Flood risk monitoring and early warning method based on millimeter wave radar

Abstract

The invention discloses a flood risk monitoring and early warning method based on millimeter-wave radar, and relates to the technical field of flood monitoring, and the method comprises the following steps: a millimeter-wave radar array is deployed in a monitoring area in advance, and the radar array comprises a plurality of millimeter-wave radar nodes which dynamically adjust the sampling frequency according to the water level change rate of a real-time area; the original echo data of the monitoring area is collected through the millimeter wave radar array, and the original echo data at least comprises the water level height, the water flow speed and the water surface fluctuation characteristics. The self-adaptive sampling frequency adjusting mechanism can dynamically adjust the data acquisition density according to the change of the water level, the resource consumption is optimized while the monitoring precision is ensured, the subtle change when the flood rises rapidly can be captured, redundant data can be reduced in the stable period of the water level, and efficient and accurate dynamic monitoring is achieved. The distributed deployment of the radar array breaks through the limitation of single-point monitoring, and realizes the comprehensive coverage of a monitoring area.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flood monitoring, in particular to a flood risk monitoring and early warning method based on millimeter wave radar. BACKGROUND

[0002] As a frequently occurring natural disaster, flood often has the characteristics of strong suddenness, wide influence range and large destructive power, which poses a serious threat to urban and rural infrastructure, ecological environment and people's life and property safety. With the intensification of global climate change and the acceleration of urbanization, the frequency of extreme rainfall events is increasing year by year, leading to a significant increase in the probability of urban waterlogging and small and medium river floods. Traditional flood monitoring and early warning methods have been unable to meet the needs of modern disaster prevention and mitigation. Existing monitoring methods rely on fixed-site hydrological monitoring equipment, which is usually costly to deploy and has limited coverage, making it difficult to achieve comprehensive monitoring in complex terrain areas or remote areas. Moreover, such equipment is easily disturbed by the environment in adverse weather conditions, resulting in data interruption or reduced accuracy, and cannot provide timely and effective information support for emergency decision-making.

[0003] Current mainstream flood warning technologies are mostly based on a single data source or simple data superposition analysis, lacking comprehensive consideration of multi-dimensional hydrological characteristics, making it difficult to accurately capture the dynamic process of flood evolution. For example, video monitoring-based methods are greatly affected by light and weather, and their recognition accuracy is greatly reduced in heavy rain, fog and other conditions. Satellite remote sensing-based monitoring can achieve wide coverage, but the time resolution is low, which cannot meet the needs of real-time warning. Traditional hydrological models rely on empirical parameters and have poor adaptability to complex underlying surface conditions, resulting in large deviations in prediction results. In addition, existing technologies focus on a single indicator (such as water level) in the risk assessment stage, and fail to consider multiple factors such as flood flow rate, inundation range and impact force, resulting in inaccurate risk level classification and difficulty in effectively guiding emergency evacuation and rescue work. Therefore, developing a flood warning method that can achieve all-weather, high-precision monitoring and comprehensive multi-source information-based accurate risk assessment has become a key requirement for improving disaster prevention and mitigation capabilities.

[0004] To address the problems in the related art, no effective solutions have been proposed so far. SUMMARY

[0005] To address the problems in the related art, the present application proposes a flood risk monitoring and early warning method based on millimeter wave radar to overcome the above technical problems existing in the prior art.

[0006] The technical solution of the present application is as follows:

[0007] A flood risk monitoring and early warning method based on millimeter wave radar, comprising the following steps:

[0008] pre-deploying a millimeter wave radar array in the monitoring area, wherein the radar array comprises a plurality of millimeter wave radar nodes dynamically adjusting a sampling frequency according to a water level change rate of a real-time area;

[0009] collecting raw echo data of the monitoring area by the millimeter wave radar array, the raw echo data at least comprising a water level height, a water flow velocity and a water surface fluctuation feature;

[0010] pre-processing the raw echo data, at least comprising noise suppression based on wavelet threshold denoising, clutter removal based on constant false alarm rate detection, and extracting a water level time series, a water flow velocity field distribution and a water surface fluctuation spectrum feature;

[0011] inputting the pre-processed data into a spatio-temporal fusion prediction model to output a flood evolution prediction result in a future preset time period, wherein the spatio-temporal fusion prediction model comprises a convolutional neural network spatial feature extraction module, a long short-term memory network time series prediction module introducing an attention mechanism and a physical constraint fusion module;

[0012] constructing a three-dimensional terrain model of the monitoring area based on digital elevation model data, and calculating a submergence depth, a flood impact force and a submergence range index according to the prediction result, generating a flood risk index through normalization processing and weighted summation, and evaluating a risk level;

[0013] triggering a corresponding level of warning signals according to the risk level, the warning signals comprising multi-level audible and visual alarms, terminal notifications and linkage control signals with an emergency command system.

