Survey PaperSolving hybrid charging strategy electric vehicle based dynamic routing problem via evolutionary multi-objective optimization
Introduction
The Vehicle Routing Problem (VRP) has been intensively investigated in supply chain management, city logistics, and other real-world applications [1], [2]. The objective of VRP is to optimize routes of vehicles so as to minimize total routing cost (e.g., cost of energy, travel distance, etc.) and satisfy constraints at the same time. In the last decade, many studies for solving VRP have been proposed. Among these designs, computational intelligence approaches have attracted more attention and interests lately. Yan et al.[3] incorporated a graph-based fuzzy assignment scheme into the evolutionary search to minimize the total travel cost. Zhang et al.[4] designed an evolutionary scatter search particle swarm optimization algorithm to solve the VRP with time windows. Wang et al.[5] developed a two-stage multiobjective evolutionary optimization algorithm to balance convergence and diversity of the population for solving VRP. Mavrovouniotis et al. [6] suggested an improved ant colony optimization (ACO) algorithm to solve combinatorial dynamic optimization problems. The ant immigrant mechanism can replace a portion of the population with some elite solutions from previous environments.
Nowadays, Electric Vehicles (EVs) have been slowly mature and applied intensively due to their environmentally friendly characteristics[7]. Traditional EVs need to charge at the fixed sites like homes or charging stations. This constraint has made EVs inconvenient and has limited their adoption due to insufficient or limited charging facilities. Recently, Wireless Charging Technology (WCT) has been steadily developed[8]. Combined with the WCT, EVs could not only save the transport time without stopping to recharge, but also extend their transport distance. However, the cost of WCT is significantly higher than that of traditional Plug-in Charging Technology (PCT). To be specific, the PCT scheme is money-saving but time-consuming [9], while the WCT scheme can save travel time but is not economical. Hence, it is expected to properly combine these two charging strategies during transportation. Therefore, the VRP of this hybrid charging strategy involves the trade-off between charging cost and travel time and can be viewed as a multiobjective problem.
Two difficulties are brought in the hybrid charging strategy based EVs. First, the availability of PCT and WCT may vary over time due to the allocation or repair of charging devices. In such a changing environment, EVs not only need to find the optimal routes to minimize travel cost but also determine charging strategies with efficiency. Second, existing evolutionary multiobjective optimization algorithms for VRP often find a large number of solutions. However, only few solutions can be implemented for real-world applications, so picking solutions for implementation among a massive number of candidates is a mentally challenging burden to Decision Makers (DMs). Especially, DMs should make decisions as quickly as possible in the dynamic changing environment.
To track the Pareto-optimal Solutions (POS) rapidly under the dynamic environment, a promising way is to reuse the archive memory[10], [11], [12], [13], [14], [15] to share the past search experiences which provide useful knowledge for enhancing the problem-solving performance in the newly changed environment. For example, a prediction technique was employed to generate the POS in the new environment and these predicted solutions can be treated as an initial population for guiding the search direction[16]. However, existing methods often reuse the whole population pool and do not carefully select solutions, which results in many computing resources being wasted on low-quality solutions. In fact, we believe that reusing only a few high-quality solutions not only can improve the performance of the predicted population but also save computing resources. Moreover, it is unnecessary to reuse the whole population, because DMs often pick only few solutions for evaluation and implementation.
From the above discussions, an advisable approach is to reuse few high-quality solutions when the environment has changed so that computational resources and the cognitive burden to DMs are relieved. In the evolutionary multi-objective optimization domain, knee points refer to a subset of non-dominated solutions with higher HyperVolume (HV) values [17] and contain valuable information for guiding the search direction towards the optimal solutions[18], [19]. Besides, knee solutions are often preferred by DMs since they don’t require any subjective preferences from DMs [20], and knee point-based Multi-Criteria Decision Making (MCDM) approaches have sufficient interpretations of the geometry[20]. An illustration of the benefits of the knee point is presented in Fig. 1. Hence, providing knee solutions can bring convenience to DMs, and reusing these high-quality knee solutions as knowledge can be effective in generating a better initial population to enhance evolutionary performance when the environment changes.
