(348f) An Integrated Prediction and Process Optimization Software Platform for Highly Efficient Neural Network Development in Chemical Engineering and Related Fields

Artificial neural networks are transforming the chemical industry with rapid applications in process design, optimization, and control. These computer based neural networks are often employed to autonomously discover hidden trends and nonlinear relationships in complex problems without explicitly programming them [1] and has been developed and extensively applied to academic and industrial problems [2]. However, researchers spend considerable amount of time through trial-and-error methods to build and train these neural networks which often results in lower accuracies. To overcome this challenge, the current research proposes a python-based software platform for automatically building and training neural networks with higher accuracies. The neural network hyperparameters such as number of neurons, activation functions, learning rates, cost functions, and optimizers are automatically optimized using a genetic algorithm embedded structure. The developed neural network was further employed as a parameter optimization objective function for searching suitable operating conditions of the cases understudy. The software package can also be used to plot both 2D and 3D contour surfaces of the predicted variables with ease. The software platform was benchmarked against 5 published researches with only a marginal error of less than 3% and higher accuracies in most of the cases. An average neural network modeling time of 10 minutes was obtained with accuracies greater than 99.8%, indicating the significant savings on time and robustness of the proposed package as well as its potential to be used for online prediction and optimization problems. Owing to the high customizability of the software package, it can be easily modified by advanced users to suit their needs.

References

[1] Lecun Y, Bengio Y, Hinton G. Deep learning. Nature 2015; 521:436–44. https://doi.org/10.1038/nature14539.

[2] Venkatasubramanian, V., 2019. The promise of artificial intelligence in chemical engineering: Is it here, finally? AIChE J. 65, 466–478. https://doi.org/10.1002/aic.16489