[0014] Further, the extracting of the water level time series further comprises: fusing and analyzing millimeter wave radar data with rainfall data, hydrological station data and / or weather forecast data to calculate data source errors, represented as:

[0015]

[0016] wherein e i (t) is an average error of the i-th data source; K is a time window size for error calculation; h i (t) is a water level measurement value of the i-th data source at time t; is a reference water level value;

[0017] performing dynamic weight allocation, represented as:

[0018]

[0019] wherein w i (t) is a weight of the i-th data source at time t; a is a regulation parameter, taking a value of 0.5; exp() is an exponential function;

[0020] The fused water level value is obtained and expressed as:

[0021]

[0022] Wherein, M is the total number of data sources.

[0023] Further, the convolutional neural network spatial feature extraction module comprises performing convolution operation on the two-dimensional water flow velocity field by using a convolutional neural network (CNN) to extract spatial distribution features of the water flow.

[0024] Wherein, the convolutional layer calculation is expressed as:

[0025]

[0026] Wherein, the pooling layer calculation is expressed as:

[0027] P i,j =max(C i·S:(i+1)·S,j·S:(j+1)·S );

[0028] Wherein, C i,j is the value of the (i,j) position of the convolutional layer output feature map; σ is an activation function, using ReLU; W m,n is the convolution kernel weight; V is the input velocity field data; b is the bias term; M and N are the size of the convolution kernel; P i,j is the (i,j) position value of the pooling layer output; and S is the pooling step.

[0029] Further, the physical constraint fusion module comprises introducing the water flow continuity equation as prior knowledge into the model to fuse and optimize the spatial features extracted by the CNN and the time features output by the LSTM.

[0030] Wherein, the water flow continuity equation constraint is expressed as:

[0031]

[0032] Wherein, the discretization constraint term is expressed as:

[0033]

[0034] Wherein, the prediction model loss function is expressed as:

[0035] L=L pred +λ·L phys ;

[0036] Wherein, is the partial derivative of the water level with respect to time; ▽ is the divergence operator; v is the flow velocity vector; r is the rainfall; L phys is the physical constraint loss term; and ht+1 , h t is the water level at time t+1 and t; v t is the flow velocity vector at time t; r t is the rainfall at time t; is the square of L2 norm; L is the total loss function; L pred is the prediction loss; L phys is the physical constraint term; λ is the constraint coefficient.

[0037] Further, the submerged depth is calculated and represented as:

[0038] d(x, y, t) = max(0, h pred (x, y, t) - DEM(x, y));

[0039] wherein d(x, y, t) is the submerged depth at time t at position (x, y); h pred (x, y, t) is the predicted water level at time t at position (x, y); DEM(x, y) is the digital elevation model data at position (x, y); max() is the maximum value function.

[0040] Further, the flood impact force is calculated and represented as:

[0041] P(x, y, t) = 0.5 · ρ · v(x, y, t) 2 · (1 + C d );

[0042] wherein P(x, y, t) is the dynamic water pressure at time t at position (x, y); ρ is the density of water; v(x, y, t) is the water flow velocity at time t at position (x, y); C d is the drag coefficient, taking 1.0.

[0043] Further, the risk level is evaluated, including the following steps:

[0044] The risk index is obtained and represented as:

[0045] RI(x, y, t) = w d · d' + w v · v' + w p · P' + w a · A r ;

[0046] wherein RI(x, y, t) is the risk index at time t at position (x, y); A r is the submerged range index; w d , w v , w p , w aThe weight coefficients of various factors are 0.4, 0.2, 0.2 and 0.2 respectively.

[0047] The risk level division is represented as:

[0048]

[0049] wherein Level(x, y, t) is the risk level of (x, y) position at time t; 0.3, 0.5 and 0.8 are risk level division thresholds.