In this paper, we propose an innovative framework combining both historical archive and knee-based MCDM strategy to handle the Hybrid Charging Strategy based Dynamic Vehicle Routing Problem (HCS-DVRP). The backbone of the proposed Transfer Knee points based MCDM algorithm (TK-MCDM) consists of three critical components, namely, knee selection, training solutions collection, and reuse function learning. First, the objective space is divided and a knee pointis identified in each division. Second, we collect training samples for training the reuse function by considering the imbalanced problem. Subsequently, a reuse function is learned to predict initial knee points in the changing environment by exploiting the collected training solutions.
The contributions of our work are as follows:
- 1.
The proposed work utilizes knee points not only for accelerating the convergence and maintaining the diversity of the population, but also relieving DMs’ burden of picking solutions. Therefore, DMs can adopt knee solutions easily for implementation, which is of great importance to those real-world dynamic applications, such as the one studied here.
- 2.
We design an adaptive learning mechanism to reuse a small number of knee points instead of a large number of common individuals, which can significantly save computing resources and improve performance.
- 3.
Since the number of knee points is far less than that of other types of solutions, which leads to the imbalanced problem[21]. To address the issue, we generate some artificial knee points to balance the minority for better training the reuse function.
The rest of this paper is organized as follows. Section II presents some existing research works about Dynamic Vehicle Routing Problem (DVRP) briefly. In Section III, we give the mathematical definition for the proposed problem, HCS-DVRP. The proposed algorithm TK-MCDM for HE-DVRP is then detailed in Section IV. Furthermore, Section V introduces a benchmark of HCS-DVRP and discusses the empirical studies conducted to verify our proposed algorithm. Finally, the conclusions are drawn in Section VI.
Section snippets
Related work
In this section, we give a brief review of the recent literatures about Dynamic Vehicle Routing Problem (DVRP). The DVRP and its variants have been widely studied in recent years. Generally, solvers for DVRP aim to meet requests from several destinations or customers and minimize the travel cost and time simultaneously.
DVRP often involves large-scale decision variables, and some works focus on dealing with high-dimensional decision variables efficiently. Kim et al. [22] proposed a Markov
System model and definitions
The formal definition of HCS-DVRP can be expressed as follows: The weighted undirected graph stands for the traffic network, where and represent the set of vertices/nodes and edges, respectively. Let be a binary variable. implies that the optimal route includes the edges , while denotes does not belong to the found optimal route. When an EV travels from node to node , the State of Batteries (SoB) of the EV is presented as
Representation and operator
The chromosome used for encoding HCS-DVRP consists of three segments, as shown in Fig. 2. The first segment represents the order of routing in which each component is the identification of the node, while the second and third segments represent the PCT charging time at node and the WCT charging time via edge , respectively. The second and third segments are encoded with real numbers.
For each segment, different operators are carried out. The order crossover (OC) and swap
The proposed benchmark HCS‐DVRP
The proposed HCS-DVRP benchmark contains 24 test instances modified by commonly used VRP benchmark[49]. These test instances have three categories: “C” (C1 and C2), “R” (R1 and R2) and “RC” (RC1 and RC2). “C” and “R” represent nodes located in clusters or the random distribution, respectively. “RC” means a mixture of “R” and “C”. Each category has two different scales: 25 and 50 nodes. The structure of the HCS-DVRP benchmark instance can be viewed in Fig. 6, where the red square represents the
Conclusion and future work
In this work, a framework combining both historical archive and knee-based MCDM strategy in solving HCS-DVRP is proposed by optimizing routes and determining charging strategies according to the changing environment. Firstly, reusing knee points can better accelerate the convergence. Secondly, it can relieve the burden for implementation. In addition, we design an adaptive reuse mechanism and address the induced imbalanced problem during reusing the knee points by generating artificial knee
CRediT authorship contribution statement
Zhenzhong Wang: Conceptualization, Methodology, Software, Writing – original draft. Kai Ye: Software. Min Jiang: Supervision, Conceptualization, Methodology, Writing – review & editing. Junfeng Yao: Supervision. Neal N. Xiong: Writing – review & editing. Gary G. Yen: Conceptualization, Methodology.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work was supported by the National Natural Science Foundation of China (No. 61673328).
References (54)
- et al.
Ant algorithms with immigrants schemes for the dynamic vehicle routing problem
Inf Sci (Ny)
(2015) - et al.
