How B.S., Teng S.Y., Leong W.D., Ng W.P.Q., Lim C.H., Ngan S.L., Lam H.L., 2019, Non-linear Programming via P-graph Framework, Chemical Engineering Transactions, 76, 499-504.
P-graph is a graph-theoretic method which is designed to solve process network synthesis (PNS) problem using combinatorial and optimisation algorithms. Due to its visual interface for data encoding and results display; and its capability of generating multiple solutions (optimal and sub-optimal) simultaneously, the utility of P-graph has expanded into a broad range of studies recently. However, this powerful graph-theoretic method still falls short of dealing with non-linear problems. The problem can be found from the cost estimation provided by P-graph software. Despite it allows users to input the sizing cost (noted as “proportional cost” in P-graph software), the capacity and the cost are assumed to be linearly correlated. This inaccurate and unreliable cost estimation has increased the difficulty of making optimal decisions and therefore lead to undesirable profit loss. This paper proposes to solve the fundamental linearity problem by implementing trained artificial neural networks (ANN) into P-graph. To achieve this, an ANN model which utilised thresholded rectified linear unit (ReLU) activation function is developed in a segregated computational tool. The identified neurons are then modelled in P-graph in order to convert the input into the nonlinear output. To demonstrate the effectiveness of the proposed method, an illustrative case study of biomass transportation is used. With the use of the trained neurons, the non-linear estimation of transportation cost which considered fuel consumption cost, vehicle maintenance cost and labour cost are successfully modelled in P-graph. This work is expected to pave ways for P-graph users to expand the utility of P-graph in solving other more complex non-linear problems.
