Deep hybrid modeling of chemical process: Application to hydraulic fracturing

Process modeling began with the use of first principles resulting in 'white-box' models which are complex but accurately explain the dynamics of the process. Recently, there has been tremendous interest towards data-based modeling as the resultant 'black-box' models are simple, and easy to construct, but their accuracy is highly dependent on the nature and amount of training data used. In order to balance the advantages and disadvantages of 'white-box' and 'black-box' models, we propose a hybrid model that integrates first principles with a deep neural network, and applied it to hydraulic fracturing process. The unknown process parameters in the hydraulic fracturing process are predicted by the deep neural network and then utilized by the first principles model in order to calculate the hybrid model outputs. This hybrid model is easier to analyze, interpret, and extrapolate compared to a 'black-box' model, and has higher accuracy compared to the first principles model. Published by Elsevier Ltd.