Reinforcement-Learning-Based Decision Support for Industrial Enterprise Employees’ Digital Twin Management System

Abstract

The article explores the problem of the design of an employee’s digital twin for a human resource management system under Industry 4.0. An employee’s digital twin is linked to the concept of individual human capital, that is, the combination of professional, intellectual, and social resources that determine employee productivity. The digital twin model includes a model for an employee’s human capital assessment and a decision support model for the employee’s individual professional trajectory design. The decision support model is based on the concept of a Markov decision process (MDP), dynamic programming methods, and reinforcement learning (RL) algorithms. RL algorithms generate an optimal control mode and represent a set of management decisions for employee development appropriate to their health and intellectual, social, and career potential. Several reinforcement learning algorithms of different classes are tested: DQN, SARSA PRO, and DDQN. The developed algorithm—Dual Deep Q-Networks (DDQN)—demonstrates the highest performance compared to other learning algorithms. This algorithm is adapted to the relevant problem and used in a decision support model. The difference between the proposed model and others is the policy of individual human capital management, aimed at increasing human capital and growing employee productivity. The results of employee digital twin implementation have practical significance: it enables the rapid mitigation of human capital risks, improves employee productivity, and enhances enterprise efficiency.