Flash flood disaster early warning method and system

Abstract

The application discloses a mountain torrent disaster early warning method and system, and the method comprises the following steps: acquiring cloud image data; processing the cloud image data based on a pre-trained deep learning model to obtain a predicted rainfall sequence; constructing a hydrological and hydrodynamic model and performing scenario simulation to demarcate a disaster threshold rainfall; and combining the predicted rainfall sequence and the disaster threshold rainfall to perform early warning. The system comprises an image acquisition module, a rainfall prediction module, a threshold determination module and an early warning module. By using the application, mountain torrent disaster early warning can be performed in advance, important disaster relief and risk avoidance time can be obtained for mountain torrent disasters, and the prediction accuracy is well guaranteed due to the combination of the rainfall prediction module and the atmospheric pressure observation and the cloud rainfall estimation method based on deep learning. The application can be widely applied in the field of natural disaster early warning.

Description

Technical Field

[0001] The present invention relates to the field of natural disaster early warning, and in particular to a flash flood disaster early warning method and system. Background Art

[0002] Flash floods are sudden, sudden floods in small and medium-sized river basins in mountainous areas caused by heavy rainfall. They are sudden and extremely destructive. As one of the most serious natural disasters worldwide, flash floods pose a significant threat to human society and the natural environment. With the impact of climate change, flash floods are becoming more frequent and more destructive.

[0003] Currently, flash flood forecasting relies primarily on meteorological observation data combined with hydrological models to provide flash flood warnings and forecasts. This method has significant shortcomings in its forecast period, with flash floods often already occurring by the time a disaster warning is issued, making it difficult to provide effective support for disaster prevention and mitigation. Summary of the Invention

[0004] In view of this, in order to solve the technical problem that most existing flash flood disaster warning methods rely on real-time rainfall data, which makes it difficult to predict the occurrence of flash flood disasters early, the present invention proposes a flash flood disaster warning method, which includes the following steps:

[0005] Acquire cloud image data;

[0006] Processing the cloud image data based on a pre-trained deep learning model to obtain a predicted rainfall sequence;

[0007] Construct hydrological and hydrodynamic models and conduct scenario simulations to determine the threshold rainfall for disasters;

[0008] An early warning is issued in combination with the predicted rainfall sequence and the disaster threshold rainfall.

[0009] In some embodiments, further comprising:

[0010] Monitor air pressure data to improve rainfall forecast accuracy.

[0011] Through this preferred step, the monitored atmospheric pressure is also sent to the server. Combining cloud photography with dynamic air pressure monitoring can greatly improve the accuracy of rainfall prediction.

[0012] In some embodiments, the cloud image data includes historical observation data and real-time cloud photography data. The step of processing the cloud image data based on the pre-trained deep learning model to obtain a predicted rainfall sequence specifically includes:

[0013] Calculating the spatiotemporal sequence of clear sky background brightness based on the recent historical observation data;

[0014] According to the real-time observation data, a cloud layer region and a clear sky region are divided;

[0015] A brightness value is extracted for the clear sky region, and a space-time interpolation is performed on the cloud layer region to obtain a clear sky background brightness sequence;

[0016] According to the real-time cloud layer photography data and the clear sky background brightness sequence, a brightness subtraction is performed to obtain a corrected cloud layer brightness;

[0017] The corrected cloud layer brightness is input into a pre-trained deep learning model to output a predicted rainfall sequence.

[0018] The rainfall prediction method is applicable to both daytime and nighttime.

[0019] Through the preferred embodiment, the clear sky background brightness at each time is calculated through a space-time interpolation method in combination with continuous photography observation, and the background brightness calculation result contains the brightness of a light scattering contribution factor, so that the actual situation can be more accurately matched, and higher reliability is achieved.

[0020] In some embodiments, the step of constructing a hydrological and hydrodynamic model and performing scenario simulation to demarcate a disaster threshold rainfall specifically comprises:

[0021] Obtaining underlying surface information of a target region;

[0022] Constructing a hydrological and hydrodynamic model according to the underlying surface information;

[0023] Performing simulation based on the hydrological and hydrodynamic model and a preset rainfall scenario to demarcate a disaster threshold rainfall.

[0024] Through the preferred step, the critical rainfall that causes disasters is calculated as a feature signal of a mountain flood disaster through the above hydrological and hydrodynamic simulation, so that the disaster is predictable. However, the existing mountain flood warning technology mainly relies on real-time hydrological and hydrodynamic simulation when rainfall occurs to predict disaster conditions, and such prediction is too lagging.

