From planning to scheduling (or from “what to do” to “how to do it”)

From planning to scheduling (or from “what to do” to “how to do it”)

In short, it can be said that Planning tells “what to do”, whereas Scheduling explains “how to do it”.

Modern planning models need to satisfy 2 complementary objectives
• Enabling the user to run a large number of cases in order to quickly evaluate various crude oils available in the market, under combinations of price and demand scenarios, in order to make more robust decisions under uncertainty
• Allowing a short-term perspective in which refinery planning is viewed as a preparation step to refinery scheduling
Princeps solution addresses these 2 objectives successively through a rigorously designed work process.

The first of these objectives is addressed using powerful parametrization and scenario generation capabilities allowing virtually limitless combinations of parametric loops.

Thanks to parallel processing, multiple scenarios can be solved in parallel using multi-core CPU, or distributed to remote servers. All planning runs results are archived in the central output database, and can be shared and analysed by all planners

3
4

The transition to scheduling:

the second objective is about facilitating the transition from planning to scheduling. A natural step in this direction is to augment a planning solution using multiperiod structures, hoping that this would lead to a “close to feasible” schedule of operations.

This approach is motivated by the observation that using a single period plan is equivalent to assuming that all crude oils are available simultaneously, leading to “false synergies” between different crudes, regardless of their arrival date. Likewise, process units are operated in “campaigns”, each having different feed qualities and operating conditions, and units’ rundowns have qualities that are variable across time, also meaning that their downstream routings cannot be constant over the whole planning horizon.

Introducing multiple periods into a planning model alleviates some of the difficulties above, in the sense that different crude slates are processed and different downstream operations are run within each given period, thereby eliminating some of the over-optimisation discussed above. However, this partially dynamic aspect of multiperiod LPs falls short of yielding close to feasible schedules. The reason can be simply understood based on the realisation that, whereas in multiperiod LP models all refinery sectors are assumed to run in synchrony within the boundaries of each time period, in real life operations, they follow asynchronous paths (they live their own life). Even within the same sector, individual operations are not possible to align within time intervals. An obvious example is the crude oil scheduling problem, where crude discharges, transfers and crude runs start and end independently of one another (in the sense: not in synchrony).

The solution:

Princeps technology is the only solution that is really and rigorously capable of bridging the gap between optimal crude selection (planning), and a feasible sequence of upstream operations (crude scheduling). The approach taken consists of 3 steps:

1- a multiperiod LP model provides an optimal planning solution, in which crude purchases are decided, and on which basis a crude cargo arrival calendar is established.
2- Compass uses this calendar as an input and automatically generates a schedule that orchestrates all movements from crude reception tanks to crude distillation unit(s) in order to ensure a feasible sequence of operations, and to achieve optimal crude blending and scheduling. Thanks to a continuous time representation, it addresses all the difficulties highlighted above and takes into account all operational constraints such as feed and rundown capacity limitations, tankage capacity limitations, rundown quality requirements and a large variety of logical and logistical constraints.
3- Capitalising on this initial step, flowers helps the scheduler achieve a feasible and enhanced overall refinery schedule using a smart simulation engine. The work process is supported by a fluid communication between planning and scheduling within a single platform, in order to import and exploit planning directives into scheduling. Throughout the process, the scheduler is assisted by powerful tools and features such as local constraint solving (threshold detection and relations propagation) and fluid communication.

Finally, at the downstream end, Optimix calculates the optimal allocation of components to final products, whether these components are tanks with dynamically varying qualities, or hot rundowns flowing directly from process units.

Post author