On-line Energy Allocation Based on Approximate Dynamic Programming for Iron and Steel Industry

Abstract

Energy allocation in iron and steel industry is the assignment of available energy to various production users. With the increasing price of energy, a perfect allocation plan should ensure that nothing gets wasted and no shortage. This is challenging because the energy demand is dynamic due to the changes of orders, production environment, technological level, etc. This paper try to realize on-line energy resources allocation under the situation of dynamic production plan and environment based on typical energy consumption process of steel enterprises. Without definite analytical model, it is a tough task to make the energy allocation plan tracks the dynamic change of production environment in real time. This paper proposes to deal with dynamic energy allocation problem by interactive learning with time-varying environment using Approximate Dynamic Programming method. The problem is formulated as a dynamic model with variable right-hand items, which is an updated energy demand obtained by on-line learning. Reinforcement learning method is designed to learn the energy consumption principle from the historical data to predict energy consumption level corresponding to current production environment and the production plan in future horizon. Using the prediction results, on-line energy allocation plan is made and its performance is demonstrated by comparison with static allocation method.