[0050] Advantages of the present application:

[0051] 1. The present application significantly improves the all-weather nature and environmental adaptability of flood monitoring by deeply combining millimeter wave radar technology with multi-source data fusion algorithm. The millimeter wave radar has the interference of rain, fog, light and other adverse conditions, and can continuously and stably collect key hydrological data such as water level and flow rate in complex weather environment, solving the problem of failure of traditional optical monitoring methods in extreme weather. At the same time, the adaptive sampling frequency adjustment mechanism can dynamically adjust the data acquisition density according to the water level change, optimizing resource consumption while ensuring monitoring accuracy, which can capture subtle changes during rapid rise of flood and reduce redundant data during stable water level, realizing efficient and accurate dynamic monitoring. The distributed deployment of radar array breaks through the limitation of single-point monitoring, realizes comprehensive coverage of the monitoring area, and provides rich spatial dimension information for subsequent flood evolution analysis.

[0052] 2. The present application constructs a spatio-temporal prediction model and a multi-dimensional risk assessment system that integrates physical constraints, greatly improving the accuracy and decision value of early warning. By combining the attention mechanism LSTM with the water flow continuity equation, the prediction model can not only capture the nonlinear time characteristics of water level changes, but also follow the physical laws of hydrological motion, making the flood evolution trend prediction more scientific and reliable. The risk assessment process comprehensively considers multiple indicators such as inundation depth, flow velocity, impact force and inundation range, and realizes fine description of flood risk through dynamic weight distribution, avoiding the limitations of single-index evaluation. The multi-source data fusion mechanism further enhances the robustness of the system, and through cross-validation and complementation of different data sources, effectively reduces the influence of single-device error on monitoring results. This whole-chain optimization from data acquisition to risk assessment makes the early warning information more timely and accurate, and can provide scientific basis for emergency command, significantly improving the response efficiency and disposal capacity of disaster prevention and reduction. BRIEF DESCRIPTION OF DRAWINGS

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a flood risk monitoring and early warning method based on millimeter-wave radar according to an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0056] According to an embodiment of the present invention, a flood risk monitoring and early warning method based on millimeter-wave radar is provided.

[0057] like Figure 1 As shown, the flood risk monitoring and early warning method based on millimeter-wave radar according to an embodiment of the present invention includes the following steps:

[0058] Step S1 involves pre-deploying a millimeter-wave radar array in the monitoring area. Multiple millimeter-wave radar nodes are deployed to form a radar monitoring network, based on the terrain features and flood-prone areas. Each radar node operates in the 24GHz or 77GHz frequency band, providing a water level measurement accuracy of 0.1m and a flow velocity measurement accuracy of 0.1m / s. The radar nodes have an adaptive sampling frequency adjustment function; when a water level change rate exceeds 0.5m / h, the sampling frequency is automatically increased from the standard 2Hz to 15Hz to capture rapidly changing hydrological characteristics.

[0059] Step S2 involves acquiring raw echo data from a millimeter-wave radar array. The millimeter-wave radar nodes continuously transmit millimeter-wave signals into the monitoring area and receive reflected echoes. The raw echo data contains hydrological information, including: calculating water level height by analyzing the time delay of the echo signal; obtaining water flow velocity through Doppler frequency shift analysis; and extracting water surface undulation characteristics through amplitude and phase changes of the echo signal.

[0060] Step S3, perform data preprocessing on the raw echo data, including the following steps:

[0061] Noise suppression is performed, including the use of wavelet thresholding algorithms to remove environmental noise;

[0062] Clutter removal is performed to remove the reflection signals of fixed targets such as the ground and buildings by a constant false alarm rate (CFAR) detection algorithm.

[0063] Feature extraction is performed to perform time-frequency analysis on the processed signals, and water level time series, two-dimensional water flow velocity field and water surface fluctuation frequency spectrum features are extracted as feature data to be input.

[0064] In addition, for the water level time series, the millimeter wave radar data are fused and analyzed with rainfall data, hydrological station data and / or weather forecast data, and a water level value after fusion is obtained by a dynamic weight distribution algorithm to extract the water level time series, thereby further improving the monitoring and prediction accuracy. The specific steps are as follows:

[0065] The data source error is calculated and represented as:

[0066]

[0067] wherein e i (t) is the average error of the i th data source; K is the time window size of error calculation; h i (t) is the water level measurement value of the i th data source at time t; is a reference water level value;

[0068] Dynamic weight distribution is performed and represented as:

[0069]

[0070] wherein w i (t) is the weight of the i th data source at time t; a is an adjustment parameter and takes a value of 0.5; exp() is an exponential function;

[0071] The water level value after fusion is obtained and represented as:

[0072]

[0073] wherein M is the total number of data sources.

