Stochastic optimization in perishable food supply chain: a holistic approach

Abstract

With continuous expansion in world population, total food demand across the globe is anticipated to increase by 56% within a span of 30 years. Predictions state that food production needs to increase by 70% and adhere to quality standards. Better informed consumers now want to have precise knowledge about the origin of food till it reaches the final shelves of the supermarket. Perishable food supply chain is a complex network involving multiple stakeholders and several interconnected stages. Presence of uncertainties like demand, outbreak, or contamination and a limited product shelf life adds further complexity and the need to uphold food quality and safety standards throughout the supply chain, from crop production to consumers. Without traceability in food supply chains, severe problems of product recall, consumer dissatisfaction, and contamination insecurities have occurred in the past. The lack of a transparent system resulted in food loss and aggravated the global challenge of feeding a growing population. In contrast to centralized systems which lack trustworthy information and deterministic optimization models, our goal is to incorporate a dynamic system like blockchain to monitor the quality of food, keeping the entire network transparent among the stakeholders, embedded with stochastic models to make it a robustly optimized supply chain. This research develops stochastic optimization models to comprehensively combat uncertainty in item quality, transportation of perishable items (not limited to the ones considered below) and provide network transparency. We consider sausage, wheat, and cheese as perishable products for our research. First, we examine a five-level sausage supply chain where at each level, the output product is manufactured by combining/mixing correct proportions of the raw materials from the previous stage. The demand for the final product is uncertain. We develop a two-stage stochastic model and analyze it using the L-shaped algorithm to improve traceability, optimize dispersion, and fulfil demand among the batches. Next, to maintain safety standards, detailed wheat and cheese supply chains are analyzed separately to filter relevant parameters responsible for food quality at any point in the network. We implement Q-learning algorithm to optimize values of parameters based on which the decision maker can choose the best or worst decision in determining the quality of the perishable product. Later, we analyze a large-scale rich tanker trailer routing problem with stochastic transit times for perishable bulk orders. Unlike classical transportation problems, bulk transportation falls under the umbrella of rich vehicle routing problems that involve several intermingled decisions. Typically, the bulk orders are characterized by a set of attributes consisting of an origin-destination location pair, pickup and delivery time windows, order specification, restrictions based on prior orders of food leading to incompatibility, and possibly special equipment (washed and prepped) or handling instructions. We propose concepts of a novel graph decomposition algorithm, column generation and shortest path to generate feasible optimal routes and account for incompatibility constraints. We overcome the challenge of network transparency by storing supply chain data in a decentralized ledger, blockchain where data is visible to all stakeholders but immune to tampering, thus enabling transparency.