Discrete honeybee mating optimization algorithm for the routing of battery-operated automated guidance electric vehicles in personal rapid transit systems
Swarm Evol Comput
(2016) - et al.
A survey of swarm intelligence for dynamic optimization: algorithms and applications
Swarm Evol Comput
(2017) - et al.
A modular neural network-based population prediction strategy for evolutionary dynamic multi-objective optimization
Swarm Evol Comput
(2021) - et al.
Ensemble prediction-based dynamic robust multi-objective optimization methods
Swarm Evol Comput
(2019) - et al.
A grey prediction-based evolutionary algorithm for dynamic multiobjective optimization
Swarm Evol Comput
(2020) - et al.
Flexible time-of-use tariff with dynamic demand using artificial bee colony with transferred memory scheme
Swarm Evol Comput
(2019) - et al.
A reinforcement learning approach for dynamic multi-objective optimization
Inf Sci (Ny)
(2021) - et al.
A knee-guided prediction approach for dynamic multi-objective optimization
Inf Sci (Ny)
(2020) - et al.
A multi-objective genetic algorithm based approach for dynamical bus vehicles scheduling under traffic congestion
Swarm Evol Comput
(2020)
A multi-objective memetic algorithm based on locality-sensitive hashing for one-to-many-to-one dynamic pickup-and-delivery problem
Inf Sci (Ny)
The dynamic vehicle routing problem: solution with hybrid metaheuristic approach
Swarm Evol Comput
A self-adaptive evolutionary algorithm for dynamic vehicle routing problems with traffic congestion
Swarm Evol Comput
A metaheuristic approach to solve dynamic vehicle routing problem in continuous search space
Swarm Evol Comput
A linear MIP model for the electric vehicle routing problem with time windows considering linear charging
2018 Annual Reliability and Maintainability Symposium (RAMS)
Evolutionary many-objective algorithm based on fractional dominance relation and improved objective space decomposition strategy
Swarm Evol Comput
Two memetic algorithms for heterogeneous fleet vehicle routing problems
Eng Appl Artif Intell
Drone routing in a time-dependent network: towards low cost and large range parcel delivery
IEEE Trans. Ind. Inf.
Placement and routing optimization for automated inspection with uavs: a study in offshore wind farm
IEEE Trans. Ind. Inf.
A graph-based fuzzy evolutionary algorithm for solving two-echelon vehicle routing problems
IEEE Trans. Evol. Comput.
An evolutionary scatter search particle swarm optimization algorithm for the vehicle routing problem with time windows
IEEE Access
A two-stage multiobjective evolutionary algorithm for multiobjective multidepot vehicle routing problem with time windows
IEEE Trans Cybern
A new coil structure and its optimization design with constant output voltage and constant output current for electric vehicle dynamic wireless charging
IEEE Trans. Ind. Inf.
An electric vehicle routing optimization model with hybrid plug-in and wireless charging systems
IEEE Access
A fast dynamic evolutionary multiobjective algorithm via manifold transfer learning
IEEE Trans Cybern
Transfer learning-based dynamic multiobjective optimization algorithms
IEEE Trans. Evol. Comput.
A knee point-driven evolutionary algorithm for many-objective optimization
IEEE Trans. Evol. Comput.
Cited by (22)
MOEA/D with customized replacement neighborhood and dynamic resource allocation for solving 3L-SDHVRP
2024, Swarm and Evolutionary ComputationA novel combinational response mechanism for dynamic multi-objective optimization
2023, Expert Systems with ApplicationsMulti-regularization sparse reconstruction based on multifactorial multiobjective optimization
2023, Applied Soft ComputingReal-time collaborative feeder vehicle routing problem with flexible time windows
2022, Swarm and Evolutionary ComputationCitation Excerpt :Zhu et al. [56] studied a multi-objective optimization problem with three conflicting objectives (i.e., route length, response time, and workload) for the dynamic pickup-and-delivery problem (DPDP). Furthermore, Wang et al. [43] offered a hybrid charging strategy to solve the DVRP. The proposed model aimed to minimize the total time of travel and the total cost of charging.
A fast sampling based evolutionary algorithm for million-dimensional multiobjective optimization
2022, Swarm and Evolutionary Computation