Rainstorm disaster early warning method and system based on rainstorm disaster bearing coefficient

Abstract

The invention relates to the technical field of disaster forecasting and early warning, and discloses a rainstorm disaster early warning method and system based on a rainstorm disaster bearing coefficient. Comprising the following steps: acquiring multi-source data and basic geographic information of an early warning area, extracting spatial-temporal characteristics based on the multi-source data and the basic geographic information, and calculating grid point forecast rainfall within a predetermined time according to the spatial-temporal characteristics; obtaining historical rainstorm disaster data of the early warning area, and calculating a grid point rainstorm disaster bearing coefficient based on the historical rainstorm disaster data and the basic geographic information; establishing an empirical model according to historical rainstorm disaster data in combination with basic geographic information to calculate a grid point rainfall threshold value; and according to the grid point forecast rainfall, the grid point rainstorm disaster bearing coefficient, the grid point rainfall threshold value and the field measurement rainfall, a rainstorm disaster prediction result of the early warning area is obtained in combination with the algorithm model, and the problem of weak monitoring capability of existing disaster forecast and early warning is solved.

Description

Technical Field

[0001] The invention relates to the technical field of weather forecasting, and in particular to a rainstorm disaster early warning method based on a rainstorm disaster bearing coefficient. Background Art

[0002] Rainfall is a very normal meteorological condition. However, during the rainy season, due to its extreme nature and short-term intensity, it is easy to cause rainstorms and floods. Rainstorms and floods are often accompanied by geological disasters such as mountain torrents, landslides and mud-rock flows, causing huge economic losses.

[0003] Existing disaster warning technology mainly uses weather forecasts to achieve protective warnings. Current weather forecasts mainly predict common weather conditions. When facing extreme weather events, because weather forecasts are only based on meteorological data, they cannot accurately provide disaster warnings for areas where disasters may occur. The monitoring and forecasting capabilities are weak and the degree of refinement is low. It is impossible to make an effective and objective evaluation of whether a flood disaster will occur in a region, and it is impossible to effectively warn of flood disasters. Summary of the invention

[0004] The present invention provides a rainstorm disaster early warning method and system based on a rainstorm disaster bearing coefficient, so as to solve the problem that the monitoring capability of the existing disaster forecast and early warning is weak.

[0005] In order to achieve the above object, the present invention is implemented by the following technical solutions: In a first aspect, the present invention provides a rainstorm disaster early warning method based on a rainstorm disaster tolerance coefficient, comprising the following steps: Step 1: Obtain multi-source data and basic geographic information of the warning area, extract spatiotemporal characteristics based on the multi-source data and basic geographic information, and calculate the grid point forecast rainfall within the predetermined time according to the spatiotemporal characteristics; Step 2: Obtain historical rainstorm disaster data in the warning area, calculate the grid point rainstorm disaster coefficient based on the historical rainstorm disaster data and basic geographic information, and establish an empirical model to calculate the grid point rainfall threshold based on the historical rainstorm disaster data combined with basic geographic information; Step 3: Obtain the rainstorm disaster prediction results for the warning area based on the grid forecast rainfall, grid rainstorm disaster resistance coefficient, grid rainfall threshold and field measured rainfall combined with the algorithm model.

[0006] Further, the multi-source data includes correction data and measurement data; The measurement data includes satellite data, ground observation data, radar data and sounding data.

[0007] Furthermore, the extraction of spatiotemporal features based on multi-source data and basic geographic information includes: extracting spatial sequences, time sequences and image data of rainfall based on multi-source data combined with spatial sequence features, encoding the spatial sequences, time sequences and image data, gridding basic geographic information, and obtaining grid geographic information; Based on the grid geographic information and the encoded spatial series, time series, image data and forecast correction data, multi-source vector data is obtained by encoding, and spatiotemporal features are extracted based on the multi-source vector data.

[0008] Furthermore, the extraction of spatiotemporal features based on multi-source vector data includes: obtaining corresponding tensor data through algorithm encoding based on the multi-source vector data, and then optimizing the tensor data to obtain spatiotemporal features in combination with a temporal attention mechanism.

[0009] Furthermore, the multi-source vector data adopts multi-modal feature extraction, and the multi-source vector data includes a vector matrix, a physical constraint matrix and an image feature matrix.