[0074] Step S4, flood evolution trend prediction is performed, including inputting the feature data obtained by preprocessing into an improved space-time fusion prediction model, and the model outputs water level change prediction and water flow evolution trend in the future 1-6 hours, including:

[0075] Spatial feature extraction is performed to perform convolution operation on the two-dimensional water flow velocity field by using a convolutional neural network (CNN), and spatial distribution features of the water flow are extracted.

[0076] wherein the convolution layer calculation is represented as:

[0077]

[0078] wherein the pooling layer calculates, denoted as:

[0079] P i,j = max(C i·S:(i+1)·S,j·S:(j+1)·S );

[0080] wherein C i,j is the value of the (i,j) position of the convolutional layer output feature map; σ is the activation function, using ReLU; W m,n is the convolution kernel weight; V is the input velocity field data; b is the bias term; M, N are the size of the convolution kernel; P i,j is the value of the (i,j) position of the pooling layer output; S is the pooling step.

[0081] For time series prediction, an LSTM network with attention mechanism is used to model the water level time series, and the attention mechanism can automatically focus on the historical time data that has a greater impact on the prediction results;

[0082] wherein the LSTM unit updates, denoted as:

[0083]

[0084] wherein the attention weight calculation is denoted as:

[0085]

[0086] wherein the output layer is denoted as:

[0087]

[0088] wherein f t is the forget gate output; i t is the input gate output; is the cell state candidate value; c t is the cell state; o t is the output gate output; h t is the hidden state; x t is the input at time t; W f ,W i ,W c ,W o are weight matrices; b f ,b i ,b c ,b o are bias terms; σ is the sigmoid activation function; tanh is the hyperbolic tangent activation function; ⊙ is the element-wise multiplication; e t,i is the attention energy value; α t,i is the attention weight; context tis a vector; v, W1, W2 are attention mechanism parameters; is a water level prediction value at t+1 time.

[0089] Physical constraint fusion is performed, the water flow continuity equation is introduced into the model as prior knowledge, and the spatial features extracted by the CNN and the time features output by the LSTM are fused and optimized to improve the physical rationality of the prediction results.

[0090] The water flow continuity equation constraint is represented as:

[0091]

[0092] The discretization constraint term is represented as:

[0093]

[0094] The prediction model loss function is represented as:

[0095] L=L pred +λ·L phys ;

[0096] wherein, is the partial derivative of the water level with respect to time; ∇ is the divergence operator; v is the flow velocity vector; r is the rainfall; L phys is a physical constraint loss term; h t+1 , h t are water levels at t+1 and t time; v t is the flow velocity vector at t time; r t is the rainfall at t time; is the square of the L2 norm; L is the total loss function; L pred is the prediction loss; L phys is the physical constraint term; λ is a constraint coefficient, which is 0.1, the model outputs the water level change prediction and the water flow evolution trend in the future 1-6 hours.

[0097] Step S5, risk level assessment, including: constructing a three-dimensional terrain model according to digital elevation model (DEM) data, superimposing and analyzing the predicted water level data and terrain data to obtain the possible flooded area and depth.

[0098] At the same time, according to the water flow velocity field, the flood impact force distribution is calculated, and the three indexes of flooded area, flooded depth and flood impact force are integrated to construct a risk assessment matrix, and the monitoring area is divided into low, medium, high and extremely high four risk levels.

[0099] The flooded depth is calculated as:

[0100] d(x,y,t)=max(0,h pred(x, y, t) - DEM(x, y));

[0101] wherein d(x, y, t) is the inundation depth at the position (x, y) at time t; h pred (x, y, t) is the predicted water level at the position (x, y) at time t; DEM(x, y) is the digital elevation model data at the position (x, y); max() is the maximum value function.

[0102] wherein the flood impact force calculation is represented as:

[0103] P(x, y, t) = 0.5 - p - v(x, y, t) 2 (1 + C d );

[0104] wherein P(x, y, t) is the dynamic water pressure at the position (x, y) at time t; p is the density of water; v(x, y, t) is the water flow velocity at the position (x, y) at time t; C d is the drag coefficient, taking 1.0.

