A collaborative and dynamic multi-source single-destination navigation algorithm for smart cities. (March 2023)

• Record Type: Journal Article
• Authors: Xiao, Ziren; Liu, Chang; Luo, Shan; Huang, Kaizhu; Gao, Honghao; Xu, Xiaolong; Wang, Xinheng
• Is Part Of: Sustainable energy technologies and assessments. Volume 56(2023)
• Publication Date: 2023-03
• Subjects: Ant colony optimisation -- Collaborative -- Multiple sources path planning; Renewable energy sources -- Periodicals; Energy development -- Technological innovations -- Periodicals; Electric power production -- Periodicals; Energy storage -- Periodicals; 333.79
• DOI: 10.1016/j.seta.2023.103032
• Languages: English
• ISSNs: 2213-1388

Abstract

Abstract: In order to enable multiple agents to select the best gathering point dynamically, the design of a collaborative and efficient method in real-time is crucial. The dynamic path planning problem from multiple sources to a single destination (DMS-SD) without prior knowledge about the target is proposed in this paper. The modified Dijkstra's algorithm (Xiao et al., 2022) in the previous work can optimally address the DMS-SD problem, which effectively generates an optimal solution in the small map. However, it requires more than 60s to compute the result in our extensive map test, which is intolerable for real-time navigation users. Therefore, we have proposed a hybrid optimisation method to address the problem more efficiently in this paper. The proposed method integrates the Ant Colony Optimisation (ACO) with Monte Carlo Tree Search (MCTS) and modifies the heuristic function to fit the hybrid algorithm. The pure MCTS algorithm can accelerate the randomised search by only exploring unvisited nodes, instead of generating every possible solution. More importantly, benefiting from limiting the maximum exploring depth, our method can approach an optimal point and generate a sub-optimal solution without any prior training used in other neural network-based methods. Experiment results show that our proposed algorithm demonstrates competitive performance with other existing state-of-the-art methods, such as reinforcement learning-based approaches, without training the neural network model. Our method also provides up to 98% reduction in computation time while obtaining sub-optimal results, comparing with the modified Dijkstra's algorithm.