Optimization of crude oil blending with neural networks and bias update scheme

Crude oil blending is an important unit operation in petroleum refining industry. There are two difficulties for commercial crude oil optimizing controllers: (1) the optimization is stable only in some special conditions, the simplex-base algorithm is not robust; (2) the optimal control should be realized on-line and it cannot be analyzed off-line based on the history data. In this paper, we propose a neural networks approach to overcome these two drawbacks. We first we use a recurrent neural networks to solve the linear programming with bias update. Then we use a static neural networks to modeling crude oil blending process based on the history data. Input-to-state stability approach is applied to access robust learning algorithms of the neural networks. Numerical simulations are provided to illustrate the successful application of neural networks on optimization.