Research On Forest Fire Danger Prediction Based On Fuzzy Pre-classification And Extreme Learning Machine

Forest fires are a type of serious natural disaster,which brings great security risks to the healthy and stable development of the natural economy and ecology environment.It is an important means to reduce loss that timely and accurate prediction and monitoring of forest fires.At present,the researches on the field of forest fire danger prediction have achieved good results,including forest fire prediction and trend analysis,real-time fire association rules mining,fire event sequence pattern detection,cluster analysis and ignition point identification,etc.These studies mainly use algorithms or techniques such as the improvement of regression methods,integration of regression methods and fusion of multi source data to improve the accuracy of fire danger prediction.However,in the case of increasing data on forest fire,it will lead to problems such as low prediction accuracy of the model,weaker robustness,and increased computational complexity of the algorithm.That is,the practicality of the above model is poor,which is not conducive to manage the growing forest fire danger database.In this paper,the qualitative and quantitative prediction methods of forest fire danger based on Fuzzy Pre-classification and Extreme Learning Machine(ELM)model are proposed,and the model is optimized from the aspects of improving the accuracy,enhancing the robustness and reducing the computational complexity respectively.The experimental results show that with the continuous expansion of forest fire danger database,the model has higher prediction accuracy and stronger robustness.The main work of this article includes:(1)The fuzzy pre-classification algorithm is used to manage the forest fire database.This paper defines the parameters of input membership function according to the value of the Fire Weather Index(FWI).For each experimental sample,the fuzzy classifier classifies it into the corresponding sub-database.The output of the classifier is respectively S group,M group,and L group.Forest fire danger rating and burned area are predicted separately for each sub-database sample.(2)The weighted Extreme Learning Machine(W-ELM)algorithm is used to realize the prediction of the forest fire danger rating.First,the t-SNE algorithm is used for feature extraction to obtain the main component sequence as the network input of the ELM.Next,the training data is re-weighted and different penalty coefficients are added to the training errors corresponding to different inputs.Memetic algorithm is also used to optimize connection weights and bias values between input layer and hidden layer in the network of ELM,and classify hidden layer nodes of W-ELM,the hidden layer divided into multiple sub-modules.Finally,the output of the ELM is the forest fire danger rating.(3)Based on the prediction of forest fire danger rating,the burned area of each category(fire danger rating)is predicted separately.The feature sets are divided into four categories,Each continuous feature is segmented based on the best mean,variance,and covariance,and a stable interval prediction is performed on the target numerical variable with a given confidence level.Each feature set is combined with the Interval Prediction Tree algorithm to predict the burned area.