[0025] In some embodiments, the hydrological and hydrodynamic model comprises a runoff generation module and a runoff concentration module.

[0026] The hydrological and hydrodynamic model used in the existing mountain flood simulation is relatively traditional, and the influence of soil moisture on runoff generation and concentration is not considered. However, the hydrological and hydrodynamic simulation of the embodiment calculates the change of soil moisture with time according to rainfall, so as to realize the dynamic change of the runoff generation and concentration model.

[0027] The application further provides a mountain flood disaster warning system, which comprises:

[0028] An image acquisition module is configured to acquire cloud layer image data;

[0029] A rainfall prediction module processes the cloud image data based on a pre-trained deep learning model to obtain a predicted rainfall sequence;

[0030] The threshold determination module is used to build a hydrological and hydrodynamic model and conduct scenario simulations to determine the threshold rainfall for disasters;

[0031] An early warning module is used to provide early warning by combining the predicted rainfall sequence and the disaster threshold rainfall.

[0032] Based on the above scheme, the present invention provides a flash flood disaster warning method and system, which uses a photographic monitoring element to obtain cloud dynamic image data, combined with a barometer element to monitor atmospheric pressure dynamic data, and inputs the above input data into a deep learning model to estimate the predicted rainfall sequence. Combined with the warning rainfall pre-simulated by the hydrological and hydrodynamic model, it is judged whether a flash flood disaster will occur, and ultimately realizes the early signal recognition and warning of flash flood disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flowchart of the steps of a flash flood disaster early warning method of the present invention;

[0034] Figure 2 This is a schematic structural diagram of a cloud photography data acquisition device according to an embodiment of the present invention;

[0035] Figure 3 It is a schematic diagram of the structure of the deep learning model according to an embodiment of the present invention;

[0036] Figure 4 is an equivalent schematic diagram of a confluence module according to an embodiment of the present invention;

[0037] Figure 5 This is a structural block diagram of a flash flood disaster warning system of the present invention; DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0039] It should be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other.

[0040] It should be understood that the terms "system," "device," "unit," and / or "module" used in this application are a method for distinguishing different components, elements, parts, portions, or assemblies at different levels. However, if other terms can achieve the same purpose, the terms may be replaced by other expressions.

[0041] As used in this application and the claims, unless the context clearly indicates an exception, the terms "a," "an," "an," and / or "the" are not intended to refer to the singular and may include the plural, unless the context clearly indicates otherwise. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements. The phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, product, or apparatus that includes the elements.

[0042] In the description of the embodiments of this application, "plurality" refers to two or more than two. The terms "first" and "second" below are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more of the features.

[0043] In addition, flow charts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps may be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0044] Reference Figure 1 , which is a flow chart of an optional example of a flash flood disaster early warning method proposed in the present invention. The method can be applied to a computer device. The early warning method proposed in this embodiment may include but is not limited to the following steps:

[0045] Step S1, obtaining cloud image data;

[0046] Step S2: processing the cloud image data based on a pre-trained deep learning model to obtain a predicted rainfall sequence;

[0047] Step S3: construct a hydrological and hydrodynamic model and conduct scenario simulation to determine the threshold rainfall for disasters;

[0048] Step S4: issuing an early warning based on the predicted rainfall sequence and the disaster threshold rainfall.

[0049] In some feasible embodiments, step S1 specifically includes:

[0050] Use photography acquisition components to continuously obtain cloud photography dynamic data.

[0051] Among them, cloud photography data collection equipment refers to Figure 2 The operating process is as follows: 1. The lens element acquires cloud imaging data, and the barometer element acquires atmospheric pressure monitoring data; 2. The data is temporarily stored in the storage element; 3. The data is sent to the server via the network module; 4. If the transmission fails, the server sends a data request, and the device resends the relevant data upon receiving the data request; 5. Expired data is deleted to free up storage space. 6. The network function allows monitoring and setting of device parameters, such as lens parameters and power parameters, for remote maintenance.

[0052] The power supply solution is suitable for mountainous areas: for areas with mains electricity, it can be installed on the roof of a building or a streetlight pole and powered by the mains; for areas without mains electricity, a photovoltaic kit can be selected, with photovoltaic panels supplying power during the day and charging the battery in the device, and using the battery for power at night.

[0053] The device in this embodiment adopts a fully sealed design to prevent the impact of outdoor moisture, dust and insects on the device; the lens cover adopts a hydrophobic design to ensure that observation capabilities are still available during rainfall; the device has a heat dissipation module, and the fan is turned on according to the temperature sensor to achieve internal air circulation, and the heat is discharged from the heat dissipation fins in the shade at the bottom to ensure that the internal temperature of the device meets the operating temperature conditions under the scorching sun.

