Severe convective weather identification method and system based on neural network

Abstract

The invention discloses a severe convective weather identification method and system based on a neural network, and relates to the technical field of meteorological monitoring. The method comprises the following specific steps: acquiring detection data including multi-source observation data and high-resolution numerical forecasting data; performing data fusion and data preprocessing on the detection data to obtain first data; performing manual labeling on the first data to establish a training label to construct a sample data set; building a convolutional neural network model, and inputting the training set into the model for training to obtain a severe convective weather intelligent identification model; evaluating the severe convective weather intelligent identification model by using the test set to obtain a final severe convective weather intelligent identification model; and inputting real-time acquired real-time observation data into the final severe convection weather intelligent identification model for severe convection weather identification. According to the method, severe convective weather identification is carried out based on the deep neural network technology, automatic and quantitative prediction can be carried out, and the reliability and timeliness of severe convective weather identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological monitoring, and in particular to a method and system for identifying severe convective weather based on a neural network. Background Art

[0002] Severe convective weather refers to sudden, violent, and extremely destructive weather, often accompanied by strong convective disasters such as thunderstorms, gales, hail, tornadoes, and local heavy rainfall. It is often very destructive. The identification, nowcasting, and early warning of severe convective weather play an extremely important role in meteorological disaster prevention and mitigation. Effectively and quickly improving the recognition and forecasting rate of severe convective weather can reduce the loss of people’s lives, property, and industrial and agricultural production. At present, the identification of severe convective weather mainly relies on traditional meteorological observation data or radar echo characteristics, which have weak early warning capabilities and low early warning accuracy. Therefore, for those skilled in the art, how to achieve accurate identification of severe convective weather is an urgent problem to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide a severe convective weather identification method and system based on a neural network to solve the problems raised in the background technology.

[0004] To achieve the above object, the present invention provides the following solutions: On the one hand, a method for identifying severe convective weather based on a neural network is provided, and the specific steps include the following:

[0005] Acquiring detection data, wherein the detection data includes multi-source observation data and high-resolution numerical prediction data;

[0006] Performing data fusion and data preprocessing on the detection data to obtain first data;

[0007] Manually annotating the first data with severe convective weather to establish training labels, thereby constructing a sample data set, and dividing the sample data set into a training set and a test set;

[0008] Building a convolutional neural network model, inputting the training set into the convolutional neural network model for training, and obtaining a severe convective weather intelligent recognition model;

[0009] Using the test set to evaluate the severe convective weather intelligent recognition model, to obtain a final severe convective weather intelligent recognition model;

[0010] The real-time observation data collected is input into the final severe convective weather intelligent identification model for severe convective weather identification.

[0011] Preferably, the multi-source observation data include satellite data, radar mosaic data, and lightning location data; and the high-resolution numerical prediction data are radar combined reflectivity data.

[0012] Preferably, the specific steps of performing data preprocessing on the detection data are:

[0013] Performing abnormal data cleaning and denoising operations on the acquired detection data to obtain a first data set;

[0014] The first data set is enhanced by using translation, rotation, and deformation to obtain the first data.

[0015] Preferably, the lightning location data is discretized data, and grid processing is performed on the lightning location data using a neighbor difference method.

[0016] Preferably, the error between the true value of the radar echo intensity in the detection data and the average value of the predicted value output during the training process is used as a parameter of the loss function in the convolutional neural network model; and the convolutional neural network model is iterated multiple times according to the loss value generated by the loss function.

[0017] Preferably, identifying severe convective weather includes identifying the type of disaster of severe convective weather and identifying the intensity of severe convective weather.

[0018] Preferably, the convolutional neural network model includes a feature extraction module, an attention mechanism module and a multi-scale residual module; wherein the feature extraction module is used to extract temporal features and contextual spatial features in the training set; the attention mechanism module is used to dynamically modulate the weights of each channel in the convolutional neural network model for feature enhancement; the multi-scale residual module is used to process the data image output by adding the output of the feature extraction module and the output of the attention mechanism module by stacking multiple residual blocks.

[0019] On the other hand, a severe convective weather identification system based on a neural network is provided, comprising a data acquisition module, a preprocessing module, a data set construction module, a training module, a testing module, and an identification module; wherein:

[0020] The data acquisition module is used to acquire detection data, wherein the detection data includes multi-source observation data and high-resolution numerical prediction data;

[0021] The preprocessing module is used to perform data fusion and data preprocessing on the detection data to obtain first data;

[0022] The data set construction module is used to manually annotate the first data with severe convective weather to establish training labels, thereby constructing a sample data set, and dividing the sample data set into a training set and a test set;

[0023] The training module is used to build a convolutional neural network model, input the training set into the convolutional neural network model for training, and obtain a severe convective weather intelligent recognition model;

[0024] The testing module is used to evaluate the severe convective weather intelligent recognition model using the test set to obtain a final severe convective weather intelligent recognition model;

[0025] The identification module is used to input the real-time observation data collected into the final severe convective weather intelligent identification model to perform severe convective weather identification.

