates, construction scale and impact radius; based on a predefined engineering impact rule library, classify engineering types as benign or malignant, and calculate impact factor values ​​according to construction scale and impact radius, where benign projects generate negative impact factors and malignant projects generate positive impact factors. S23: Steps for calculating the water and sediment effect: analyze the water and sediment content monitoring data, and use the water and sediment dynamic function model to calculate the water and sediment enhancement effect coefficient; S24: Parameter standardization step, normalizes and formats the human activity impact factor and water and sediment enhancement effect coefficient to generate a standardized dynamic correction parameter set; S25: Parameter output and fusion steps, which transmit the standardized dynamic correction parameter set to the flash flood risk prediction engine in real time.

[0030] In this embodiment, the present invention achieves real-time analysis, quantification, and parameterization of human activities and water and sediment effects by accessing structured engineering approval data and real-time water and sediment data, combined with a rule base and dynamic model. This generates standardized dynamic correction parameters that can be fused, greatly enhancing the timeliness and adaptability of risk prediction.

[0031] As a preferred option, the hydrodynamic function model, when in operation, specifically includes: S31: Call the baseline risk curve, which retrieves the baseline water level-risk relationship curve based on historical clean water data from the built-in database; S32: Based on the current water and sediment content data, the sediment influence coefficient is calculated using the Lorenz curve model; S33: The basic risk value corresponding to each water level in the benchmark water level-risk relationship curve is weighted and superimposed with the sediment influence coefficient to generate a dynamic water level-risk relationship curve. The principle formula is as follows: ; in, This represents the dynamic risk value of the dynamic water level-risk relationship curve. The base risk value for the benchmark water level-risk relationship curve. For predefined weighting coefficients, The sediment influence coefficient; S34: Input the current real-time water level data into the dynamic water level-risk relationship curve to calculate the comprehensive risk value at the current moment after correction for water and sediment content; S35: Automatically adjust the system's early warning trigger threshold based on the dynamic water level-risk relationship curve.

[0032] In this embodiment, the present invention establishes a baseline risk curve and introduces the Lorenz model to quantify the impact of sediment, thereby achieving dynamic adjustment of the risk curve based on real-time water and sediment content and adaptive optimization of the threshold, which significantly improves the accuracy of risk assessment and the sensitivity of early warning under different sediment conditions.

[0033] As a preferred embodiment, the flash flood risk prediction module, based on preprocessed data and dynamic risk factors, uses a built-in AI model to calculate and output the probability value of flash floods occurring within a future period. Specifically, this includes: S41: Synchronously receive the standardized environmental data stream from the data processing module and the standardized dynamic correction parameter set from the dynamic risk factor calculation module; S42: The standardized environmental data stream and the dynamically corrected parameter set are spatiotemporally aligned and feature-stitched, and a unified input feature vector containing static environmental features and dynamic influence features is generated through a weighted fusion algorithm; S43: Input the unified input feature vector in parallel into the physical information neural network model and the hydrological mechanism model. The physical information neural network model outputs data-driven probability predictions, and the hydrological mechanism model outputs calculated probability values ​​based on physical equations. S44: Based on the current data quality indicators and model confidence, dynamically assign weights to the probability values ​​output by the two models, and use an adaptive weighted average algorithm to generate preliminary ensemble probability values; S45: Based on the dynamic influence factor in the unified input feature vector, the initial integration probability value is nonlinearly corrected; S46: Repeat steps S43-S45 at multiple future time points to generate time-series prediction results containing the probability values ​​of flash floods at different forecast periods, and transmit the time-series prediction results to the early warning decision module.

[0034] In this embodiment, the present invention utilizes physical information neural networks and hydrological mechanism models in parallel, and adopts dynamic weight integration and nonlinear correction strategies to achieve complementary advantages between data-driven and mechanism models, generating high-precision probabilistic time series predictions with multiple forecast periods, thus comprehensively improving the robustness, accuracy and practicality of the predictions.

