Prediction Model of Iron Ore Pellet Process Based on MTCN-Informer
The prediction of finished pellet flow is the key to the production process,which determines the efficiency and output of the whole production.Iron ore pellet chain grate machine—rotary kiln is one of the important processes for producing iron ore to prepare high-quality ferroalloy.It has the characteristics of large time lag,complex parameters,complex coupling relationship,etc.,and the flow rate of finished pellets fluctuates violently,making the flow rate of pellets difficult to predict.For this reason,we use the moving average filter to smooth the fluctuating data,and the mutual information method performs feature selection on complex parameters,and then uses the Informer pellet flow prediction model based on the self-attention mechanism,which reduces the time of the traditional self-attention mechanism complexity and improves the efficiency of model training.At the same time,in view of the problem that the probabilistic sparse self-attention mechanism of the Informer model is difficult to grasp the long-term sequence fluctuations,the extended information dependence of the long-term sequence is extracted through the TCN time convolution network,and the context information is processed by combining the Informer encoding and decoding network,thereby completing accurate prediction of pellet flow.Through the experimental analysis of the actual factory data,it can be seen that compared with traditional deep learning models such as recurrent neural networks,the proposed integrated model is the best in terms of prediction accuracy and stability.
pellet flow predictionfeature selectiontemporal convolutional networkencoding and decoding networkself-attention mechanism
国家科技期刊平台
NSTL
万方数据
