Reaserch On Mill Load Identification Of Combined Grinding System

The mill is the core equipment of combined grinding system. However, most of the mills are low efficiency and high energy consumption, the grinding process has the characteristics of high coupling, so the accurate identification of the mill load is very important. In order to get the optimal model of mill load, four kinds of recursive least square algorithm, RBF neural network and T-S fuzzy model were adopted to identify the mill load. The main work completed of this subject is summed up as follows:Summarize the development of mill load identification and the combined grinding system, analysis a large number of historical data, I get a conclusion that the mill’s main motor current well reflects the mill load. Analysis the main variables that influence the main motor current, and they are given total, powder selecting machine’s speed, circulating blower’s speed, discharge valve’s open degree and hoist current of fly ash storage. Finally I get a conclusion that given total and powder selecting machine’s speed are key variables, the other three are the uncertain factors. In this paper, all of the load model’s input variables are given total and powder selecting machine’s speed, the output variable is the main motor current.First of all, I use four kinds of recursive least square algorithm to get the mill load and they are: recursive least square, forgetting factor recursive least squares, finite memory recursive least square method and the bias compensation recursive least square. By analysising simulation results, the models based on forgetting factor recursive least squares and bias compensation recursive least square can well track the main motor current in this working condition. The modle based on bias compensation recursive least square is most accurate. The other two’s fitting error is relatively larger and is not suitable for modeling in this working condition.Then I use RBF neural network to get the mill load. RBF neural network has the advantage of nonlinear system’s well approximation performance, so I use three kinds of RBF neural network models based on Gauss function, multi quadric function and inverse multi quadric function. Network’s center, weight and expansion coefficient are both trained with the gradient descent method. The number of neurons is determined through repeated experiments. By analysising the mean error, mean square error and other performance indicators, RBF neural network models based on Gauss kernel function is most accurate and it’s suitable for modeling in this working condition.Finally, I use T-S fuzzy model to get the mill load. Input variables are divided into four sub space with the fuzzy C-mean clustering algorithm. The fuzzy consequent’s parameters are identified with the weighted least squares. In order to contrast three identification methods’ effectiveness and model’s accuracy, I get the models with weighted least squares method and RBF neural network based on Gauss kernel function in the same period of history datas. Compared to the models based on weighted least squares method and RBF neural network model, the simulation result shows that T-S fuzzy model can well reflect the main motor current in this working condition.