Fiodar
Kazhamiaka,
PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
In 2018, nearly two-thirds of newly installed global power generation has come from renewable energy sources. Distributed installations of solar photovoltaic (PV) panels have been at the forefront of this global energy transition. In many places, the cost of solar power has dropped below the price of fossil fuels such as coal. The main challenge in incorporating this growing source of clean and cheap energy is its high variability; it must often be used in conjunction with an expensive energy storage system to help match electricity supply and demand. Despite the growing focus on energy storage and its role in helping meet the ambitious renewable energy targets set by climate-conscious policy makers, the relatively high capital cost of combined PV-storage systems has limited their widespread adoption.
The high cost of PV-storage systems may be offset by the value they provide to system owners. The combination of PV panel and energy storage components adds complexity and flexibility to an energy system where both supply and demand are stochastic and depend on many factors, and this contributes to the technical challenges of system design and operation. Practical methods for increasing the value of PV-storage systems through effective system design and operation are the focus of this dissertation. The proposed approach to solving these problems involves the use of system models and data describing the system's operating environment. Our research consists of two main components: the theoretical component, i.e., modelling, and the practical component, i.e., analysis of energy systems on the basis of models and data.
In this thesis, we construct new battery models that enable simulation and optimization studies of PV-storage systems. These models are then used to develop several advanced methods for designing and operating PV-storage systems based on available solar generation and electricity consumption data. We study the problem of determining the combined sizes of PV panel and storage components to meet a given system load target at the lowest possible cost. We also study the problem of system operation, with the objective of increasing system value via minimization of operating expenses. The sizing and control methods developed in this thesis are based primarily on system simulation, mathematical programming, and neural networks, and are evaluated on datasets of PV generation and electricity consumption measurements of buildings.
Among our contributions are an accurate battery model for simulation studies that can be easily calibrated. We further derive models for optimization studies with various degrees of accuracy and complexity, including a linear model with higher accuracy than existing linear models. For system sizing, we develop a novel approach to jointly size a PV-storage system to reliably meet a load performance target at the lowest possible cost. For system operation, we develop a set of algorithms for low-cost system operation, a mixed-timescale approach to reduce the online computational complexity of model predictive control, and a system controller designed to encode a deep neural network with a model predictive control policy and capable of refining its performance over time while adapting to changes in the operating environment.