[0026] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: severe convective weather identification based on deep neural network technology can be automatically and quantitatively predicted, thereby improving the reliability and timeliness of severe convective weather identification, which is of great significance for timely warning and prevention of disasters caused by severe convective weather; by fusing multi-source data, the deep learning model can provide more accurate and timely forecast results. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0028] Figure 1 is a flow chart of the method of the present invention;

[0029] Figure 2 It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0031] The purpose of this invention is to provide a severe convective weather identification method based on neural network. Figure 1 As shown, the specific steps include the following:

[0032] S1. Acquire detection data, wherein the detection data includes multi-source observation data and high-resolution numerical prediction data;

[0033] S2. Performing data fusion and data preprocessing on the detection data to obtain first data;

[0034] S3, manually labeling the first data with severe convective weather to establish training labels, thereby constructing a sample data set, and dividing the sample data set into a training set and a test set;

[0035] S4. Build a convolutional neural network model, input the training set into the convolutional neural network model for training, and obtain a severe convective weather intelligent recognition model;

[0036] S5. Using the test set to evaluate the severe convective weather intelligent recognition model, and obtain the final severe convective weather intelligent recognition model;

[0037] S6. Inputting the real-time observation data collected into the final severe convective weather intelligent identification model to identify severe convective weather.

[0038] Furthermore, the multi-source observation data includes satellite data, radar mosaic data, and lightning location data; the high-resolution numerical forecast data is radar combined reflectivity data. Among them, the satellite data is meteorological satellite data, and in this embodiment, the sunflower-8 satellite imager observation data is used; the radar mosaic data is Doppler radar national combined reflectivity mosaic data, including this reflectivity mosaic data, liquid water content data, etc.; the lightning location data comes from the ground lightning observation data provided by the National Lightning Monitoring Network.

[0039] Furthermore, in step S2, the specific steps of performing data preprocessing on the detection data are:

[0040] S21, performing abnormal data cleaning and denoising operations on the acquired detection data to obtain a first data set;

[0041] S22. Perform data enhancement on the first data set by using translation, rotation, and deformation to obtain first data.

[0042] High-quality data is the basis of application. To ensure the quality of the training data set, the multi-source observation data collected and compiled in this data set have all been quality controlled. Data cleaning After quality control of the observation data, the data is re-examined and verified from the perspective of data set application. The acquired detection data is cleaned from the perspectives of data integrity, data consistency, time continuity, logical relationship rationality, non-convective weather process filtering, and Internet data verification. Duplicate information, incomplete information on spatiotemporal attributes, erroneous information on logical relationships between elements, time discontinuity information, and weather process information caused by non-convective factors are eliminated to ensure the validity and correctness of the multi-source training samples of the data set.

[0043] It should be noted that the lightning location data is discrete data, and it is necessary to use the adjacent difference method to grid the lightning location data. When lightning is used as a prediction factor, if n lightning occurs within a range of Rkm around a grid point, the number of lightning at this grid point is recorded as n, and the lightning density value on the grid point is obtained; when lightning is used as a mark, if lightning occurs within a range of Rkm around a grid point, this grid point is marked as 1, otherwise it is marked as 0. In this embodiment, R is set to 20km. Tests show that if R is too small, there will be more missed reports; if R is too large, the false reports will be more serious.

[0044] Furthermore, in order to improve the accuracy of model recognition, the error between the actual value of the radar echo intensity in the detection data and the average value of the predicted value output during the training process is used as the parameter of the loss function in the convolutional neural network model; according to the loss value generated by the loss function, the convolutional neural network model is iterated multiple times.

[0045] Furthermore, the convolutional neural network model includes a feature extraction module, an attention mechanism module and a multi-scale residual module; wherein the feature extraction module is used to extract temporal features and contextual spatial features in the training set; the attention mechanism module is used to dynamically modulate the weights of each channel in the convolutional neural network model for feature enhancement; the multi-scale residual module is used to process the data image output by adding the output of the feature extraction module and the output of the attention mechanism module by stacking multiple residual blocks.

[0046] Furthermore, the identification of severe convective weather includes identifying the type of disaster of severe convective weather and identifying the intensity of severe convective weather.

