Energy consumption prediction in cement calcination process: A method of deep belief network with sliding window. (15th September 2020)

• Record Type: Journal Article
• Authors: Hao, Xiaochen; Guo, Tongtong; Huang, Gaolu; Shi, Xin; Zhao, Yantao; Yang, Yue
• Is Part Of: Energy. Volume 207(2020)
• Publication Date: 2020-09-15
• Subjects: Deep belief network -- Sliding window -- Energy consumption prediction -- Multiple-index prediction; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.118256
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Electricity consumption and coal consumption are two important indicators in the cement calcination process. Modeling predictions of cement energy consumption support efforts aimed at understanding energy use and energy conservation. However, due to the three characteristics of cement: time-varying delay, non-linearity and uncertainty, it is very difficult to establish accurate energy consumption prediction models. To solve the above problems, a multiple-index energy consumption prediction model based on sliding window deep belief network (SW-DBN) is proposed in this paper. Specifically, to avoid studying complex problem of time-varying delay, the sliding window method is introduced to deep belief network, which combines the previous and current variable data into time series data. As a result, all temporal information related to the energy consumption data is fed to the input layer of deep belief network. Then deep belief network is utilized to establish the multiple-index energy consumption prediction model on the temporal information, which is capable of predicting electricity consumption and coal consumption simultaneously. Experimental results show that the proposed model obtains improvement for multiple-index energy consumption prediction model in cement calcination process. Highlights: Sliding window technique is applied to solve the time-varying delay problem. Nonlinear features extracted by deep belief network can improve model performance. The proposed method is applied to predict two indexes simultaneously. Sliding window deep belief network performs better than other traditional models. The prediction results can provide useful information for optimization system.