[0105] In addition, normalization processing is performed, represented as:

[0106]

[0107] wherein d', v', P' are the normalized inundation depth, flow velocity and dynamic water pressure; d max is the maximum reference inundation depth; v max is the maximum reference flow velocity; P max is the maximum reference dynamic water pressure; min() is the minimum value function.

[0108] The risk index is obtained, represented as:

[0109] RI(x, y, t) = w d d' + w v v' + w p P' + w a A r ;

[0110] wherein RI(x, y, t) is the risk index at the position (x, y) at time t; A r is the inundation range index; w d , w v , w p , w a are the weight coefficients of each factor, taking values of 0.4, 0.2, 0.2, 0.2 respectively.

[0111] The risk level division is calibrated, represented as:

[0112]

[0113] wherein, Level(x, y, t) is the risk level of the position (x, y) at time t; 0.3, 0.5, 0.8 are the risk level division threshold values.

[0114] Step S6, generating and issuing a warning signal, including: when the real-time monitoring data or the predicted data reaches the preset risk threshold value, the system automatically triggers a warning of the corresponding level:

[0115] Low risk: record abnormal data and continue monitoring;

[0116] Medium risk: yellow warning information is issued on the electronic display screen in the monitoring area;

[0117] High risk: send a warning SMS to the residents in the area and start the sound and light alarm device;

[0118] Extremely high risk: link the city emergency command system, issue a red warning, and start the emergency evacuation plan.

[0119] With the above technical solution, taking the city waterlogging monitoring and warning system in operation as an example, a millimeter wave radar array is deployed in a low-lying area of a city prone to waterlogging, including 8 radar nodes, covering an area of about 5 square kilometers. The radar node uses a 24GHz frequency band, with a horizontal beam width of 30° and a vertical beam width of 10°, and a maximum detection distance of 500 meters. The specific system working process is as follows:

[0120] Radar deployment is performed in advance, and 8 radar nodes are installed on streetlight poles or building tops to ensure that the monitoring range covers the entire low-lying area and the surrounding main drainage channels. Through time synchronization technology, the data collection times of each radar node are consistent.

[0121] Data collection is performed, and the radar nodes collect echo data at a regular frequency of 2Hz. When it is monitored that the water level in a certain area rises by more than 0.3 meters within 10 minutes, the radar nodes in the area and adjacent areas automatically increase the sampling frequency to 15Hz.

[0122] Data preprocessing is performed, db4 wavelet basis function is used to decompose the original echo signal for 3 layers, rain and fog interference is removed through threshold denoising; the reflection signal of fixed targets such as ground and buildings is removed by using unit average CFAR algorithm; the processed signal is subjected to Fourier transform and wavelet transform, and the average value of water level, water flow velocity field distribution and main frequency component of water surface fluctuation every 5 minutes are extracted.

[0123] Wherein, the water level average value fuses the real-time data of 5 rainfall stations in the region and the future 2 hours of rainfall forecast data. By calculating the error sum of squares of each data source and historical data, the weight coefficient is dynamically adjusted, when the error of millimeter wave radar data is less than 0.05 meters, the weight coefficient is set to 0.6, the weight of rainfall data is set to 0.3, and the weight of other data is set to 0.1, and the water level average value every 5 minutes after fusion is extracted;

[0124] Performing flood prediction, comprising the following steps:

[0125] The water level time series of the past 2 hours is input into the attention mechanism LSTM network, which contains 3 hidden layers, each hidden layer has 64 neurons;

[0126] The CNN is used for feature extraction of the water flow velocity field, and the CNN contains 2 convolution layers and 1 pooling layer;

[0127] The continuity equation of water flow is introduced as a constraint condition, and the time prediction result of LSTM and the spatial feature of CNN are fused, and the water level prediction result of the future 3 hours is output, and the time resolution is 10 minutes.

[0128] Performing risk assessment, comprising the following steps:

[0129] Loading the DEM data of the region, the resolution is 1 meter x 1 meter;

[0130] Superimposing the predicted water level data with the DEM, calculating the submergence depth of each region and calculating the flood impact force;

[0131] Building a risk assessment matrix, as follows:

[0132] When the submergence depth is <0.3 meters and the impact force is <500 N / m 2 , it is low risk;

[0133] When 0.3 meters ≤ submergence depth <0.5 meters or 500 N / m 2 ≤ impact force <1000 N / m 2 , it is medium risk;

[0134] When 0.5 meters ≤ submergence depth <1.0 meters or 1000 N / m 2 ≤ impact force <2000 N / m 2 , it is high risk;

[0135] When the submergence depth is ≥1.0 meters or the impact force is ≥2000 N / m 2 , it is extremely high risk.