[0047] Furthermore, in step S5, the calculation formula for evaluating the severe convective weather intelligent identification model is:

[0048]

[0049] Among them, S represents a sample score of the severe convective weather intelligent identification model, α represents the true value of the predicted sample, β represents the predicted value of the predicted sample, β is a number greater than or equal to 0. If the calculated β is less than 0, it is considered to be 0. For a data containing several samples, the evaluation score is obtained by summing S and then taking the average.

[0050] On the other hand, a severe convective weather identification system based on a neural network is provided. Figure 2 As shown, it includes a data acquisition module, a preprocessing module, a data set construction module, a training module, a testing module, and a recognition module; wherein,

[0051] A data acquisition module, used to acquire detection data, wherein the detection data includes multi-source observation data and high-resolution numerical prediction data;

[0052] A preprocessing module, used for performing data fusion and data preprocessing on the detection data to obtain first data;

[0053] A data set construction module, used to manually annotate the first data with severe convective weather to establish training labels, thereby constructing a sample data set, and dividing the sample data set into a training set and a test set;

[0054] The training module is used to build a convolutional neural network model, input the training set into the convolutional neural network model for training, and obtain a severe convective weather intelligent recognition model;

[0055] A testing module, used to evaluate the severe convective weather intelligent recognition model using a test set to obtain a final severe convective weather intelligent recognition model;

[0056] The recognition module is used to input the real-time observation data collected into the final severe convective weather intelligent recognition model to perform severe convective weather recognition.

[0057] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A severe convective weather identification method based on neural network, characterized in that: The specific steps include the following: Acquiring detection data, wherein the detection data includes multi-source observation data and high-resolution numerical prediction data; Performing data fusion and data preprocessing on the detection data to obtain first data; Manually annotating the first data with severe convective weather to establish training labels, thereby constructing a sample data set, and dividing the sample data set into a training set and a test set; Building a convolutional neural network model, inputting the training set into the convolutional neural network model for training, and obtaining a severe convective weather intelligent recognition model; Using the test set to evaluate the severe convective weather intelligent recognition model, to obtain a final severe convective weather intelligent recognition model; The real-time observation data collected is input into the final severe convective weather intelligent identification model for severe convective weather identification.

2. A severe convective weather identification method based on neural network according to claim 1, characterized in that: The multi-source observation data include satellite data, radar mosaic data, and lightning location data; the high-resolution numerical prediction data are radar combined reflectivity data.

3. The method for identifying severe convective weather based on a neural network according to claim 1, characterized in that: The specific steps of performing data preprocessing on the detection data are: Performing abnormal data cleaning and denoising operations on the acquired detection data to obtain a first data set; The first data set is enhanced by using translation, rotation, and deformation to obtain the first data.

4. The method for identifying severe convective weather based on a neural network according to claim 2, characterized in that: The lightning location data is discretized data, and the lightning location data is gridded by using a neighboring difference method.

5. The method for identifying severe convective weather based on a neural network according to claim 1, characterized in that: The error between the true value of the radar echo intensity in the detection data and the average value of the predicted value output during the training process is used as a parameter of the loss function in the convolutional neural network model; and the convolutional neural network model is iterated multiple times according to the loss value generated by the loss function.

6. The method for identifying severe convective weather based on a neural network according to claim 1, characterized in that: Identifying severe convective weather includes identifying the type of disasters caused by severe convective weather and identifying the intensity of severe convective weather.

7. The method for identifying severe convective weather based on a neural network according to claim 1, characterized in that: The convolutional neural network model includes a feature extraction module, an attention mechanism module and a multi-scale residual module; wherein the feature extraction module is used to extract the temporal features and contextual spatial features in the training set; the attention mechanism module is used to dynamically modulate the weights of each channel in the convolutional neural network model for feature enhancement; the multi-scale residual module is used to process the data image output by adding the output of the feature extraction module and the output of the attention mechanism module by stacking multiple residual blocks.

8. A severe convective weather identification system based on neural network, characterized in that: It includes data acquisition module, preprocessing module, data set construction module, training module, testing module and recognition module; among them, The data acquisition module is used to acquire detection data, wherein the detection data includes multi-source observation data and high-resolution numerical prediction data; The preprocessing module is used to perform data fusion and data preprocessing on the detection data to obtain first data; The data set construction module is used to manually annotate the first data with severe convective weather to establish training labels, thereby constructing a sample data set, and dividing the sample data set into a training set and a test set; The training module is used to build a convolutional neural network model, input the training set into the convolutional neural network model for training, and obtain a severe convective weather intelligent recognition model; The testing module is used to evaluate the severe convective weather intelligent recognition model using the test set to obtain a final severe convective weather intelligent recognition model; The identification module is used to input the real-time observation data collected into the final severe convective weather intelligent identification model to perform severe convective weather identification.

Images