[0010] Furthermore, the calculation of the grid point forecast rainfall within a predetermined time based on the spatiotemporal characteristics includes: performing feature decoding through a deep temporal neural network based on the spatiotemporal characteristics, and calculating the grid point forecast rainfall within the predetermined time.

[0011] Furthermore, in step 2, the calculation of the grid rainstorm disaster resistance coefficient based on historical rainstorm disaster data and basic geographic information includes: analyzing historical rainstorm disaster data, obtaining rainstorm disaster factors through machine learning, gridding the basic geographic information, obtaining grid geographic information, fusing the rainstorm disaster factors and grid geographic information, and calculating the grid rainstorm disaster resistance coefficient in combination with the deterministic coefficient method.

[0012] Furthermore, the method of establishing an empirical model based on historical rainstorm disaster data in combination with basic geographic information to calculate the grid rainfall threshold includes: obtaining rainfall data of rainstorm meteorological disasters based on the historical disaster data, gridding the basic geographic information to obtain grid geographic information, establishing an empirical model based on the rainfall data in combination with the grid geographic information, and calculating the grid rainfall threshold.

[0013] Furthermore, in step 3, the algorithm model calculates the disaster probability of the grid points in the warning area according to the grid point forecast rainfall, the grid point rainstorm disaster tolerance coefficient, the grid point rainfall threshold and the on-site measured rainfall, and issues a warning signal.

[0014] In a second aspect, the present invention provides a rainstorm disaster warning system based on a rainstorm disaster tolerance coefficient, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method when executing the computer program.

[0015] Beneficial effects: The present invention provides a rainstorm disaster warning method and system based on rainstorm disaster tolerance coefficient. By calculating grid point forecast rainfall based on basic geographic information and multi-source data, combining the grid point disaster tolerance coefficient calculated with historical rainstorm disaster data, the grid point rainfall threshold and the rainfall measured on site, an empirical model is used to perform disaster warning from multiple data perspectives, thereby improving the refinement and accuracy of disaster warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a rainstorm disaster early warning method based on a rainstorm disaster bearing coefficient of the present invention; Figure 2 It is a schematic diagram of the structure of the AutoEncoder network used in the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate quantity restrictions, but indicate the existence of at least one. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0019] See also Figure 1 The embodiment of the present invention provides a rainstorm disaster warning method based on a rainstorm disaster bearing coefficient, comprising the following steps: Step 1: Obtain multi-source data and basic geographic information of the warning area, extract spatiotemporal characteristics based on the multi-source data and basic geographic information, and calculate the grid point forecast rainfall within the predetermined time according to the spatiotemporal characteristics; Among them, multi-source data include correction data and measurement data; The measurement data include satellite data, ground observation data, radar data and sounding data; In the correction data of multi-source data, the grid forecast correction product is mainly used, which mainly includes: precipitation, phase state, temperature, UV wind, relative humidity, thunderstorm, short-term heavy precipitation, hail, thunderstorm gale. It is mainly based on the precipitation meteorological element products of every hour in 24 hours output by the high-resolution rapid cycle assimilation numerical forecast system, multi-source observation, and multi-source data at the kilometer level such as radar and satellite. In this embodiment, 14-channel radar data and geographic information at four locations A, B, C, and D in the warning area at four time nodes 1, 2, 3, and 4 are mainly collected. Please refer to Tables 1 and 2 for details.

[0020] Table 1: Collected radar data

[0021] Table 2: Geographic information collected

[0022] The measurement data is integrated based on the meteorological geographic information data metafile, including: vector format meteorological map base map, terrain data of different scales, 30-meter surface coverage data, township boundary data, global terrain data, global 90-meter DEM data, comprehensive thematic data, etc. By mining the correspondence between meteorological data and geographic location, please refer to Figure 2 , the AutoEncoder network is used to encode the absolute geographic location, and the specific geographic location is used as an important feature factor; an efficient encoding and semantic extraction model is obtained, high-precision terrain information is mapped into a high-dimensional feature vector, a high-dimensional terrain feature vector dataset is constructed, and a reusable feature vector dataset is formed.

