Hybrid Predictive Control for a Dial-a-Ride System

Abstract

AbstractIn this chapter, we develop a family of solution algorithms based upon computational intelligence for solving the dynamic multi-vehicle pickup and delivery problem. The problem is formulated under a hybrid predictive control (HPC) scheme, which considers the prediction of the future demand and traffic conditions of the transport system.A generic expression of the system objective function is used to measure the benefits of dispatch decisions of the proposed scheme when solving for more than a two-step-ahead problem under unknown future demand conditions. The demand prediction is based on a systematic fuzzy clustering methodology resulting in appropriate probabilities for uncertain future service requests.The potential uncertainty in travel time resulting from unexpected incidents in the transport network is incorporated into the vehicle routing decisions, incorporating the vehicle position and its speed as indicators of traffic conditions. The unpredictable congestion events generate a more complex dynamic routing problem that is handled through both fault-detection and isolation and fuzzy, fault-tolerant control approaches. Because the dynamic problem considered is NP-hard, we propose the use of evolutionary algorithms that provide near-optimal solutions for one-, two-, and three-step-ahead problems and generate promising results in terms of computation time and accuracy.One extension of the HPC framework is a more generic formulation considering a multi-objective function specification for the dial-a-ride problem under the premise that the dynamic objective should consider two dimensions: user and operator costs. Because these two components are usually directed at opposite goals, the problem is formulated and solved through multi-objective model predictive control.KeywordsPareto FrontFuzzy ClusterTraffic ConditionUser CostPrediction HorizonThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.