On optimal pooling of renewable energy sources through risk-averse portfolio optimization methods

Abstract

This dissertation investigates approaches to achieve optimal pooling of renewable energy sources through risk-averse portfolio optimization methods. The first chapter aims at evaluating the potential for an approach targeted at addressing the issue of limited predictability of wind energy generation, as opposed to intermittency, which has been previously considered in the literature. Specifically, a portfolio optimization model for intelligently constructing a wind energy portfolio for a given harvesting region with the goal of reducing the prediction error is proposed. The mathematical model, based on Conditional Value-at-Risk (CVaR) optimization methodology, is used to evaluate potential improvement in (day ahead) generation predictability for a collection of locations in the USA. The study concludes that pooling indeed can significantly reduce wind energy generation forecasting error, with the effect largely dependent on the size of the harvesting region. Further, if advanced optimization techniques are used, it is possible to balance this reduction with average generation output. The second chapter aims to evaluate the impact of reducing limited predictability on battery sizing through creating a ’proof-of-concept’ experiment. In particular, a heuristic approach is proposed to the problem of simultaneous optimization of generation portfolio and battery sizing. The mathematical models is a bi-level problem, based on conditional risk value (CVaR) optimization methodology on the first level, and a operational planning problem to evaluate installed battery capacity on the second. The study concludes that the heuristic approach with pooling significantly reduces required battery capacity according to operational plans, and the effect varies on the size of the harvesting region or the degree of the technology combination. Further, it is possible to diversify the pooling leading to overall operational cost reduction. Consequently, the results imply that the positive effect of pooling diverse wind resources can be an important factor in planning for generation expansion projects. Lastly, the third chapter considers a similar multilevel modeling approach, but solves it exactly. The mathematical model is used to evaluate potential cost improvement in a Virtual Power Plant (VPP). In particular, at the first level, Mean conditional risk value (Mean-CVaR) optimization model is used to create an optimal portfolio for minimizing intermittency. The second level of the optimization procedure is based on linear programming for operation planning to minimize the total operation cost. The bi-level problem is solved by employing optimality conditions. The study shows that the multilevel model leads to significant savings in total operation cost compared to the benchmark (heuristic methodology). Further, the total cost is also significantly decreased as the pooling region is increased. This improvement is directly related to the size of the harvesting region. Consequently, this research contributes to the operational planning in VPPs through a multilevel model that advanced pooling approaches.