[0136] According to the risk assessment result, the warning is issued, as follows:

[0137] Low risk: internal records of the system, no external warning;

[0138] Medium risk: yellow warning is issued through electronic display screens within the region, reminding citizens to pay attention to waterlogging;

[0139] High risk: send SMS warning to registered users in the region, start sound and light alarm devices on both sides of the road, and prompt vehicles and pedestrians to detour;

[0140] Extremely high risk: send linkage signal to the city emergency command center, start red warning, dispatch emergency drainage equipment, open emergency shelters, and issue evacuation notices through media such as radio and television.

[0141] In summary, with the above technical solutions of the present application, the following effects can be achieved:

[0142] 1. The present application combines millimeter wave radar technology with multi-source data fusion algorithm, significantly improving the all-weather and environmental adaptability of flood monitoring. Millimeter wave radar has the ability to interfere with rain, fog, and other harsh conditions, and can continuously and stably collect key hydrological data such as water level and flow rate in complex weather conditions, solving the problem of failure of traditional optical monitoring methods in extreme weather. At the same time, the adaptive sampling frequency adjustment mechanism can dynamically adjust the data acquisition density according to the water level changes, optimizing resource consumption while ensuring monitoring accuracy, capturing subtle changes during rapid rise of flood, and reducing redundant data during stable water level, achieving efficient and accurate dynamic monitoring. The distributed deployment of radar arrays breaks through the limitations of single-point monitoring, achieving comprehensive coverage of the monitoring area and providing rich spatial dimension information for subsequent flood evolution analysis.

[0143] 2. The present application constructs a spatio-temporal prediction model and a multi-dimensional risk assessment system that integrates physical constraints, significantly improving the accuracy and decision value of the warning. By combining the attention mechanism LSTM with the water flow continuity equation, the prediction model can capture the nonlinear time characteristics of water level changes and follow the physical laws of hydrological motion, making the flood evolution trend prediction more scientific and reliable. The risk assessment process takes into account multiple indicators such as flooding depth, water flow speed, impact force, and flooding range, and achieves fine description of flood risk through dynamic weight distribution, avoiding the limitations of single-index assessment. The multi-source data fusion mechanism further enhances the robustness of the system, through cross-validation and complementation of different data sources, effectively reducing the influence of single-device error on monitoring results. This whole-chain optimization from data collection to risk assessment makes the warning information more timely and accurate, providing scientific basis for emergency command, significantly improving the response efficiency and disposal capacity of disaster prevention and reduction.

[0144] The above descriptions are only the preferred embodiment of the application, not intended to limit the application. Based on the disclosure of the specification and embodiments, other embodiments of the disclosure will be easily conceived by those skilled in the art. The application is intended to cover any variations, uses, or adaptive changes of the disclosure that follow the general principles of the disclosure and include common knowledge or conventional technical means in the technical field not disclosed by the disclosure. The specification and embodiments are only considered as exemplary, and the true scope and spirit of the disclosure are indicated by the claims.

[0145] It should be understood that the present disclosure is not limited to the precise structures as herein described and illustrated in the drawings, and that various modifications and changes can be made without departing from its scope. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A flood risk monitoring and early warning method based on millimeter-wave radar, characterized in that, Includes the following steps: A millimeter-wave radar array is pre-deployed in the monitoring area, which includes multiple millimeter-wave radar nodes that dynamically adjust the sampling frequency according to the real-time water level change rate in the area. The millimeter-wave radar array collects raw echo data of the monitoring area, and the raw echo data includes at least water level height, water flow velocity, and water surface undulation characteristics. The raw echo data is preprocessed, including at least noise suppression based on wavelet threshold denoising, clutter removal based on constant false alarm rate detection, and extraction of water level time series, water flow velocity field distribution and water surface fluctuation spectrum characteristics. The preprocessed data is input into the spatiotemporal fusion prediction model, which outputs the flood evolution prediction results within a preset time period. The spatiotemporal fusion prediction model includes: a convolutional neural network spatial feature extraction module, a long short-term memory network time series prediction module with an attention mechanism, and a physical constraint fusion module. A three-dimensional terrain model of the monitoring area is constructed based on digital elevation model data. The inundation depth, flood impact force, and inundation range index are calculated based on the prediction results. The flood risk index is generated through normalization and weighted summation, and the risk level is assessed. Based on the risk level, a corresponding warning signal is triggered, including multi-level audible and visual alarms, terminal notifications, and linkage control signals with the emergency command system.

