Both rising and more volatile energy prices are strong incentives for manufacturing companies to become more energy-efficient and flexible. A promising approach is the intelligent control of Industrial Energy Supply Systems (IESS), which provide various energy services to industrial production facilities and machines. Due to the high complexity of such systems widespread conventional control approaches often lead to suboptimal operating behavior and limited flexibility. Rising digitization in industrial production sites offers the opportunity to implement new advanced control algorithms e. g. based on Mixed Integer Linear Programming (MILP) or Deep Reinforcement Learning (DRL) to optimize the operational strategies of IESS.This paper presents a comparative study of different controllers for optimized operation strategies. For this purpose, a framework is used that allows for a standardized comparison of rule-, model- and data-based controllers by connecting them to dynamic simulation models of IESS of varying complexity. The results indicate that controllers based on DRL and MILP have a huge potential to reduce energy-related cost of up to 50% for less complex and around 6% for more complex systems. In some cases however, both algorithms still show unfavorable operating behavior in terms of non-direct costs such as temperature and switching restrictions, depending on the complexity and general conditions of the systems.

中文翻译：

---

工业能源供应系统优化控制算法的比较研究

不断上涨的能源价格和更不稳定的能源价格都是制造企业提高能效和灵活性的强烈动力。一种有前途的方法是工业能源供应系统（IESS）的智能控制，该系统为工业生产设施和机器提供各种能源服务。由于这种系统的高度复杂性，广泛的常规控制方法经常导致次优的操作行为和有限的灵活性。工业生产现场数字化的兴起提供了实施新的高级控制算法的机会，例如基于混合整数线性规划（MILP）或深度强化学习（DRL）的优化IESS的运行策略。优化的运营策略。以此目的，通过使用一个框架，该框架通过将基于规则，基于模型和基于数据的控制器连接到具有不同复杂度的IESS动态仿真模型来进行标准化比较。结果表明，基于DRL和MILP的控制器具有巨大的潜力，对于不太复杂的系统，与能源相关的成本最多可降低50％，对于更复杂的系统，可降低6％左右。但是，在某些情况下，根据系统的复杂性和一般条件，两种算法仍会在非直接成本（例如温度和切换限制）方面显示不利的操作行为。结果表明，基于DRL和MILP的控制器具有巨大的潜力，对于不太复杂的系统，与能源相关的成本最多可降低50％，对于更复杂的系统，可降低6％左右。但是，在某些情况下，根据系统的复杂性和一般条件，两种算法仍会在非直接成本（例如温度和切换限制）方面显示不利的操作行为。结果表明，基于DRL和MILP的控制器具有巨大的潜力，对于不太复杂的系统，与能源相关的成本最多可降低50％，对于更复杂的系统，可降低6％左右。但是，在某些情况下，根据系统的复杂性和一般条件，两种算法仍会在非直接成本（例如温度和切换限制）方面显示不利的操作行为。