Stochastic planning of electricity and gas networks: An asynchronous column generation approach

Planning networks within a multi-stage stochastic framework is becoming critical for improving the economic performance of investment decisions against the present levels of uncertainty. This problem, however, has been proved extremely challenging to be solved on real networks, especially when considering the interactions among various energy vectors. In this context, this paper proposes the use of Dantzig-Wolfe decomposition and parallel asynchronous column generation to solve a multi-stage stochastic planning of an integrated power and natural gas system, including non-linear effects of gas compressors reformulated in a mixed integer linear programming fashion. We compare the computational performance of the proposed approach against two alternatives: a parallel synchronous column generation approach and the counterfactual, monolithic approach, where the mixed integer linear program (without decomposition) is directly solved by a commercial solver. Our sources of long-term uncertainty are the locations and volumes of (i) new renewable generation (which may depend on policy objectives, regulatory incentives, etc. that are constantly evolving) and (ii) new demands. The model also ensures that the planned energy infrastructure can effectively be operated reliably against a large array of operating conditions originated by high variability of renewable generation outputs, multiple demand levels and hydro inflows. Through various case studies, we discuss and demonstrate the importance of stochastic and integrated planning of electricity and natural gas systems along with the benefits of asynchronous algorithms and decomposition techniques that can be parallelized.