[0054] In some feasible embodiments, the following further comprises:

[0055] Monitor air pressure data.

[0056] The above device contains a barometer, and the monitored atmospheric pressure is also sent to the server.

[0057] Traditional cloud photography observations only estimate water content based on cloud shape characteristics, and their accuracy in predicting rainfall is not high. However, the present invention combines cloud photography with dynamic air pressure monitoring, which can greatly improve the accuracy of rainfall prediction.

[0058] In some feasible embodiments, step S2 specifically includes:

[0059] S2.1. Calculate the spatiotemporal sequence of clear sky background brightness based on the recent historical observation data;

[0060] S2.2. Dividing the real-time observation data into a cloud area and a clear sky area;

[0061] S2.3. Extracting brightness values ​​for the clear sky area and performing spatiotemporal interpolation on the cloud area to obtain a clear sky background brightness sequence;

[0062] S2.4. Subtracting the clear sky background brightness sequence from the real-time cloud photography data to obtain corrected cloud layer brightness;

[0063] S2.1-S2.4 are data preprocessing steps, the server calculates the spatiotemporal sequence of the clear sky background brightness of each cloud photography device at 0 o'clock every day, uses the continuous observation data within the adjacent 24 hours, divides the image into a cloud layer region and a clear sky region according to a classification algorithm, extracts the brightness value (including the night sky) of the clear sky region, and performs spatiotemporal interpolation on the part without clear sky to calculate the clear sky background brightness sequence at each time of the day within 24 hours. For the extreme case of continuous absence of clear sky, the last clear sky background brightness calculation result is used.

[0064] When the server receives the real-time cloud photography data, the real-time cloud photography brightness is subtracted from the clear sky background brightness calculated above to obtain the corrected cloud layer brightness. The cloud layer brightness sequence at any time and within the past 24 hours is used as the input data of the next step of the deep learning model.

[0065] The cloud photography data preprocessing method has the characteristics that the continuous change value of the clear sky background brightness within 24 hours is completed by using the continuous observation of the clear sky within the adjacent 10 days. Most of the existing radiation correction methods are based on the atmospheric transmission theory of solar radiation, and the actual background brightness deviates greatly from the theoretical radiation brightness due to the influence of near-ground complex factors (such as light scattering). The above method calculates the clear sky background brightness at each time by using the spatiotemporal interpolation method, and the background brightness calculation result contains the brightness of the light scattering and other contributing factors, which can more accurately match the actual situation and has higher reliability.

[0066] S2.5. Inputting the corrected cloud layer brightness into the pre-trained deep learning model to output a predicted rainfall sequence.

[0067] The model architecture combines a convolutional neural network and a long short-term memory network, and the specific model structure is as shown in FIG. 2. Figure 3 The input data is a 12-hour cloud image sequence and a pressure monitoring sequence. The LSTM layer is used to process the cloud time dynamics and the pressure time dynamics. The 3D convolution layer is used for cloud amount perception. The Dropout layer is used to prevent model overfitting. The intermediate layer is used to match the cloud time dynamics and the pressure time dynamics. Finally, the full connection layer is used to output a 6-hour future rainfall sequence.

[0068] In some feasible embodiments, the prior training step of the deep learning model is as follows:

[0069] The rain gauge observation station and cloud layer photography hardware equipment are laid out, rain record and cloud layer photography observation and atmospheric pressure monitoring are carried out for a period of time, the obtained rain record, cloud layer dynamic process data and atmospheric pressure dynamic process data are used for deep learning prior training, and a deep learning model is obtained. For regions with similar climate conditions, the trained deep learning model can be directly used.

[0070] The cloud layer rainfall prediction deep learning model is characterized in that: the input data is background brightness, photography image and atmospheric pressure, and the output data is a predicted rainfall sequence; on the model structure, the convolutional neural network and the long short-term memory network are fused, the spatiotemporal dynamics of the cloud layer have a perception ability, the atmospheric pressure dynamic process is perceived, and the rainfall prediction accuracy can be greatly improved.

[0071] In some possible embodiments, the step S3 specifically comprises:

[0072] Before the disaster, a plurality of rainfall scenarios are established in advance, a hydrological and hydrodynamic model is combined, a mountain flood disaster is preplayed, and the conditions (i.e. a disaster rainfall threshold) for the occurrence of a mountain flood disaster in a target region are determined.

[0073] The hydrological and hydrodynamic model mainly comprises a runoff generation module and a confluence module.

