Title:

Shortest path algorithms for dynamic transportation networks

Over the last decade, many interesting route planning problems can be solved by finding the shortest path in a weighted graph that represents a transportation network. Such networks are private transport networks or timetabled public transportation networks. In the shortest path problem, every network type requires different algorithms to compute one or more than one shortest path. However, routing in a public transportation network is completely different and is much more complex than routing in a private transport network, and therefore different algorithms are required. For large networks, the standard shortest path algorithms  Dijkstra's algorithm (1959) and Bellman's algorithm (1958) are too slow. Consequently, faster algorithms have been designed to speed up the search. However, these algorithms often consider only the simplest scenario of finding an optimal route on a graph with static real edge costs. But real map routing problems are often not that simple  it is often necessary to consider timedependent edge costs. For example, in public transportation routing, consideration of the timedependent model of these networks is mandatory. However, there are a number of transportation applications that use informed search algorithms (where the algorithm uses heuristics that guide the search toward the destination), rather than one of the standard static shortest path algorithms. This is primarily due to shortest paths needing to be rapidly identified either because an immediate response is required. For example, the A* algorithm (Nilsson, 1971) is widely used in artificial intelligence. Heuristic information (in the form of estimated distance to the destination) is used to focus the search towards the destination node. This results in finding the shortest path faster than the standard static search algorithms. Road traffic congestion has become an increasingly significant problem in a modern society. In a dynamic traffic environment, traffic conditions are timedependent. For instance, when travelling from home to the work, although an optimal route can be planned prior to departure based on the traffic conditions at that time, it may be necessary to adjust the route while en route because traffic conditions change all the time. In some cases, it is necessary to modify the travelling route from time to time and replan a new route from the current location to the destination, based on the realtime traffic information. The challenge lies in the fact that any modification to the optimal route to adapt to the dynamic environment necessitates speeding up of the search efforts. Among the algorithms suggested for the dynamic shortest path problem is the algorithm of Lifelong Planning A* algorithm (LPA*) (Koenig, Likhachev and Furcy, 2004). This algorithm has been given this name because of its ability to reuse information from previous searches. It is used to adjust a shortest path to adapt to the dynamic transportation network. Search space and fast shortest path queries can be used for finding fastest updated route on road and bus networks. Consequently, the efficient processing of both types of queries is of firstrate significance. However, most search methods focus only on one type of query and do not efficiently support the other. To address this challenge, this research presents the first novel approach; an Optimised Lifelong Planning A* (OLPA*) algorithm. The OLPA* used an appropriate data structure to improve the efficiency of the dynamic algorithms implementation making it capable of improving the search performance of the algorithm to solve the dynamic shortest path problem, which is where the traveller may have to recompute the shortest path while travelling in a dynamic transportation environment. This research has also proposed bidirectional LPA* (BLPA*) algorithm. The proposed algorithm BLPA* used bidirectional search strategy and the main idea in this strategy is to divide the search problem into two separate problems. One search proceeds forwards from the start node, while the other search proceeds backwards from the end node. The solution requires the two search problems to meet at one middle node. The BLPA* algorithm has the same overall structure as the LPA* algorithm search, with some differences that the BLPA* contains a priority queue for each direction. This research presented another algorithm that designed to adaptively derive the shortest path to the desired destination by making use of previous search results and reducing the total execution time by using the benefits of a bidirectional search strategy . This novel algorithm has been called the bidirectional optimised Lifelong A* algorithm (BiOLPA*). It was originally proposed for road transport networks and later also applied to public transportation networks. For the road transport network, the experimental results demonstrate that the proposed incremental search approach considerably outperforms the original approach method, which recomputed the shortest path from scratch each time without utilization of the previous search results. However, for public transportation, the significant problem is that it is not possible to apply a bidirectional search backwards using estimated arrival time. This has been further investigated and a better understanding of why this technique fails has been documented. While the OLPA* algorithms give an impressive result when applied on bus network compared with original A* algorithms, and our experimental results demonstrate that the BiOLPA* algorithm on road network is significantly faster than the LPA*, OLPA* and the A* algorithms, not only in terms of number of expansion nodes but also in terms of computation time.
