Browsing by Author "Roux, Marcel"
Now showing 1 - 1 of 1
Results Per Page
- ItemLoad management of electric water heaters in a smart grid through forecasting and intelligent centralised control(Stellenbosch : Stellenbosch University, 2018-03) Roux, Marcel; Booysen, M. J.; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.ENGLISH ABSTRACT: Globally utilities are facing increasing demand and numerous challenges arise with the supply and management thereof. The South African electricity utility Eskom is at present still facing difficulty with meeting the country's growing demand. One of the largest consumers of energy in the residential sector has been identified as the domestic electric water heater (EWH). To manage significant peak loads, electricity utilities employ demand side management (DSM) strategies to throttle demand in order to maintain stability in the electrical grid. Ripple control is such a strategy which is a blunt, unidirectional control scheme which toggles electrical supply to zones of EWHs at times with no consideration for the comfort of individual consumers. Smart grid (SG) technology is on the rise and emerging Internet of things (IoT) technology augments the adoption of SG to address the problem of DSM. The data collected from a SG is of high value for knowledge discovery and many advantages can be obtained from effective analysis of this data. This study utilises data obtained through the Geasy project which presents a smart EWH controller to enable the monitoring and control of EWHs with resolution of 1 minute. This study presents a three-part look at different aspects of the data aimed towards the development of a cogent, data-driven bidirectional DSM application. Of fundamental importance to data analysis is to assess the current quality of the data, due to the "garbage in, garbage out" principle. High quality data is required for analysis. After investigating potential data quality impacting factors, the Geasy data was used to develop a numerical data cleaning framework with scalability in mind. The implemented routines were tailored to the specific needs of the data fields considered, such as removing erroneous spikes and filling in missing data according to the most suitable processes. The cleaned data had vastly superior data quality and indicates that the developed data cleaning framework may provide a baseline for more advanced data cleaning steps to be employed before data warehousing. Next, the aspect of predictive scheduling was investigated. The temporal structure of one of the largest drivers of EWH usage, the hot water usage, was investigated using statistical methods including time series decomposition, autocorrelation and partial autocorrelation plots. The decomposition of the usage data indicated a strong seasonal component that indicated potential for forecasting. Linear seasonal autoregressive integrated moving average models were used to create models of the temporal structure of the usage data. Box-Jenkins parameter identification proved highly effective in estimating good, general-purpose seasonal forecasting models. The obtained forecasting results were shown to predict a daily water volume of 225 L, compared to the observed 272 L, which indicates an error of 17.3 %. However, correcting the forecast volume with the normalised observed training volume reduced the volume error to 0 %. Continuing the exploration of the value of the SG data, a DSM application was developed to balance the utility and consumer need in real time. During the development of the algorithms, a computationally efficient EWH thermal model was revised to provide improved scalability through vectorisation which also enabled the algorithms to consider multiple, micro-simulated EWHs during the macro evaluation of a microgrid. The approach uses actual individual hot water consumption patterns, measured real-time water heater temperatures and individual EWH properties as the main determinants in a cost function for a centralised scheduler. The application was evaluated against various demand and temperature limits, with actual consumption measured in a field trial of 34 EWHs for a period of measurements spanning 28 days at 1 minute resolution. For a temperature limit of 60° C, the application reduces the peak load from a measured 47 kW to 20 kW (vs. 106 kW for full ripple control). The number of undesired cold events decreases by 83.3 %, improving consumer experience, while the total grid energy consumption only increases by 12 %.