Comparison of Agent-Based Scheduling to Look-Ahead Heuristics for Real-Time Transportation Problems

Planning and controlling various transportation networks often yields uncertainties such as transportation times and the arrival of urgent orders among many others. This is why it is considered essential to employ a flexible, stable and robust system for planning and controlling the transportation networks. Flexibility, in this case, implies that the schedule adjustments in response to any information updates are easy to perform. Stability, on the other hand, means that the minor information updates should only impact a small part of the schedule. Finally, being robust implies that the overall network performance should maintain its acceptability under various scenarios including those that are unexpected, such as the urgent orders (Cheng, Tajan & Lau 2012, p. 309). This paper features a summary of the arguments, ideas, and calculations presented in the article.

Suggested Scheduling Approach

The authors note that traditional approaches such as the operations research (OR) based global optimization methods are used to develop the integral transport schedules (Mes, van der Heiken & van Haretn 2005, p. 2). However, they do not agree that it is the most suitable option for the planning and control of the present transport networks which are stochastic and dynamic. They cite various reasons for this argument; one, they note that most optimization algorithms require significant information beforehand. Two, they argue that these global optimization algorithms may be at risk of destruction in the event that information updates are performed. Therefore, even a minor information changes may lead to a great impact on the vehicle schedules. Three, the algorithm requires a lot of time which may eventually not lead to the timely response to any unexpected occurrences such as the demand for rush orders. Lastly, the flexible transportation networks may feature quite a number of independent organizational units. These units tend to work autonomously and are rarely cooperative. This means that the sharing of information will be a difficult process, hence making the traditional centralized approach inappropriate for this specific situation.

The authors suggest an alternative approach which features the Multi-Agent System (MAS) ((Mes, van der Heiken & van Haretn 2005, p. 2). This system is different in the sense that it consists of control units which are independent, yet linked together through physical or functional entities such as the vehicle or order. They argue that it presents a better solution to the problem of controlling such complex networks as it encourages flexibility, reliability, and even adaptability. Unlike the traditional methods, this system features units which act autonomously but still interacts with each other.

The use of Related Literature

To establish what is already known on the research topic, these authors conducted a literature review on specific scholars. First, they start by considering what different authors have concluded about transport planning. The problem of assigning jobs to vehicles is considered to existing as a result of vehicle routing problems (Mes, van der Heiken & van Haretn 2005, p. 4). Most of the variants of this problem are quite difficult to solve effectively within a short time. Most of the researchers focus their work on the static and deterministic problems; here, all information is known.

When the data is stochastic and dependent on time, the process of planning will feature a policy of how the routes should evolve rather than planning the route itself (Mes, van der Heiken & van Haretn 2005, p. 4). These policies have already been studied by various authors who have been acknowledged in the article. In this case, dealing with the transport problem involves using the available data at that point in time, to solve the model, which will then be re-optimized when new data is presented in future. Since the fast response is often required, a simple heuristics method is employed to provide an overview.

In relation to demand situations that are oversaturated, these authors refer to the work of Kim, Mahmasanni, and Jaillet (2002, p. 69). These scholars present a hybrid approach to the problem of reassigning and sequencing the accepted loads, where they suggest the combination of fast heuristics for initial assignment with the optimization-based approach.

Next, the review of literature focuses on agent-based logistic planning. Here, an agent is described as hardware or software based computer system which has specific properties of autonomy, reactivity, social ability, and even pro-activeness (Mes, van der Heiken & van Haretn 2005, p. 4). Therefore, the MAS is a series of agents which interact in order to come up with a solution to a complex problem. An example of how such an interaction may be achieved is through the use of market mechanism where the resource agents are expected to bid for orders through the dynamic pricing of orders. This is the approach which is employed in the rest of the paper.

In conducting a review of the literature on this subtopic, the authors note that only two scholars engaged in a comparison of their agent approach with an exact solution produced by CPLEX (Mes, van der Heiken & van Haretn 2005, p. 5). In their conclusion, they argued that the centralized models are no longer effective as effective since there are decentralized models which help to deal with the problems of computational inefficiency and degradation in the quality of the resulting solution.

This section identifies specific transport areas where the MAS has been employed effectively, such as in the railroad schedule, real-time coupling and sharing of train wagons, and even in air cargo assignment.

