A decentralised control model for the risk-based management of dangerous good transport flows on road Claudio Roncoli, Chiara Bersani, Roberto Sacile Department of Communication, Computer and System Sciences (DIST) University of Genova Genova, Italy Claudio.Roncoli@Unige.it, Chiara.Bersani@Unige.it, Roberto.Sacile@Unige.it Abstract—A decentralised approach to control the risk in a transportation network where dangerous goods may be transported is presented. The approach is based on a graph representation of the territory where each vertex represents a sub-region of a defined extension, and links represents the dangerous goods flow from a sub-region to another. The mathematical model of the system is presented according to a quadratic objective function, a linear state equation, and other linear constraints. The model is then decomposed in a series of sub-problems according to a classical dual decomposition approach related to convex optimization. A simple example is reported as case study. Keywords-Dangerous goods transport; Optimisation problem; Decentralised control I. INTRODUCTION Dangerous Goods (DG) transport represents about the 4.1% of total tons-kilometre performance of road goods transport in 2007 in the EU-27 and, in any mode of transport, accidents involving dangerous goods can produce hazards such as fire, explosion, chemical leak and environmental pollution, as well as possibly affecting areas beyond the actual scene of the accident [1]. The classical approaches to define the risk for DG transport introduce the probabilities of accident - although it is very low - and reported accident data are very scarce [2]. Besides, in [3], an integrated methodology to estimate DG transportation accident frequency by utilising publicly available databases and expert knowledge has been implemented. The estimation process addresses route-dependent and route-independent variables. A multi-objective optimisation model in DG transport have been studied in [4]: here the Decision Maker (DM) has to plan, for each day, the deliveries of DG from depots to several other destinations following a risk-averse routing approach. This model also takes into account the risk for domino effect arising from the simultaneous presence of one or more vehicles on the same link at the same time, with the purpose of improving the minimisation of the overall maximum exposure. A different approach to DG transportation risk analysis was recently proposed by [5], where the authors identified the most influential variables and countermeasures for two consequences of a DG accident: the economic cost and release quantity. This approach focuses on an exploratory data modelling based on US DG accident data. The authors concluded that the most influential variables are related to the failure of the container. The authors in [6] presented a model for time-dependent DG flow assignment which, unlike classical approaches, also considers the desired planning of DG trips, that could be provided by transport companies. Time-dependency is widely considered: value of risk, maximum speed, and planned flows are defined for each time interval. The case study demonstrated the effectiveness of the model applied to an area sensitive to DG transportation. In addition, to improve transport safety, transport productivity, travel reliability, informed travel choices, environmental performance and network operation flexibility, with particular attention to DG transportation, intelligent transportation systems (ITS) has been applied. The technological and economic development has produced, in the last decades, the increasing of complexity in traffic network definition and description above all in the DG transportation context. These large scale systems are often composed by many interacting subsystems and thus the required computational complexity prevents from controlling it with a centralised control structure [7]. A transportation system, which consists of geographically distributed systems, can be represented by a set of subsystems or by a System of Systems (SoS) [8] as well as a DG transportation system [9]. In these cases, whereby the set of agents solves small subproblems, dual decomposition approach [10] is attractive. The dual decomposition method allows the distribution of decision making to the subsystems, to implement a configuration for each subsystems with local coordination, and communications between neighbouring subsystems. The method of dual decomposition has been applied since the sixties, and a complete reference is the sixth chapter of [11]. In this paper, a transportation system is implemented as a network characterised by nodes, which represent subsystems, and links, that model dynamic relationships among subsystems. The current state of each node depends on discrete dynamics and evolves according to the local state of each specific subsystem and on the local state of the neighbouring This work has been sponsored by ENI R&M. 978-1-4673-0750-5/12/$31.00 ©2012 IEEE