Theory Of Big Data Law Mining And Its Application In Coal Blending For Coking

For weaknesses such as difficult integration analysis and difficult extraction of effective values for massive experimental data in the industrial blended coal coking field,this paper combines current research hotspots and predicts results in an efficient and accurate manner by using different neural networks and improved networks based on research on different neural network models.This paper further improves practical efficiency of the algorithms,saves the experimental time and production cost,and provides the coke quality prediction model based on the neural network.This model integrates data cleaning,data analysis and quality prediction functions for industrial production in the coke field.The main research contents are described as follows:First,to better analyze the experimental data,this paper studies theories and improvements on different neural network algorithms,discusses the data collection,integration and cleaning methods and flows in the coke field under the big data environment,and gives the cleaning instances of the coke blending and coking data.Secondly,to accurately analyze and process the experimental data,this paper proposes the cascade BP neural network coke quality prediction model with cross validation based on the original BP neural network model.This model can predict coke quality parameters according to the blended coal quality parameters by using the cross validation cascade idea.The forward and backward feedback interconnection structure among different tiers can ensure accurate determination and inclusions of incorrect neurons.The general prediction error of the new model is within 5%,so this model is precise and effective.Thirdly,to accelerate the approximation search and improve the coke quality prediction effect,this paper designs the normalized network training method based on Sigmoid function sensitive area for computing complexity and training error performance in order to delete weights,which have extremely small influences on performance and may lead to over-fitting.This paper also builds the coke quality prediction mode based on BP neural network,cascade BP neural network with cross validation and RBF neural work and compares their prediction effects.Finally,to improve the low confidence level of the coke quality prediction results driven by the data,the inherent causal relationships of blended coal coking are hierarchically described.The initial neural network structure is created based on the known domain knowledges,next the cascade neural network is used to train the initial network for perfect structure.The experimental simulation results show that the average error of the coke St,d,M10 and CRI predicted by using the cascade feedforward network structure based on the domain knowledge representation method is within 5%,which is significantly better than the prediction results of the traditional network structure.