Prioritised A* search in real-time elevator dispatching

Under the typical operating conditions of an elevator system, there is insufficient time to consider all dispatching alternatives and the major elevator companies normally adopt empirical techniques to reduce complexity and achieve acceptable performance. The current work has been able to demonstrate that in practical circumstances an optimal solution to the real-time elevator-dispatching problem can be obtained. The elevator dispatching problem is formulated as a heuristic search and is implemented using a novel extension of the popular A* search, termed prioritised A*, that retains the desirable admissibility and monotonicity of A*. PA* makes best use of the limited time available by ensuring the dispatcher considers the most important aspect of the problem first, namely to give each elevator its first assignment. In a manufacturing process, this is equivalent to ensuring that each machine is immediately given its first job, while the determination of the detailed order of the remaining jobs is refined later. This study has obtained access to extensive data records collected from installed elevator systems and their analysis has led to the identification of new passenger models able to deliver suitable high quality predictive data to improve the operations of the dispatcher.