[0023] Specifically, the spatial sequence, time series and image data of rainfall are obtained based on multi-source data combined with spatial sequence feature extraction, the spatial sequence, time series and image data are encoded, the basic geographic information is gridded, and the grid geographic information is obtained; Based on grid geographic information, encoded spatial series, time series and image data, forecast and correction data are encoded to obtain multi-source vector data. Based on the multi-source vector data, algorithmic encoding is used to obtain the corresponding tensor data, and then combined with the temporal attention mechanism, the tensor data is optimized to obtain spatiotemporal features.

[0024] Multi-source vector data adopts a multimodal feature extraction method, which includes vector matrix, image feature matrix and physical constraint representation matrix. The vector matrix is ​​obtained by encoding the measurement data, the image feature matrix is ​​mainly obtained by encoding the weather radar puzzle data, and the physical constraint representation matrix is ​​formed by encoding the physical constraints between surface runoff, reservoir capacity, water level, precipitation and river runoff to form a probability mask matrix between 0 and 1.

[0025] Spatiotemporal feature extraction is to simultaneously obtain the feature information of spatial sequence data in the time dimension and the spatial dimension. This information refers to the tensor data corresponding to the description of spatiotemporal features obtained after each group of multi-source fusion data is encoded by the algorithm. The hybrid encoder is divided into a spatial transformer block and a temporal converter block. The spatial transformer block focuses on calculating the self-attention between each data, with the aim of learning the relationship between each data at each time node and promoting the association between geographic information and meteorological data. The temporal converter block calculates the self-attention between each time point, focusing on learning the temporal correlation information of the fused data, so that the acquired data is more accurate and effectively used for disaster warning.

[0026] Based on the spatiotemporal characteristics, feature decoding is performed through a deep time series neural network to calculate the grid point forecast rainfall within the predetermined time.

[0027] The hybrid spatiotemporal coding is used to construct a feature extractor, which pays attention to the internal connection of various data at the unit time node and the connection of data in time. Based on this spatiotemporal feature extractor, the prediction accuracy of rainfall is improved.

[0028] Spatial correlation learning of hybrid spatiotemporal coding: First, a linear embedding layer is used to project various data of each time node onto a high-dimensional feature. Then the spatial representation of this node is fed back into the spatial self-attention mechanism to simulate the correlation between all data and output a high-dimensional representation.

[0029] Temporal correlation learning of hybrid spatiotemporal coding: In order to inject effective motion trajectories into the learned representation, this embodiment considers the temporal correspondence of the data so as to model the correlation of the same joints over a long time series. Different categories of data are separated in the time dimension so that the evolution of each node data is a separate marker, and different categories of data of the fused data are modeled in parallel. From the perspective of the time dimension, the different evolutions of each type of data of the fused data are modeled separately to better represent the temporal correlation. In addition, by treating each type of data of the fused data as a separate marker, the dimensionality of the model can be reduced from N×dim to dim, and longer sequences can also be processed in the model.

[0030] The attention calculation of the query Q, keyword K, and value matrix V in each head of the transformer block in the hybrid spatiotemporal encoding MixSTE is expressed by the following formula; ; in , represents the attention mechanism, represents the normalized exponential function, N represents the type of fused data, is the dimension of each data, and T represents the matrix transpose. Multi-head attention is defined as follows: ; ; The linear projection weight is , MSA stands for multi-head self-attention mechanism, Concat stands for concatenation, and h stands for the number of independent groups. In the Transformer encoder of this embodiment, each joint marker From low dimension The Union Projection. Joint label p through the matrix Embed location information: ; Above The matrix dimension is transformed according to the different task requirements of modeling spatial correlation and temporal correlation. Representative layer standardization, Represents a linear embedding layer.

[0031] The feature decoding structure uses the deep time series model LSTNet, and adds the traditional autoregressive linear model on the basis of the nonlinear neural network part, making the nonlinear deep learning model more robust to time series with scale violations.

[0032] The structure consists of five modules: convolutional component, recursive component, recursive-skip layer, temporal attention layer, and autoregressive component.

[0033] Convolutional component: This component is a convolutional network without pooling, which aims to extract short-term patterns in the time dimension as well as local dependencies between variables.

[0034] Recursive component: The recursive component is a gated recurrent unit (GRU) that uses the RELU function as the activation function for updating the hidden state. The purpose is to extract the temporal features of the features extracted by the convolutional layer with time order through the GRU.

[0035] Recursive-skip layer: Traditional GRU has difficulty capturing long-term patterns, so a skip connection layer can be used. That is, through interval sampling, while keeping the length of the sampling sequence unchanged, it can look back for a longer time to capture long-term features.

