Surrogate Models for Maximizing Net Present Value of Renewable Energy Sources

点击翻译按钮获取中文摘要

Hydro-Québec supplies electricity to customers in remote off-grid locations by operating diesel plants. Cost and environmental concerns motivate the consideration of using several types of renewable energy sources. The optimal combination of different energy sources can be determined by solving an optimization problem to maximize the net present value (NPV) generated by the off-grid electrical network over its lifetime while accounting for construction, fuel, maintenance and dismantling costs. However, this requires numerical simulations that are associated with high computational cost and discontinuities in both functions and variables, which does not allow the computation of gradients. IREQ has tried conventional techniques to address these challenges in order to solve the simulation-based optimization problem in practical time frames, but has not had any success. This project aims at developing and implementing a surrogate-assisted, derivative-free optimization framework to alleviate these challenges. We will develop surrogate models that are based on Kernel Smoothing (KS) and Radial Basis Functions (RBF) and exploit our knowledge of the problem and its structure. These surrogate models will then be integrated with a particular class of derivative-free optimization algorithms termed Mesh-Adaptive Direct Search (MADS). Our optimization framework will use the surrogate models to explore the design space locally and to accelerate local convergence, and will use the expensive simulations for final decision-making at each iteration. The outcome of this research will provide Hydro-Québec with a powerful tool for off-grid network optimization that can consider renewable energy sources to offer economically viable, robust and environmentally benign energy solutions for isolated communities.