[0074] (1) The runoff generation module divides the underlying surface into a pervious surface and an impervious surface. The pervious surface includes grassland, wetland and forest land and the like, and is mainly full of runoff on the underlying surface, and the infiltration equation is as follows:

[0075]

[0076] A+P<W' mm Time:

[0077]

[0078] A+P>W' mm Time:

[0079] R PA =P-(WM-k·API)

[0080] Wherein, API is a previous rainfall index, which represents the change of soil moisture with time, A is the original water storage capacity of the catchment unit (mm), P represents the rainfall; W' mm is the maximum water storage capacity of the catchment unit (mm); k is the attenuation coefficient of the previous rainfall index with a given value; WM is the average water storage capacity of the catchment unit (mm); B is the unevenness of the water storage capacity on the catchment unit; R PA is the runoff of the pervious surface (mm).

[0081] The impervious surface refers to the built-up area such as roads, and the infiltration formula on the underlying surface is as follows:

[0082]

[0083] Where R DA is the flow rate of the direct impervious surface (mm); F is the infiltration volume (mm); f is the infiltration rate (mm / h); x is the duration of rainfall (h). In the model, rainfall within 4 hours after the rainfall moment is considered continuous rainfall; A DA 、B DA and C DA These are all calculation parameters for infiltration rate.

[0084] Indirect impervious surfaces refer to impervious surfaces in built-up areas such as buildings and roads that flow through permeable surfaces before being discharged into stormwater pipes. The main innovation of this infiltration model is the introduction of API to improve the calculation of CN values ​​in the SCS-CN model to adapt to different antecedent rainfall scenarios:

[0085] CN'=(a·API+b)·CN

[0086] Where CN' is the characteristic value of the improved runoff curve; a and b are empirical values ​​that can be determined through actual measurement or parameter adjustment.

[0087]

[0088] Where R NDA is the indirect impervious surface runoff; S is the lost rainfall.

[0089] (2) The confluence module simulates the process of net rain from all parts of the basin to the outlet section. The model generalizes the basin into a nonlinear reservoir, such as Figure 4 As shown in Figure 1, the confluence is calculated by solving the continuity equation and the Manning equation.

[0090] The water balance equation of a nonlinear reservoir is as follows:

[0091]

[0092] Where, d is the depth of ponding water, m; i is the rainfall intensity, mm / s; e is the surface evaporation rate, mm / s; f is the infiltration rate, mm / s; q is the surface runoff, mm / s;.

[0093] The surface runoff flow calculation formula (Manning formula) is as follows:

[0094]

[0095] Where Q is the runoff (m3); n is the Manning coefficient; S is the slope of the watershed; R X is the hydraulic radius (m); A X is the cross-sectional area of ​​the nonlinear reservoir (m2).

[0096] where R X and A X The following formulas are used for calculation:

[0097] A X =W(dd s )

[0098] R X =dd s

[0099] Where W is the characteristic width of the watershed (m); d is the water depth (m); and ds is the maximum depression storage depth (m).

[0100] Dividing Q by the basin area gives the expression for the surface runoff flow q of the basin:

[0101]

[0102] In some feasible embodiments, step S4 specifically includes:

[0103] The hydrological and hydrodynamic simulation of a certain mountainous basin showed that flash floods would occur when the rainfall reached 50 mm. Cloud photography data acquisition equipment was installed in the basin to conduct dynamic monitoring and prediction of rainfall. At a certain moment, it was identified that a shower was about to occur and the expected rainfall exceeded the 50 mm threshold, indicating that an early flash flood signal had appeared, and a flash flood disaster warning was immediately issued.

[0104] Current flash flood signal recognition and warning methods widely used in engineering projects, such as those based on ground-based rain gauges, weather radar, or remote sensing satellites, fail to balance construction costs, forecast period, and accuracy, and suffer from numerous limitations in practical applications. Based on the above-mentioned solution, the present invention offers significant advantages by combining low cost, timely forecasting, and high accuracy.

[0105] like Figure 5 As shown, a flash flood disaster early warning system includes:

[0106] An image acquisition module, used to acquire cloud image data;

[0107] A rainfall prediction module processes the cloud image data based on a pre-trained deep learning model to obtain a predicted rainfall sequence;

[0108] The threshold determination module is used to build a hydrological and hydrodynamic model and conduct scenario simulations to determine the threshold rainfall for disasters;

[0109] An early warning module is used to provide early warning by combining the predicted rainfall sequence and the disaster threshold rainfall.

[0110] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0111] A flash flood disaster warning device:

[0112] at least one processor;

[0113] at least one memory for storing at least one program;

[0114] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned flash flood disaster early warning method.

