Research And Application Of WELM In Coal Mine Water Inrush Prediction

There are many factors that affect coal mine water inrush.There are certain overlaps between the many influencing factors,which led to the indefinite and fuzzy similarity of most influencing factors,and factors formed a nonlinear relationship.Therefore,it is difficult to solve the problem of coal mine water inrush prediction by using classical mathematical models.In response to this phenomenon,many scholars proposed that using artificial neural network model to solve the problem of coal mine water inrush prediction.Classical artificial neural network models such as BP neural network and SVM neural network have certain deficiencies in practical model training.For example,the BP neural network sometimes falls into a local extremum,which affects training success.Meanwhile,the convergence speed of the algorithm is slower.The support vector machine(SVM)model is only suitable for the problem with small sample's size,and it is difficult to solve the problem of multiple classifications.Extreme learning machine(ELM)network model makes up for the deficiencies of the above two models.The ELM algorithm can not only obtain the globally unique optimal solution,but also ensure the training speed and the prediction speed are very fast.At the same time,it can also handle the multi-classification problem of a large number of sample data.However,the historical data of water inrush in coal mines is unbalanced data,that is,most data were collected without water inrush occurred,and a few data were collected when water inrush occurred.The classic ELM model cannot handle this problem effectively.Weighted Extreme Learning Machine(WELM)not only inherits the advantages of ELM,but also can effectively handle unbalanced data by weighting samples.Therefore,this paper uses the WELM model to solve the problem of coal mine water inrush prediction.The main work of the dissertation is as follows:(1)Studying on texture and existing prediction technology about coal mine water inrush;Analyzing and summarizing the main factors affecting coal mine water inrush;Collecting data,preprocessing the original data,including using local linear embedding to reduce the dimension of the data,and normalizing the data using a normalized method.(2)Building WELM model.Experiment to determine the activation function of hidden layer neurons in WELM network;Experiment to optimize parameters of WELM network by using differential evolution algorithm;This paper compares the DE-WELM prediction model with other models to demonstrate the feasibility and superiority of the model in theory.(3)Inputing processed data into the established DE-WELM model for training,validation and prediction,and analyzing the prediction results.Meanwhile,inputing processed data into the SVM prediction model for experiment.Comparing the results of the two experiments to verify the superiority of the DE-WELM model in solving coal mine water inrush prediction.