Prediction of electricity consumption in cement production: a time-varying delay deep belief network prediction method

Abstract

An important energy consumption index in cement production process is electricity consumption whose accurate prediction is of great significance to optimize production. However, it is difficult to establish an accurate electricity consumption forecasting model in cement production, for some problems such as the time delay, uncertainty and nonlinearity existing in the cement manufacturing process. To address the problems, we propose an electricity consumption prediction model based on time-varying delay deep belief network (TVD-DBN). In order to eliminate the influence of time-varying delay in the cement production process prediction, time series containing the time-varying delay is integrated into the input layer. In addition, we use the restricted Boltzmann machine (RBM) to capture the features, and after the pretraining of RBM, the gradient descent algorithm is used to fine-tuning the parameters of network. Through the above methods, the forecast of electricity consumption is realized in cement manufacturing process. Experiment results show that our approach TVD-DBN has higher accuracy, stronger robustness and better generalization ability in the prediction of cement electricity consumption compared with the least squares support vector machine and the deep belief network.