[0115] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0116] A storage medium stores processor-executable instructions, which are used to implement the above-mentioned flash flood disaster early warning method when executed by the processor.

[0117] The contents of the above method embodiments are all applicable to the present storage medium embodiment. The functions specifically implemented by the present storage medium embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0118] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A flash flood disaster early warning method, characterized in that: The following steps are involved: Acquire cloud image data and monitor air pressure data, wherein the cloud image data includes recent historical observation data and real-time cloud photography data; The cloud image data is processed based on a pre-trained deep learning model to obtain a predicted rainfall sequence, which specifically includes: calculating a clear sky background brightness sequence based on the recent historical observation data; dividing the real-time cloud photography data into cloud regions and clear sky regions; extracting brightness values ​​from the clear sky regions and performing spatiotemporal interpolation on the cloud regions to obtain a clear sky background brightness sequence; performing brightness subtraction based on the real-time cloud photography data and the clear sky background brightness sequence to obtain a corrected cloud brightness; inputting the corrected cloud brightness into the pre-trained deep learning model to output a predicted rainfall sequence; The model architecture integrates convolutional neural networks and long short-term memory networks. The input data is a cloud image sequence and an air pressure monitoring sequence. The LSTM layer is used to process the temporal dynamics of cloud layer and air pressure. The 3D convolution layer is used for cloud cover perception. The dropout layer is used to prevent model overfitting. The intermediate layer is used to match the temporal dynamics of cloud layer and air pressure. The fully connected layer outputs the future predicted rainfall sequence. Construct hydrological and hydrodynamic models and conduct scenario simulations to determine the threshold rainfall for disasters; An early warning is issued in combination with the predicted rainfall sequence and the disaster threshold rainfall.

2. A flash flood disaster early warning method according to claim 1, characterized in that: The step of constructing a hydrological and hydrodynamic model and conducting scenario simulation to determine the threshold rainfall for disasters specifically includes: Obtain underlying surface information of the target area; constructing a hydrological dynamic model based on the underlying surface information; A simulation is performed based on the hydrological dynamic model and a preset rainfall scenario to determine the threshold rainfall for disasters.

3. A flash flood disaster early warning method according to claim 2, characterized in that: The hydrological dynamic model includes a runoff generation module and a runoff confluence module.

4. A flash flood disaster early warning method according to claim 3, characterized in that: In the runoff generation module, the underlying surface is divided into permeable surface, directly impermeable surface and indirectly impermeable surface, where: The infiltration equation for a permeable surface is as follows: A+P <W' mm hour: A+P>W' mm hour: R PA =P-(WM-k·API) Among them, API is the previous rainfall index, which represents the change of soil moisture over time, A is the original water storage capacity of the water catchment unit, P represents the rainfall; W' mm is the maximum water storage capacity of the catchment unit; k is the attenuation coefficient of the previous rainfall index of a given value; WM is the average water storage capacity of the catchment unit; B is the uneven distribution of water storage capacity on the catchment unit; R PA The flow rate of permeable surface; The infiltration equation for a directly impervious surface is as follows: Among them, R DA is the flow rate directly generated by the impervious surface; F is the infiltration volume; f is the infiltration rate; x is the duration of rainfall; A DA 、B DA and C DA are all calculation parameters of infiltration rate, Δt represents infiltration time, and P represents rainfall; The infiltration equation for an indirect impervious surface is as follows: CN′=(a·API+b)·CN Where, CN' is the characteristic value of the improved runoff curve; a and b are preset empirical values; Where R NDA is the indirect impervious surface runoff; S is the lost rainfall.

5. A flash flood disaster early warning method according to claim 4, characterized in that: In the confluence module, the surface runoff flow rate of the basin is expressed as follows: Where W is the characteristic width of the watershed; d is the depth of accumulated water; d s is the maximum depression water storage depth; S is the basin slope; A is the catchment area; n is the roughness.

6. A flash flood disaster warning system, characterized in that: Used to execute the flash flood disaster early warning method according to claim 1, comprising: An image acquisition module, used to acquire cloud image data; A rainfall prediction module processes the cloud image data based on a pre-trained deep learning model to obtain a predicted rainfall sequence; The threshold determination module is used to build a hydrological and hydrodynamic model and conduct scenario simulations to determine the threshold rainfall for disasters; An early warning module is used to provide early warning by combining the predicted rainfall sequence and the disaster threshold rainfall.

7. A flash flood disaster warning device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the flash flood disaster warning method as described in any one of claims 1 to 5.

Images