The proto_lab has been researching the intelligent planning of production resources with the help of AI for almost two years. The aim is to automatically assign manufacturing orders to machines with available capacities, depending on the process step. Now, the research results on the Job Shop Scheduling Problem (JSSP) could be published as a scientific paper through MDPI (publisher of scientific open-access journals). JSSP falls into the category of NP-hard combinatorial optimization problems (COP), where finding a solution through exhaustive search is no longer feasible. Simple heuristics such as first-in, first-out, largest processing time first and metaheuristics such as tabu search are often used. However, these methods are inefficient for very complex problems, as they are either far from the optimum or consume a lot of time. In recent years, deep reinforcement learning (DRL) has become increasingly important for solving COPs and there are promising results in the field of AI research in terms of solution quality and computational efficiency.
In the proto_lab, a novel approach to solve the JSSP has been developed, exploring the generalization of objectives and the effectiveness of the solution using DRL. The Proximal Policy Optimization (PPO) algorithm is applied, which adopts the policy gradient paradigm that has been proven to be effective in constrained job dispatch. To achieve better generalized learning of the problem, a new method called Order Swapping Mechanism (OSM) has been integrated into the environment. The publication extensively analyzes the performance of the proposed approach by using a set of available benchmark instances and comparing the results with the work of other groups.