[0036] Temporal Attention Layer: Recursive-skip layers require a predefined hyperparameter, which is disadvantageous in non-periodic or time series where the period length changes dynamically over time. To alleviate this problem, this embodiment adopts another approach, namely the attention mechanism. It learns a weighted combination of the hidden representations at each window position in the input matrix.

[0037] Autoregressive component: Due to the nonlinear nature of convolutional and recursive components, a major drawback of neural network models is that the scale of the output is insensitive to the scale of the input. In certain real-world datasets, the scale of the input signal keeps changing in a non-periodic manner, which greatly reduces the prediction accuracy of the neural network model. The final prediction of LSTNet can be decomposed into a linear part and a nonlinear part, where the linear part focuses on the local scale problem and the nonlinear part focuses on the recursive pattern. In the LSTNet architecture, the classic autoregressive (AR) model is used as the linear component.

[0038] Essentially, this is a linear layer. Finally, the output of the neural network part is added to the AR component to get the final prediction of LSTNet.

[0039] In this embodiment, based on the above, the grid point forecast rainfall at 30 minutes, 60 minutes, 90 minutes and 120 minutes is calculated according to the spatiotemporal characteristics, as shown in Table 3; Table 3: Grid point forecast rainfall within the scheduled time

[0040] Step 2: Obtain historical rainstorm disaster data in the warning area, calculate the grid point rainstorm disaster coefficient based on the historical rainstorm disaster data and basic geographic information, and establish an empirical model to calculate the grid point rainfall threshold based on the historical rainstorm disaster data combined with basic geographic information; Specifically, the rainfall data of rainstorm meteorological disasters are obtained based on historical rainstorm disaster data, the basic geographic information is gridded to obtain grid geographic information, an empirical model is established based on the rainfall data combined with the grid geographic information, and the grid rainfall threshold is calculated. For details, please see Table 4.

[0041] Table 4: Calculated grid point rainfall thresholds

[0042] Meteorological disasters such as floods, landslides, and mudslides have a strong intrinsic correlation with precipitation. Forecasting and warning of large-scale rainstorm meteorological disasters based on regional rainfall characteristics is an important way to prevent rainfall-type meteorological disasters. Among them, the most critical issue is to determine the critical rainfall threshold that induces rainfall-type meteorological disasters. The study of the critical rainfall threshold of rainfall-type meteorological disasters uses historical statistical methods to establish an empirical model.

[0043] The critical rainfall for disasters is confirmed based on the classification of topography. Therefore, the gridded geographic information is incorporated into the model calculation. According to the precipitation in all the short-term historical rainstorm meteorological disaster data, the correlation between disasters and precipitation is analyzed, and the critical rainfall threshold for disasters is assigned to the grid geographic information according to the precipitation and correlation at the disaster site.

[0044] Calculating the grid rainstorm disaster coefficient based on historical rainstorm disaster data and basic geographic information includes: analyzing historical rainstorm disaster data, obtaining rainstorm disaster factors through machine learning, gridding basic geographic information, obtaining grid geographic information, integrating rainstorm disaster factors and grid geographic information, and calculating the grid rainstorm disaster coefficient based on the deterministic coefficient method.

[0045] Specifically, by analyzing historical rainstorm disaster data, the rainstorm disaster factors are obtained through the machine learning Apriori association analysis algorithm; The coefficient of certainty method (CF) is a probability function, which is generally used to determine the contribution rate of each indicator when an event occurs. The calculation formula is as follows: ; Where: It represents the conditional probability of a certain disaster occurring in impact factor a, usually expressed as the ratio of the number of disasters occurring at a grid point in impact factor a to the area of ​​the grid point. It represents the probability of debris flow disaster occurring in a region, usually calculated by dividing the number of occurrences of a certain disaster in the entire region by the area of ​​the entire region.

[0046] Step 3: Obtain the rainstorm disaster prediction results for the warning area based on the grid forecast rainfall, grid rainstorm disaster resistance coefficient, grid rainfall threshold and field measured rainfall combined with the algorithm model.

