AIOLOS: Artificial Intelligence powered framework for OnLine prOduction Scheduling

英国的化学工业在国家经济中起着至关重要的作用，年营业额达500亿英镑。为了在区域和全球范围内保持竞争力，基于优化的调度方法经常被应用于实现过程利润的显著增加、能源成本的降低、库存管理效率的提高和客户满意度的提高。然而，在化工生产中，需求波动、订单到货急、到期日变更、设备故障等频繁的中断是不可避免的。当这些干扰出现时，预先确定的最优调度可能变得次优甚至不可行的。由于使用基于启发式的反应性调度方法来应对频繁的中断，英国化学工业每年损失的利润估计在数亿英镑左右。现有的基于优化的调度方法要么需要较高的计算费用来生成调度，从而使它们无法管理在线调度中的意外中断；或者直接使用糟糕的启发式或知识来快速决策，这通常会导致保守的时间表，从而导致重大的财务损失。更重要的是，这些方法不能有效地适应在线调度中经常出现的某些中断，如设备故障和急单到达，限制了它们的潜在应用。这项研究将提供下一代自主在线调度框架，以应对化学制造业中不同类型的中断。该框架将生成高质量的调度规则，通过集成新颖的机器学习技术和鲁棒的数学规划方法，及时为新出现的不确定性（例如，< 5分钟）提供最优或接近最优的在线调度解决方案。这也将允许确定一个解决方案，以尽量减少能源消耗。这项研究将通过曼彻斯特大学和伦敦大学学院在过程系统工程和机器学习方面的专业知识的无缝合作来解决。拟议的框架将在与英国和中国的工业合作伙伴的密切互动中进行测试。利润的提高预计至少为3%，可能高达15%，相当于英国化学工业的利润预计每年增加7000万至3.2亿英镑。

The chemical industry in the UK plays a vital role in the nation's economy with a total annual turnover of £50 billion. To remain competitive both regionally and globally, optimisation-based scheduling methods are often applied to achieve a significant increase in process profit, reduction in energy cost, improvement in the efficiency of inventory management, and enhanced customer satisfaction. However, frequent disruptions such as demand fluctuation, rush order arrivals, due date changes, and equipment malfunction are unavoidable in chemical manufacturing. When these disruptions are present, a pre-determined optimal schedule can become suboptimal or even infeasible. With the use of heuristic-based reactive scheduling methods in response to frequent disruptions, the UK chemical industry loses an estimated profit in the order of hundreds of millions of pounds every year. The existing optimisation-based scheduling methods either require high computational expense to generate a schedule, thus rendering them incapable of managing unexpected disruptions in online scheduling; or directly use poor heuristics or knowledge for fast decision-making which usually leads to a conservative schedule resulting in significant financial losses. More importantly, these methods cannot effectively accommodate certain disruptions such as equipment malfunction and rush order arrivals that often occur in online scheduling, restricting their potential application.This research will deliver a next generation autonomous online scheduling framework in response to different types of disruptions in the chemical manufacturing industry. The framework will generate high-quality dispatching rules to provide optimal or near-optimal online scheduling solutions for emerging uncertainties in a timely manner (e.g., < 5 minutes) through integration of novel machine learning techniques and robust mathematical programming approaches. This will also allow for the identification of a solution to minimise energy consumption. The research will be addressed via a seamless collaboration between The University of Manchester and University College London with expertise in process systems engineering and machine learning. The proposed framework will be tested in close interactions with industrial partners in the UK and China. The improvement in profit is expected to be at least 3% and potentially up to 15%, corresponding to an estimated annual increase in profit between £70 million and £320 million for the UK chemical industry.