Contribution to the Literature

Considering the areas that have already been answered, these authors note a significant gap in the literature which they seek to help in bridging. The contributions are focused on specific issues for agent-based transport scheduling including; combining the soft time windows with the incomplete information to improve order handling times, a study of how the overall system performance is impacted due to additional intelligence agents, a comparison of the MAS to the traditional approaches for real time transport planning, an analysis of the robustness of performance, and lastly the analysis of the impact of order characteristics on the resulting costs (Mes, van der Heiken & van Haretn 2005, p. 6).

Cost Structure and Performance Measurements

The authors of the article provided a cost structure for the new MAS system, as well as the performance measurements that will be used. The key performance indicators that will be used include; the total number of orders delivered before their due time, the stability of the service level, the percentage of driving loaded, and the relative additional costs (Mes, van der Heiken & van Haretn 2005, p. 8).

In relation to the cost structure, it is assumed that fixed costs are identical; hence they are not relevant for the decisions of scheduling (Taghaddos et al. 2012, p. 37). However, there are relevant cost structures which include; variable cost per time unit and the penalty costs.

Schedules

The authors define a vehicle schedule as a sequence of actions involving the following; first, it refers to a vehicles move along a specific arc when loaded (Mes, van der Heiken & van Haretn 2005, p. 8). Second, a vehicle moves along a specific arc when empty. Third, when a vehicle waits at a specific node for a given period. It is argued that a schedule will always end with this last option as it will be expected to return to the loading joint empty, after making a delivery.

The schedule of vehicles is updated at specific times, following specific events. It was updated when the first action is completed and when it needed to match external load with an available vehicle capacity (Mes, van der Heiken & van Haretn 2005, p. 9). Rarely, periodic planning was needed due to the control.

Agent-Based Planning Concepts

In the research conducted, the vehicles were assigned to jobs using the market-like negotiation protocol which coordinated the agents decisions. This concept of planning may be defined using three key choices (Mes, van der Heiken & van Haretn 2005, p. 9); one, it depended on the agents used to distinguish between the tasks and goals. Two, the products to trade also defined the type of planning concept. Three, the concept was determined by the market mechanism to be defined.

The multi-agent structure used in this research featured four agent types. It is described in the figure below;

Figure SEQ Figure \* ARABIC 1: Agent Structure for Transportation Network(Mes, van der Heiken & van Haretn 2005, p. 9)

The structure features one agent per vehicle and one agent per order. In addition, the fleet manager agent is included to aid in the collection and analysis of and processing time data for the vehicles (Wang 2011, p. 19). It also has the capacity to distribute this information to the vehicles when required to do so. The goal of the vehicle agents is to maximize profits by deploying capacity, while the goal of the job agent is to arrange transportation of the remaining loads before the deadline at the lowest cost possible. However, each one of these vehicles maintains its own schedule.

According to the authors, introducing hierarchy into this system has the capacity to improve how these agents coordinate (Mes, van der Heiken & van Haretn 2005, p. 10). Hierarchy, therefore, may be defined at the job level and the resource level. The job level is characterized by a shipper agent that is responsible for a set of orders. An example of a task at this point is to reallocate transport capacity to ensure orders are completed before their due times and at the lowest possible times.

In relation to the marketplace, product definition must be conducted. There are different options which must be distinguished (Mes, van der Heiken & van Haretn 2005, p. 10-11), including; (i) order transportation from one location to another, which is to be loaded at the release time and delivered before the due time.(ii) transport capacity of a unit load that is available at a given node, during a specific time, that must be used within the specified timeframe. (iii) Transport capacity of a specific number of vehicle loads that can be used within a given time frame. (iv) Lastly, it may feature the transport capacity of a unit load from one point to the other, which must be picked up at a given time and delivered before the deadline. For this research, the first option is settled for as it is the most simplistic approach for bidding. It also ensures flexibility of schedules, especially when they are required to make significant changes to manage rush orders.

For the distributed scheduling, there are several auctioning mechanisms that have been proposed. For example; there is bargaining, sealed-bid auctions, and the open outcry auctions. For the purpose of this research, the authors settled for the Vickery auction which is under the sealed-bid auctions (Mes, van der Heiken & van Haretn 2005, p. 11). The Vickery auction is where the bidder is given the price of one but the best offer. It was chosen because of the ease of use as it requires a single bidding round. In addition, when under mild conditions the optimal bid is the net cost price of the bidder. This bidder will make a profit from the margin between the two bids, hence this mechanism also provides acceptable profits.

Considering the above descriptions, it is evident that market mechanism is implemented as follows; whenever an order arrives, the required job agent starts the auction by requesting for bids from the v...