Control of perishable inventory system with uncertain perishability process using neural networks and robust multicriteria optimization

The inventory systems are highly variable and uncertain due to market demand instability, increased environmental impact, and perishability processes. The reduction of waste and minimization of holding and shortage costs are the main topics studied within the inventory management area. The main difficulty is the variability of perishability and other processes that occurred in inventory systems and the solution for a trade-off between sufficient inventory level and waste of products. In this paper, the approach for resolving this trade-off is proposed. The presented approach assumes the application of a state-feedback neural network controller to generate the optimal quantity of orders considering an uncertain deterioration process and the FIFO issuing policy. The development of the control system is based on state-space close loop control along with neural networks. For modelling the perishability process Weibull distribution and FIFO policy are applied. For the optimization of the designed control system, the evolutionary NSGA-II algorithm is used. The robustness of the proposed approach is provided using the minimax decision rule. The worst-case scenario of an uncertain perishability process is considered. For assessing the proposed approach, simulation research is conducted for different variants of controller structure and model parameters. We perform extensive numerical simulations in which the assessment process of obtained solutions is conducted using hyper volume indicator and average absolute deviation between results obtained for the learning and testing set. The results indicate that the proposed approach can significantly improve the performance of the perishable inventory system and provides robustness for the uncertain changes in the perishability process.