Predicting Calorific Value And Ash Fusion Temperature Of Coal Based On Neural Network

This paper mainly studies the predictive modeling methods of calorific value and ash fusion temperature based on neural network.The calorific value of coal can provide technical basis for calculating parameters like boiler heat balance, boiler thermal efficiency and boiler capacity etc. The calorific value of coal has the different implication such as bomb calorific value, gross calorific value; low heat value and constant humidity ashless gross calorific value. Among them, bomb calorific value can be immediately measured by bomb calorific value calorimeter. Based on the detailed study of working principle and main structure of bomb calorific value calorimeter, it is shown that this technique has a serious shortage, which demands high environmental requirement and high cost. Based on industrial analysis of coal, some scholars obtained an empirical formula for the calorific value of coal by using the linear regression method. They found that the primary factors which influence the calorific value are the water content, ash content and volatile content in coal. This paper analyzes these industrial components.Ash fusion temperature of coal is within the temperature range characterized by deformation temperature (DT), soften temperature (ST) and flowing temperature (FT). The common methods to measure ash fusion temperature are pyramid method and thermal microscope method. In order to predict ash fusion temperature more quickly and efficiently, some scholars developed several empirical formulas based on the effect of chemical composition. This paper dealt with the effect of oxides:SiO2、Al2O3、Fe2O3、 TiO2、Na2O、K2O、MgO and CaO。According to the study of calorific value and ash fusion temperature of coal, it is found that the relationship between various effective factors and calorific value or ash fusion temperature is not linear. In this case, empirical formulas would result in unneglectable error. The neural network based on in-depth study of structures and algorithms can easily solve the nonlinear problem.The paper mainly analyzes the structures and algorithms of BP neural network and RBF neural network. On this basis, this paper makes the prediction models of net calorific value (Qnet, ar) and soften temperature (ST). Through experiments, it is found that the prediction based on Neural Network is superior to those based on empirical formulas. RBF neural network has better prediction performance with fewer samples. In addition, the quality of the samples, includes quantity and comprehensiveness, has stronger influence on predictive ability and generalization ability of models.In order to further improve prediction models’performances, this paper studies particle swarm optimization (PSO) and uses it to optimize the weight of RBF neural network. Then, the paper uses the optimization model to make predications of calorific value and ash fusion temperature, and gets the prediction results which are better than those based on conventional RBF algorithm.