| This thesis presents a multi-agent cooperative search approach to optimize the operation of large and complex natural gas pipeline networks. The proposed approach is based on a biologically-inspired computational model, namely particle swarm optimization, and the main objective is to determine the control parameters of a natural gas transmission pipeline network that result in optimal operation while maintaining the desired throughput and satisfying given system constraints. Solving this problem is not trivial given the large number of decision variables and constraints, the nature of the objective functions and the limited time available to obtain the solution. The used cooperative search approach improves on the pure competition of search agents by the sporadic exchange of solutions which are integrated into the search state of the search agents and also used to improve their search control. The approach exploits the strength of particle swarm optimization in each agent to deal with high-dimensional problems that include a mix of discrete and continuous decision variables. Incorporation of domain knowledge into the search allowed the identification of solutions that can be pre-evaluated without having to use the time consuming pipeline simulations used to measure the quality of a solution. The experimental evaluation with real problem instances from TransCanada PipeLines Ltd. show that the developed system meets the industry's time requirements and, for large and complex pipeline networks, it reliably outperforms the current state-of-the-art interactive method by creating solutions that require at least 12% less energy, reducing not only the transportation costs but also the amount of greenhouse gas emissions being dissipated to the atmosphere. |
