Computational intelligence in extra low voltage direct currrent pico-grids

Abstract

The modern power system has gone through a lot of changes over the past few years. It is no longer about providing one-way power from sources to various loads. Power monitoring and management have become an increasingly essential task with the growing trend to provide users more information about the status of the loads within their energy consumption so that they can make an informed decision to reduce usage and cost or request desired maintenance. Computational intelligence has been successfully implemented in the electrical power systems to aid the user, but these research studies about this are generally conducted on the conventional alternative current (AC) macro-grids. Until now, little work has been done on direct current (DC) and the focus on smaller DC grids has been even less. In recent years, the evolution of electrical power system has seen the proliferation of direct current (DC) appliances and equipment such as buildings, households and office loads. This number keeps increasing with the advancement in technology and consumer lifestyles changes. Given that DC power supplies are getting more popular in the form of photovoltaic panels and batteries, it is possible for Extra Low Voltage (ELV) DC households or office pico-grids to come into use soon. This research recognises and addresses this research gap in the monitoring and managing of the DC picogrids. It recommends and applies the bottom-up monitoring and management approach in smaller scale grids and in larger scale grids. It innovatively categorises the loads in the grids into dumb loads that do not have intelligence and communication features and smart loads that have these features. While targeting at these ELV DC pico-grids, this research presents solutions that provide users useful information on load classification, load disaggregation, anomaly warning and early fault detection. It provides local and remote sensing with the alternative use of hardware to lessen the computational burden from the main computer. The inclusion of remote monitoring has opened a window of opportunities for Internet of Things (IoT) implementation. These solutions involve the blending of computational intelligence techniques with enhanced algorithms, such as K-Means algorithm, k-Nearest Neighbours (kNN) classification, Naïve Bayes Classification (NBC) Theorem, Statistical Process Control (SPC) and Long Short-Term Memory Recurrent Neural Network (LSTM RNN). As demonstrated in this research, these solutions produce high accuracy results in load classification and early anomaly detection in both AC and DC pico-grids. In addition to the load side, this research features a short-term PV energy forecasting technique that is easily comprehensible to users. This research contributes to the implementation of the Smart Grid with possible IoT features in DC pico-grids.