Dynamic optimization in business-wide process control

The chemical marketplace is a global one with strong competition between man- ufacturers. To continuously meet the customer demands regarding product quality and delivery conditions without the need to maintain very large stor- age levels chemical manufactures need to strive for production on demand. In this thesis we research how market-oriented production can be realized for the particular class of multi-grade continuous processes. For this class of processes production on demand is particularly challenging due to the the complex trade- off between performing costly and time-consuming changeovers and maintaining high storage levels. The first requirement for market-oriented production is that production management cooperates with purchasing and sales management. We propose the use of a scheduler as a decision support system in a cooperative organization constituted by these players. In such a scheduler, decision making is represented using decision variables and their effect on the company-wide objective, which is chosen to be the added value of the company, is modeled. The scheduler then selects a decision strategy that is optimal with respect to the objective and presents this strategy to the decision makers who use it to base their actual decision taking on. The company-market interaction is modeled using a transaction-based mod- eling framework. Therein not the actual market behavior is modeled but the expected effect of the interaction of the company with the market. Two types of transactions can be modeled in this framework: orders, which result from contracts with suppliers and customers, and opportunities, which express the expected sales and purchases. Two different approaches to the modeling of production decisions are taken, the choice of which depends largely on the im- plementation of the process control hierarchy that is assumed. In the first approach, production management and control is performed by a single level controller and the control decisions are the minute to minute manipulation of the valves. This approach is academically interesting, though practically in- tractable due to the combination of long horizons and fast sampling times. In the second approach the process control hierarchy consists of a scheduling layer at which it is determined what products will be produced when, and a process control layer which determines how this production is realized. This approach is taken in the rest of the thesis.