[0035] As a preferred approach, when performing nonlinear correction on the initial integrated probability value based on the dynamic influencing factors in the unified input feature vector, the final flash flood probability value is generated by introducing human activity influencing factors as probability multiplier factors and water and sediment enhancement effect values ​​as probability exponent correction terms. The principle formula is as follows: ; in, This represents the final probability value of a flash flood. To initially integrate probability values, To correct the weight parameters, As a factor influencing human activities, This is the scaling factor. This represents the water and sediment enhancement effect value.

[0036] In this embodiment, the present invention constructs a probability correction formula that integrates human activity factors and water and sediment effect values, thereby realizing the quantitative and interpretable adjustment of dynamic risk factors on prediction results, significantly enhancing the physical interpretability of flash flood probability values ​​and their adaptability to complex real-world situations.

[0037] As a preferred embodiment, the disaster simulation and optimization module, when in operation, specifically includes: S51: After a disaster occurs, the system automatically retrieves and integrates high-precision pre-disaster terrain data, multi-source monitoring data throughout the disaster process, and external environmental data to construct a complete data chain for the disaster process. S52: Based on the complete data chain, the hydrodynamic model and the geological disaster model are driven to perform parallel calculations in the digital twin engine to simulate and reproduce the evolution process of flash floods, the inundation range, the change of impact force and the occurrence process of secondary disasters; S53: Spatially and temporally overlay and compare the simulated disaster evolution process, inundation range, peak water level and disaster intensity with the actual recorded disaster information to identify the areas of difference and key points of divergence between the simulation results and the actual disaster. S54: Based on the aforementioned discrepancy regions and key divergence points, analyze the root causes of the discrepancies and identify the types of system defects, including: perception layer defects, model layer defects, engineering defense defects, and emergency response defects. S55: Based on the type of system defect located, access the pre-set optimization measures knowledge base, automatically match and generate targeted reinforcement solutions, including sensor network optimization deployment suggestions, model parameter calibration strategies, engineering structure reinforcement measures and emergency plan revision guidelines; S56: Output the generated enhancement plan to the system management platform in the form of a structured report, and automatically create optimization task work orders.

[0038] In this embodiment, the present invention uses digital twin technology to drive the post-disaster process review and simulation comparison, realizes the automatic identification of system defects and the intelligent generation of optimization solutions, greatly improves the system's self-diagnosis, self-learning and continuous optimization capabilities, and promotes the iterative evolution of the monitoring and early warning system.

[0039] Example 1: Routine Monitoring and Early Warning Scenario: A mountainous river basin in a certain province, designated as a key national-level area for the prevention and control of flash floods, has deployed this IoT-based intelligent monitoring and early warning system for flash floods.

[0040] Specific implementation plan: Sensing Layer: At key locations within the watershed, such as riverbanks, steep slopes, and confluence areas, multiple sets of intelligent sensor nodes have been deployed. Each set of nodes is equipped with a rainfall sensor to monitor rainfall intensity in real time, a water level sensor to monitor sudden rises in river levels, a soil moisture sensor to monitor soil saturation, and a turbidity meter to monitor changes in sediment content in the water during the initial stages of flash floods. These sensor nodes work collaboratively in a self-organizing network to continuously collect environmental data.

[0041] Network layer: By integrating 4G / 5G public networks and LoRa long-range wireless communication technology, all environmental data collected by the perception layer is transmitted in real time and reliably to the cloud platform layer for processing.

