Energy optimization and prediction modeling of petrochemical industries: An improved convolutional neural network based on cross-feature

The petrochemical industry is the top priority of the national economy and sustainable development. For the purpose of improving the energy efficiency in the petrochemical industry, an energy optimization and prediction model based on the improved convolutional neural network (CNN) integrating the cross-feature (CF) (CF–CNN) is proposed. The CF can combine the correlation between features to obtain the input of the CNN, which can avoid over-fitting problems caused by fewer features. Then the CNN is designed as a three-layer structure and the Rectified Linear Unit (ReLU) is introduced to achieve better generalization capability and stability with boiler fluctuations in the petrochemical industry. The developed method has better performances of modeling accuracy and applicability than that of the back-propagation (BP) neural network and the extreme learning machine (ELM) on University of California Irvine (UCI) benchmark datasets. Furthermore, the developed method is applied to establish an energy optimization and prediction model of ethylene production systems in the petrochemical industry. The experimental results testify the capability of the proposed method. Meanwhile, the average relative generalization error is 2.86%, and the energy utilization efficiency increases by 6.38%, which leads to reduction of the carbon emissions by 5.29%.