Predictive residential energy management with photovoltaic power generation and battery energy storage

["UCTD", "Photovoltaic power generation", "Battery energy storage", "Power supply", "Predictive control"]

Coetzee, Stef ; Mouton, Toit

Stellenbosch University

Others  :  https://scholar.sun.ac.za:443/bitstream/10019.1/105979/1/coetzee_predictive_2019.pdf

瑞士|英语

来源: Stellenbosch University

【 摘 要 】

ENGLISH ABSTRACT: Over the course of the past decade, South African national energy utility Eskom hasincreased its average electricity rate more than fourfold as it finds itself in financial difficulty,brought about by a myriad causes. During the same time period, the cost of solarphotovoltaic arrays and battery energy storage has fallen by more than two thirds.In this thesis, a residential energy management system which incorporates small-scalesolar photovoltaic power generation and battery energy storage is developed. The primarygoal of the system is to increase self-sufficiency of a given household through managementof the battery energy storage unit and two controllable loads: an air-handling unit and anelectric water heater. Such a system would be able to shield residences, at least in part,from the energy utility's ongoing challenges.A grid-connected household, featuring each of the controllable electrical entities mentioned,as well as a photovoltaic array, and a generic non-controllable load is described.Due to the intermittent nature of solar radiation, potential solar power generation is inevitablylost because of power supply-demand misalignment. Model predictive control,a popular process-control technique, is exerted over the residential system in pursuit ofresolving this misalignment.At a sampling time of ten minutes, a predictive controller capable of an hour (or sixsteps) of model-based foresight is formulated and tested in simulation. A rules-based hierarchicalcontroller is used as baseline against which the predictive control scheme isevaluated. The controller's ability to reduce solar power curtailment is confirmed by evaluatingits performance with relevant data from each of the four seasons (from September2017 to August 2018), for prediction horizons one through six.