[0042] Platform Layer: Upon receiving the data, the data processing module immediately initiates analysis and cleaning of the raw data, and merges time-aligned rainfall, water level, soil moisture, and turbidity data to form a high-quality, standardized data stream. Simultaneously, the dynamic risk factor calculation module accesses information from the local government's engineering project database, discovering a recent "slope stabilization and tree planting" project in the upstream of the basin that has been registered and is under construction. The module quantifies this as a benign negative impact factor based on a rule base. Simultaneously, this module also receives real-time data from the turbidity meter and calculates a slight water and sediment enhancement effect coefficient using a hydrodynamic function model. These standardized dynamic correction parameters are sent in real-time to the flash flood risk prediction module. This module performs feature concatenation between the preprocessed environmental data stream and the dynamic correction parameter set, inputting them in parallel into the physical information neural network model and the hydrological mechanism model. After dynamic weight allocation and nonlinear correction formula calculations, the final output flash flood probability values ​​for the next 1 hour, 2 hours, and 3 hours are 15%, 30%, and 45%, respectively. The early warning decision module receives the time-series prediction results and determines that there is a 30% probability that the threshold for "attention level" has been exceeded in the next 2 hours, and then automatically generates a blue early warning message.

[0043] Application Layer: Once the early warning command is triggered, warning information is disseminated to township officials and residents within the watershed through various channels, including the early warning platform, SMS, and broadcasts, reminding them to closely monitor rainfall and water levels and prepare for evacuation. This early warning, by taking into account the positive impact of slope stabilization projects, featured more accurate probability calculations, avoiding over-warning and achieving scientific and efficient disaster prevention and control.

[0044] Example 2: Responding to Unexpected Engineering Activities and Post-Disaster Post-Disaster Review Scenario: The same system underwent a more complex test in another rainstorm.

[0045] Specific implementation plan: Sensing layer: Sensors detected continuous heavy rainfall and a rapid rise in water levels. Data from the turbidity meter showed a sharp increase in the sediment content of the water within a short period.

[0046] Platform Layer: The data processing module handles various types of data. When accessing external engineering data, the dynamic risk factor calculation module detects unreported "slope-cutting for housing construction" (a serious activity) and "dumping soil to block the river channel" (another serious activity) near a river channel via satellite remote sensing. The module immediately quantifies these as two high-value positive malignant impact factors based on a rule base. Simultaneously, based on sharply increasing sediment content data, the hydrodynamic function model calculates an extremely high hydro-sediment enhancement effect coefficient. These strong risk correction parameters are urgently sent to the flash flood risk prediction module. After integrating severe meteorological and hydrological data and extremely high dynamic risk parameters, the module's AI model, through a nonlinear correction formula (where both the human activity impact factor η and the sediment enhancement effect value λ become very large), calculates that the probability of a flash flood in the next hour has sharply increased to 85%. The early warning decision module immediately triggers the highest-level red alert.

[0047] Application Layer: A piercing alarm sounded through the village-wide public address system, and strong reminders were sent to every resident via SMS and app push notifications. The government immediately activated its emergency response plan, organizing the mandatory evacuation of residents from danger zones. Shortly afterward, a raging flash flood carrying large amounts of mud, sand, and rocks arrived as expected, but thanks to the prior evacuation, major casualties were successfully avoided.

[0048] Following the disaster, the disaster simulation and optimization module automatically activated. It retrieved pre-disaster topographic data, full-process monitoring data, and external environmental data, recreating the entire flash flood evolution process within the digital twin engine. By comparing the simulated inundation range and peak water level with the actual disaster situation, it was found that the system's prediction of the impact force at a downstream bend had an error of approximately 10%. The reason was analyzed to be that the topography at that location had undergone minor changes due to recent engineering activities, and the model had not been updated in time. Based on this, the module located the defect and automatically generated solutions from the optimization measures knowledge base: 1. Add an impact force sensor at the downstream bend; 2. Update the local topographic data in the digital twin model. This structured report was submitted to management to guide targeted optimization of the system, thereby achieving more accurate predictions in the next disaster.

[0049] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. An intelligent monitoring and early warning system for flash floods based on the Internet of Things, characterized in that: include: The perception layer consists of multiple sets of intelligent sensor nodes deployed in the monitoring area, used to collect environmental data; The network layer, which is communicatively connected to the perception layer, is used to transmit the environmental data to the platform layer; The platform layer is used to process the environmental data transmitted by the network layer and combine it with the stored historical data to obtain the probability of flash floods, and output early warning commands based on the probability of flash floods. The application layer is used to receive early warning instructions issued by the platform layer and to publish information.

2. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 1, characterized in that: The platform layer specifically includes: A11: Data processing module, used to preprocess the raw environmental data collected by the perception layer, including cleaning, fusion and formatting. A12: Flash flood risk prediction module, which uses a built-in AI model to calculate and output the probability value of flash floods occurring in the future based on preprocessed data and dynamic risk factors. A13: Early warning decision module, used to receive probability values ​​output by the risk prediction engine, automatically judge and generate early warning information of different levels according to preset threshold rules, and trigger the release process; A14: Dynamic risk factor calculation module, used to quantify the dynamic impact of human activities and fluid characteristics on risk, and transform them into corrected parameters that can be understood and integrated by the prediction engine; A15: Disaster Simulation and Optimization Module, used to review the entire disaster process after a disaster using digital twin technology, conduct comparative analysis to locate system defects, and automatically generate targeted engineering and monitoring optimization solutions.

3. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 2, characterized in that: The intelligent sensing nodes in the perception layer include the following devices: A21: Rain gauge sensor for real-time monitoring of rainfall intensity and cumulative rainfall; A22: Water level sensor, used to monitor water level changes in water bodies including rivers and reservoirs in real time; A23: Video sensor used to provide real-time visual footage of the site; A24: Soil moisture sensor used to monitor soil moisture content; A25: Turbidity meter and laser sediment concentration sensor, used to monitor the sediment content in water bodies.

4. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 3, characterized in that: The data processing module performs preprocessing on the raw environmental data collected by the perception layer, including cleaning, fusion, and formatting. Specifically, this includes: S11: Receive raw environmental data packets uploaded by each sensor node in the perception layer; S12: Parse and decode the original data packet to extract the physical measurement values ​​and the corresponding timestamps and device ID information; S13: Perform data cleaning on the extracted physical measurement values; S14: Perform time alignment and formatting on the cleaned data; S15: Fusion processing of data from different sensor types but at the same spatiotemporal reference; S16: Perform quality assessment on the fused data records and mark them with quality labels, then output the standardized high-quality data stream to the flash flood risk prediction module and the dynamic risk factor calculation module.

5. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 4, characterized in that: The dynamic risk factor calculation configuration is as follows: External data representing the type and scale of human engineering activities are accessed and quantified into benign or malignant impact factors according to a predefined rule base; real-time water and sediment content data collected by the perception layer are accessed and their enhancement effect on water flow impact force is calculated according to the water and sediment dynamics model. The rule base for quantifying human engineering activities into impact factors is defined as follows: slope stabilization, afforestation, and diversion channel construction projects correspond to negative benign impact factors, used to reduce the probability of flash flood risk; slope cutting for housing construction, roadbed excavation, and dumping soil to block river channels correspond to positive malignant impact factors, used to increase the probability of flash flood risk.

6. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 5, characterized in that: The dynamic risk factor calculation module, in quantifying the dynamic impact of human activities and fluid characteristics on risk and transforming it into corrective parameters that can be understood and integrated by the prediction engine, specifically includes: S21: Data access step, receiving in real time structured engineering activity data from an external government engineering approval database, as well as real-time water and sediment content monitoring data streams from the turbidity meter and laser sediment concentration sensor in the sensing layer; S22: Engineering impact quantification steps, analyze engineering activity data, extract key information such as engineering type, geographical coordinates, construction scale and impact radius; based on a predefined engineering impact rule library, classify engineering types as benign or malignant, and calculate impact factor values ​​based on construction scale and impact radius; S23: Steps for calculating the water and sediment effect: analyze the water and sediment content monitoring data, and use the water and sediment dynamic function model to calculate the water and sediment enhancement effect coefficient; S24: Parameter standardization step, normalizes and formats the human activity impact factor and water and sediment enhancement effect coefficient to generate a standardized dynamic correction parameter set; S25: Parameter output and fusion steps, which transmit the standardized dynamic correction parameter set to the flash flood risk prediction engine in real time.

7. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 6, characterized in that: When the hydrodynamic function model is in operation, it specifically includes: S31: Call the baseline risk curve, which retrieves the baseline water level-risk relationship curve based on historical clean water data from the built-in database; S32: Based on the current water and sediment content data, the sediment influence coefficient is calculated using the Lorenz curve model; S33: The basic risk value corresponding to each water point in the benchmark water level-risk relationship curve is weighted and superimposed with the sediment influence coefficient to generate a dynamic water level-risk relationship curve; S34: Input the current real-time water level data into the dynamic water level-risk relationship curve to calculate the comprehensive risk value at the current moment after correction for water and sediment content; S35: Automatically adjust the system's early warning trigger threshold based on the dynamic water level-risk relationship curve.

8. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 7, characterized in that: The flash flood risk prediction module, based on preprocessed data and dynamic risk factors, uses a built-in AI model to calculate and output the probability of flash floods occurring within a future period. Specifically, this includes: S41: Synchronously receive the standardized environmental data stream from the data processing module and the standardized dynamic correction parameter set from the dynamic risk factor calculation module; S42: The standardized environmental data stream and the dynamically corrected parameter set are spatiotemporally aligned and feature-stitched, and a unified input feature vector containing static environmental features and dynamic influence features is generated through a weighted fusion algorithm; S43: Input the unified input feature vector in parallel into the physical information neural network model and the hydrological mechanism model. The physical information neural network model outputs data-driven probability predictions, and the hydrological mechanism model outputs calculated probability values ​​based on physical equations. S44: Based on the current data quality indicators and model confidence, dynamically assign weights to the probability values ​​output by the two models, and use an adaptive weighted average algorithm to generate preliminary ensemble probability values; S45: Based on the dynamic influence factor in the unified input feature vector, the initial integration probability value is nonlinearly corrected; S46: Repeat steps S43-S45 at multiple future time points to generate time-series prediction results containing the probability values ​​of flash floods at different forecast periods, and transmit the time-series prediction results to the early warning decision module.

9. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 8, characterized in that: When performing nonlinear correction on the initial integrated probability value based on the dynamic influencing factors in the unified input feature vector, the final flash flood probability value is generated by introducing human activity influencing factors as probability multiplier factors and water and sediment enhancement effect values ​​as probability exponent correction terms. The principle formula is as follows: ; in, This represents the final probability value of a flash flood. To initially integrate probability values, To correct the weight parameters, As a factor influencing human activities, This is the scaling factor. This represents the water and sediment enhancement effect value.

10. The IoT-based intelligent monitoring and early warning system for flash floods according to claim 9, characterized in that: The disaster simulation and optimization module, when in operation, specifically includes: S51: After a disaster occurs, the system automatically retrieves and integrates high-precision pre-disaster terrain data, multi-source monitoring data throughout the disaster process, and external environmental data to construct a complete data chain for the disaster process. S52: Based on the complete data chain, the hydrodynamic model and the geological disaster model are driven to perform parallel calculations in the digital twin engine to simulate and reproduce the evolution process of flash floods, the inundation range, the change of impact force and the occurrence process of secondary disasters; S53: Spatially and temporally overlay and compare the simulated disaster evolution process, inundation range, peak water level and disaster intensity with the actual recorded disaster information to identify the areas of difference and key points of divergence between the simulation results and the actual disaster. S54: Based on the aforementioned areas of difference and key points of divergence, analyze the root causes of the differences and identify the types of system defects; S55: Based on the type of system defect located, access the pre-set optimization measures knowledge base, automatically match and generate targeted reinforcement solutions; S56: Output the generated enhancement plan to the system management platform in the form of a structured report, and automatically create optimization task work orders.

Citation Information

Cited By

  • Mountain torrent evolution and risk assessment method and system based on inducement characteristic change

    CN121638921A

  • Control method and system for deep-well pump

    CN121916174A