Predictive Control Algorithms

Escaping Scheduling Errors via Partial Look-ahead

Since previously mentioned above, the heuristic priority rules that are normally utilized in industrial practice only identify the local status of the jobs in the queue to be sequenced.  No  information  about  the  future  passing  of  the  jobs  via  the network of queues is utilized.

Therefore, priority rules cause scheduling errors. One exacting, frequently happening type of error is known by the practitioners the shading effect. This can be identified vary from the subsequent illustration.

Now lets us suppose that in the queue in front of a highly loaded station there are two categories of jobs, A and B. Operations relating to jobs of type A contain a high priority since they are previously critical. Operations of type B have lower priority. Consequently, a priority  based  local  scheduling  algorithm  would  initial  allocate  the  operations  of category A and after that those of category B, given that nothing else happens. Here, this may occurred that the jobs of type A next have to visit a station that is also highly loaded and upon that they compete along with jobs of category C, all or several of that are even more critical. Conversely, jobs of category B next visit a machine that is ready to process them and might still be idle as extensive as no jobs of category B are finished on the first machine. From the accessible information to the scheduler of initial machine, to  prefer  operations  of  category A  to  assist  them  meet  their  due  dates  is  accurate. Conversely, because of the downstream situation, the result is to reasons additional delays for jobs of category B with no any gains for jobs of category A.

In practice, the condition may be more complex. The bottlenecks in the system might be dynamic, that is the reverse condition on the downstream stations may also arise, consequently an easy bonus for jobs of category B will not help. Or at several instance there might be competition amongst urgent jobs of category C, but at another, not. And the error might not become observable in the next processing steps but only later. As explained this is extremely complicated to explore all possible sequence on all stations even over a extremely short horizon.

The solution which is proposed currently is of predictor - corrector type. The necessary is not to start along with the uncontrolled system, but along with the system controlled via a basic control algorithm. The basic control algorithm is a main concerning rule. After that the potential errors are focused. It implies that the strategy starts on along with bottleneck machines and  simulates  the  solution  on  the  solutions  that  the  jobs  which  are  currently waiting will encounter on the subsequently station. This needs the stimulation of all other stations from which potentially competing jobs may originate. But still only a small part of the system is identified. This type of analysis yields the possible earliest possible beginning times for the subsequent processing steps and, thus, the actual (in practice at least enhanced) due dates for the operations waiting opposite our bottleneck station. Scheduling on this station is after those done based upon these due dates quite than on the global slack of the jobs or on the operation due dates which are determined beforehand. The result of the partial simulation is a dynamic operational due date that is DODD for all operations. To facilitate optimize the sequence of the jobs in the queue; the DODDs can be utilized in many ways. Three strategies were tested as:

(a) The easiest possible scheduling rule on the station for that the simulation  is  complete  is  to  employ  DODD  as  priority  criterion,  that is  to schedule the operation along with the earliest DODD first. It is consistent along with the above reported results which the ODD rule is a good option if the max, rms, and mean tardiness are all relevant.

(b) The  queue  is  first  ordered  as per  to  a  local  priority  rule.  An operation from the original queue is merely moved to the top of the queue if:

• This has an earlier DODD than they represent top candidate;
• No operation that originally was scheduled before this operation misses its DODD.

(c) Again the queue is ordered as per to a local priority rule. This operation is marked, if the machine for the subsequently operation of similar job is a static bottleneck. All marked operations are moved to the top of the queue; after that similar moving procedure as under B is applied to the modified queue.

Strategy B may reasons several idling of downstream machines in order to ignore additional delays of critical jobs. Strategy C sets a higher emphasis on ignoring idling of bottleneck resources since it in a high load situation is severely found to overall performance.