[0047] For details, please refer to Table 5, and combine the algorithm model to obtain the rainstorm disaster prediction results of the four locations A, B, C, and D in the warning area; Table 5: Rainstorm disaster prediction results

[0048] The algorithm model calculates the disaster probability of the grid points in the warning area based on the grid forecast rainfall, grid rainstorm disaster tolerance coefficient, grid rainfall threshold and on-site measured rainfall, and issues a warning signal.

[0049] In this embodiment, the model combines the current accumulated precipitation at a certain grid point with the rolling output value of the precipitation forecast model to calculate and determine the probability that the total precipitation at the grid point will reach the critical threshold of the disaster precipitation.

[0050] ; In the above formula is the probability of precipitation disasters, is the current accumulated precipitation, is the disaster bearing coefficient of the grid point geographic information, Indicates hourly precipitation forecast starting from now.

[0051] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A rainstorm disaster early warning method based on rainstorm disaster bearing coefficient, characterized in that: The steps include: Step 1: Obtain multi-source data and basic geographic information of the warning area, extract spatiotemporal characteristics based on the multi-source data and basic geographic information, and calculate the grid point forecast rainfall within the predetermined time according to the spatiotemporal characteristics; Step 2: Obtain historical rainstorm disaster data in the warning area, calculate the grid point rainstorm disaster coefficient based on the historical rainstorm disaster data and basic geographic information, and establish an empirical model to calculate the grid point rainfall threshold based on the historical rainstorm disaster data combined with basic geographic information; Step 3: Obtain the rainstorm disaster prediction results for the warning area based on the grid forecast rainfall, grid rainstorm disaster resistance coefficient, grid rainfall threshold and field measured rainfall combined with the algorithm model.

2. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to claim 1 is characterized in that: The multi-source data includes correction data and measurement data; The measurement data includes satellite data, ground observation data, radar data and sounding data.

3. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to claim 2 is characterized in that: The method of extracting spatiotemporal features based on multi-source data and basic geographic information includes: extracting spatial sequences, time sequences and image data of rainfall based on multi-source data combined with spatial sequence features, encoding the spatial sequences, time sequences and image data, gridding basic geographic information, and obtaining grid geographic information; Based on the grid geographic information and the encoded spatial series, time series, image data and forecast correction data, multi-source vector data is obtained by encoding, and spatiotemporal features are extracted based on the multi-source vector data.

4. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to claim 3 is characterized in that: The method of extracting spatiotemporal features based on multi-source vector data includes: obtaining corresponding tensor data through algorithm encoding based on the multi-source vector data, and then optimizing the tensor data to obtain spatiotemporal features in combination with a temporal attention mechanism.

5. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to claim 4 is characterized in that: The multi-source vector data adopts multi-modal feature extraction, and the multi-source vector data includes a vector matrix, a physical constraint matrix and an image feature matrix.

6. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to any one of claims 1 to 5, characterized in that: The method of calculating the grid point forecast rainfall within a predetermined time according to the spatiotemporal characteristics includes: performing feature decoding through a deep temporal neural network based on the spatiotemporal characteristics, and calculating the grid point forecast rainfall within a predetermined time.

7. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to any one of claims 1 to 5, characterized in that: In step 2, the calculation of the grid rainstorm disaster resistance coefficient based on historical rainstorm disaster data and basic geographic information includes: analyzing historical rainstorm disaster data, obtaining rainstorm disaster factors through machine learning, gridding basic geographic information, obtaining grid geographic information, fusing the rainstorm disaster factors and grid geographic information, and calculating the grid rainstorm disaster resistance coefficient in combination with the deterministic coefficient method.

8. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to any one of claims 1 to 5, characterized in that: The method of establishing an empirical model based on historical rainstorm disaster data combined with basic geographic information to calculate the grid rainfall threshold includes: obtaining rainfall data of rainstorm meteorological disasters based on the historical rainstorm disaster data, gridding the basic geographic information to obtain grid geographic information, establishing an empirical model based on the rainfall data combined with the grid geographic information, and calculating the grid rainfall threshold.

9. The rainstorm disaster early warning method based on the rainstorm disaster bearing coefficient according to any one of claims 1 to 5, characterized in that: In step 3, the algorithm model calculates the disaster probability of the grid points in the warning area according to the grid point forecast rainfall, the grid point rainstorm disaster tolerance coefficient, the grid point rainfall threshold and the on-site measured rainfall, and issues a warning signal.

10. A rainstorm disaster warning system based on a rainstorm disaster tolerance coefficient, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images