Co-evolutionary Decision-Making Modeling Via Integration of Machine Learning and Optimization

Abstract

A dynamic model configuration platform for executing virtual production that dynamically responds to various manufacturing scenarios is presented. By configuring a virtual supply chain using the proposed platform, customer satisfaction and higher productivity can be achieved by verification of manufacturing scenario, product configuration of products, supply chain configuration. Preliminary examination can be easily conducted from various viewpoints of environmental load, risk response, and safety, etc. Using the proposed platform, a co-evolutionary decision-making modeling framework is introduced for estimating appropriate objective functions and constraints of decision-makers from event logs for automatically generating optimization models from big data. Given the input/output data and the optimization model, the problem of finding unknown coefficients of the optimization model that match the given data is called the inverse optimization problem. Machine learning models such as statistical models and neural networks can be used to construct mathematical models from big data. However, it is not easy to explain the causal relationship of the obtained results with these models alone. A valid optimization model requires the existence of a feasible solution and the guarantee of physical and causal consistency with the actual system of transition relations. Therefore, we construct a feedback loop for machine learning and optimization, complement each other’s performance by using the optimization results for machine learning, and realize real-time optimization and online additional learning. A main feature of the proposed method is that by verifying specifications using discrete event system theory, model checking of mathematical models constructed from event logs is performed. A necessary data is selected from a large amount of data, and an optimized model that meets the specifications is constructed.