Load profiling on time and spectral domain : from big data to smart data

With the promotion of demand side responses (DSRs) and low carbon technologies (LCTs), there is a growing interest in visualising the demand information at individual consumer and low voltage (LV) network level, where demands are less aggregated and highly volatile. Yet, traditional load profiling techniques, which are carried out on small data, are struggling to meet the requirements on accuracy and granularity. This thesis contributes to this area by extending traditional load profiling to a big-data context, where refined load profiles (smart data) can be extracted by two novel load profiling techniques for LV networks and individual consumers. The refined load profiles aim to: i) economically visualise LV networks with limited smart-grid monitoring data; ii) transform the smart metering data into a high-detail granular representation of the customers’ daily demand. For the LV networks, this thesis develops a novel concept, LV network templates, which aim to visualise the LV networks in a cost-effective manner. A novel three-stage load profiling method is proposed as: clustering, classification and scaling. By using statistical time-series analysis, three steps are undertaken: i) cluster a vast amount of load data according to their load shapes; ii) classify un-monitored substations to the most similar cluster without sample metering; iii) and also scale them to the right magnitude without sample metering. Through this method, limited representative monitoring data can be used to develop a library of typical load profiles for un-monitored networks, thus saving the cost of extensive monitoring for every single substation. In addition, it is the first load profiling method that can accurately express both load shapes and magnitudes for LV networks. Regarding the customer’s demand representation, the developed time-series analysis needs to be updated due to the volatile and uncertain nature of smart metering data, including inter-related factors such as overall load shapes, sudden spikes and magnitudes. Therefore, an innovative spectral load profiling is proposed to decompose these factors into different spectral levels, characterised by spectral features. By analysing the extracted features on each spectral level separately through multi-resolution analysis, the interference among different factors can effectively be prevented. The proposed method, for the first time, is able to fully capture the energy characteristics at the household level. The developed LV network load templates provide an economical but straightforward way to quantify the available headroom of unmonitored substations over time, providing quantitative information for distribution network operators to integrate LTCs at the minimal costs. The spectral load profiling gives an insight into customer’s energy behaviours with high granularity and accuracy. It can support the customer-specified DSR, tariff design, smart metering validation and load forecasting.