A dynamic forecast method for linz donaniz gasholder level based on recurrent neural network

作者：Lv Zhimin; Wang Ziyang^*; Liu Qin

来源：Journal of Information and Computational Science, 2014, 11(2): 649-658.

DOI：10.12733/jics20102853

• Dynamic behaviors
• Dynamic characteristics
• Elman neural network
• Gasholder level predictions
• Incremental learning
• Iron and steel enterprise
• Modeling and forecast
• Trend forecasting

A recurrent neural network with dynamic characteristic is proposed to forecast the gas holder level and to solve the gas holder prediction problems in iron and steel enterprises. The gas holder is not only concerned with the input gas production and consumption, but also related to the previous gasholder level and dynamics of gas pipe network, thus we adopted Elman neural network with dynamic behavior and added the feedback connection between the output layer and the input layer. In this paper, Elman neural network is improved to adapt the practical mechanism of the gas holder. In addition, the traditional cycle learning method is replaced by the incremental learning method to train samples so as to enhance the learning ability of the model and achieve the results of dynamic modeling and forecast. The validation experiment proved that the method is suitable for forecasting gasholder level.

• 出版日期2014
• 单位北京科技大学