2. The flood risk monitoring and early warning method based on millimeter-wave radar according to claim 1, characterized in that, The extraction of water level time series also includes: fusing and analyzing millimeter-wave radar data with rainfall data, hydrological station data, and / or meteorological forecast data, and calculating the data source error, expressed as: Among them, e i (t) represents the average error of the i-th data source; K is the size of the time window for error calculation; h i (t) represents the water level measurement value of the i-th data source at time t; For reference water level values; Dynamic weight allocation is represented as: Among them, w i (t) represents the weight of the i-th data source at time t; α is an adjustment parameter with a value of 0.5; exp() is an exponential function; The merged water level value is represented as: Where M represents the total number of data sources.

3. The flood risk monitoring and early warning method based on millimeter-wave radar according to claim 1, characterized in that, The convolutional neural network spatial feature extraction module includes using a convolutional neural network (CNN) to perform a convolution operation on the two-dimensional water flow velocity field to extract the spatial distribution features of the water flow. The calculation of the convolutional layer is represented as follows: The pooling layer calculation is represented as follows: P i,j =max(C i·S:(i+1)·S,j·S:(j+1)·S ); Among them, C i,j σ is the (i,j) position value of the output feature map of the convolutional layer; σ is the activation function, using ReLU; W m,n V represents the convolution kernel weights; B represents the input velocity field data; C represents the bias term; M and N represent the kernel size; P represents the convolution kernel weights. i,j is the (i,j) position value output by the pooling layer; S is the pooling step size.

4. The flood risk monitoring and early warning method based on millimeter-wave radar according to claim 1, characterized in that, The physical constraint fusion module includes introducing the water flow continuity equation as prior knowledge into the model and fusing and optimizing the spatial features extracted by CNN and the temporal features output by LSTM. The continuity equation constraint for water flow is expressed as: The discretization constraint term is expressed as: The loss function of the prediction model is expressed as: L=L pred +λ·L phys ; in, This is the partial derivative of the water level with respect to time; Here, is the divergence operator; v is the velocity vector; r is the rainfall; L phys For physical constraint loss terms; h t+1 h t v represents the water level at times t+1 and t; t Let r be the velocity vector at time t; t Let be the rainfall at time t; L is the square of the L2 norm; L is the total loss function; L pred To predict losses; L phys λ represents the physical constraint term; λ is the constraint coefficient.

5. The flood risk monitoring and early warning method based on millimeter-wave radar according to claim 1, characterized in that, The calculated flooding depth is expressed as: d(x,y,t)=max(0,h pred (x,y,t)-DEM(x,y)); Where d(x,y,t) is the flooding depth at position (x,y) at time t; h pred (x,y,t) represents the predicted water level at time t at position (x,y); DEM(x,y) represents the digital elevation model data at position (x,y); max() is the function to take the maximum value.

6. The flood risk monitoring and early warning method based on millimeter-wave radar according to claim 5, characterized in that, The flood impact force is calculated and expressed as: P(x,y,t)=0.5·ρ·v(x,y,t) 2 ·(1+C d ); Where P(x,y,t) is the dynamic water pressure at position t (x,y); ρ is the density of water; v(x,y,t) is the water velocity at position t (x,y); C d The drag coefficient is set to 1.

0.

7. The flood risk monitoring and early warning method based on millimeter-wave radar according to claim 6, characterized in that, The risk assessment includes the following steps: The risk index is represented as: RI(x,y,t)=w d ·d′+w v ·v′+w p ·P′+w a A r ; Wherein, RI(x,y,t) is the risk index at position t of (x,y); A r The flooding extent index; w d w v w p w a The weighting coefficients for each factor are 0.4, 0.2, 0.2, and 0.2, respectively. The risk level classification is represented as follows: Where Level(x,y,t) is the risk level at time t at position (x,y); 0.3, 0.5, and 0.8 are the risk level classification thresholds.

Images