Hybrid Neural Network and Particle Swarm Optimization Model Applied To Assist Production Planning in A Large Petrochemical Plant

Process and production planning play an important role in petrochemical production systems. planning models are essential to optimize the combination of multiple non-linear production processes involved and therefore improve the commercial competitiveness of the such plant. artificial neural networks offer an effective petrochemical plant planning tool, especially when configured in hybrid form as a back propagation artificial neural network coupled with an optimizer to assist with feature selection. a plant with eight feedstock inputs and thirteen petrochemical products is evaluated, firstly to show the capabilities of a basic backpropagation network model in predicting product outputs. the involvement of a particle swarm optimizer assists in filtering the dataset to remove outlying data records and identifying the input variables that are influential in determining specific product output volumes. the hybrid back propagation network-particle swarm optimization model assists by determining the logical relationship between input and output variables and expressing them in the form of an index matrix. the matrix leads to improved predictions of production outputs and faster convergence of the planning model. the modified back propagation network achieved maximum, minimum, and average relative errors of 59.1%, 0.0%, and 9.9%, respectively. prediction errors in that range are considered acceptable for the collective production processes of a large-scale petrochemical complex evaluated with a nonlinear planning program.