Research On Lightning Disaster Risk Assessment And Lightning Activity Prediction Based On Machine Learning

In recent years,lightning disasters have caused enormous economic losses and casualties all over the world,posing great threat to human activities.Therefore,how to reduce the occurrence of lightning disasters effectively has become the focus of researchers.Lightning disaster risk assessment and lightning prediction research are two major approaches to improve the lightning defense ability of the whole society.This paper analyzes the current research status and shortcomings of lightning disaster risk assessment,and introduces neural network algorithm to improve its performance.This paper uses ConvLstm network to build a prediction framework,and proposes an algorithm to solve the problem of data sparsity,which improves the accuracy of lightning prediction.The main research results are as follow:For the problems of unscientific weight distribution and rough evaluation resolution in traditional lightning disaster risk assessment methods.Firstly,this paper gathers a large number of experimental data based on the lightning disaster data from 2015 to 2016,combined with the ADTD lightning location data,geographical environment data and demographic and economic data over the past few years.Through the operation of geographic raster slicing and data rasterization,the situation of each geographic raster is quantified,which lays the foundation for the realization of fine-grained evaluation.Secondly,this paper introduces the neural network algorithm and took the lightning disaster data as the label,combined with the above multi-type experimental data to fit the high-order function between each impact factor and the actual situation,so that the fitting function could express the relationship between the impact factor and the actual lightning disaster more accurately.Also,this paper uses the variable-controlling method to explore the impact of the selection of negative sample data after clustering operation and the combination of different impact factors on the final evaluation accuracy.Finally,this paper presents the evaluation results visually.For the problem of low accuracy of lightning prediction.First,in terms of data,this paper combines ADTD lightning location data of Changsha and its surrounding areas from 2010 to2019 as samples.The raster operation is carried out and divided according to the time interval to construct a three-dimensional data structure based on the time space dimension.Second,this paper proposes a data enhancement algorithm to solve the problem of sparse data matrix in space dimension.The algorithm not only solves the problem of data sparsity,but also makes the data matrix of each time capable of integrating the spatial information of previous multiple times,which makes the expression of time information more comprehensive.Finally,this paper builds a parallelogram prediction framework with the help of ConvLstm model,so that the model can continuously carry out real-time prediction with less computation.According to the calculation,the accuracy of the model is 74.6%,and show that our model has a good accuracy in the area with dense data.