Co-optimising network and storage systems investments through stochastic optimisation via column generation algorithms

Abstract

In the context of higher participation of renewable generation in power systems worldwide, it is critical to capture the variable nature of these energy sources in investment planning models. Furthermore, an optimal investment plan of complementary generation, transmission, and storage infrastructure for the integration of renewable generation has to recognise the flexible means necessary to deal with its variable outputs. To do so, investment planning models have to consider higher time resolution and a more detailed model of operation, which renders models intractable. Further computational complexities are needed to capture the increased levels of long-term uncertainty, due to evolving policy and market parameters, such as subsidies to renewables, investment costs of generation and storage technologies, among others. Hence, we propose a multi-stage stochastic network expansion program and its associated decomposition algorithm, which is able to co-optimise network and energy storage assets, properly capturing long-term uncertainties through a scenario tree representation of various possible future evolutions of model s parameters. Additionally, the proposed model considers high resolution in the operation, with an hourly representation, and incorporates unit commitment constraints, to properly capture the inflexibilities of the current infrastructure. Due to these features, the model is able to plan for future flexible systems, such as energy storage systems, needed to deal with the variability of increased volumes of renewable generation. To handle the increased computational burden produced by the operational details considered, we represent the yearly operation of the system by a set of typical days/weeks, and solve the problem utilising a Dantzig-Wolfe decomposition with an improved column generation approach. The novel characteristic of our algorithm is the day/week-based decomposition utilised to generate new columns, which is beyond the classic scenario tree node-based decomposition reported in existing literature. Through various case studies on three different power networks, we validate our model, study key features of planning network and storage facilities under uncertainty, and demonstrate the scalability of the proposed approach. In this vein, we use the IEEE 24-busbar network for validation and derivation of key insights of planning future flexible networks. Then, we test computational performance of our algorithm on the IEEE 118-busbar network, demonstrating the benefits of the day/week-based decomposition against the classic scenario tree node-based decomposition. Finally, we study the Australian power system where investments in large pumped storage hydro facilities are being coordinated with investments in key transmission corridors. Our case studies demonstrate the significant option value of storage facilities, helping to defer investments in new corridors, waiting for more information to be available in the future that will support a better decision making. Our case studies also show the distortions caused by neglecting operational details in network expansion planning, particularly, the value of flexibility is considerably decreased, as investment on flexible assets is significantly lower than case studies with higher operational details. Finally, our enhanced model of Dantzig-Wolfe decomposition is able to solve instances of multi-stage stochastic planning problem that can not be solved by the most recent version of the algorithm available in the literature.