Connection imbalance in low voltage distribution networks

On British electricity distribution networks, the phase to which single phase loads and generators are connected is, in most cases, unknown. There is concern that large imbalances in connection will limit the capacity of the network to support distributed generation as well as the electrification of heating and transport. The roll-out of Smart Metering in Britain, expected to be completed by the end of 2020, provides Distribution Network Operators with a means to predict the phase of single phase connections and more accurately assess the impact of increased distributed generation. This thesis examines these possibilities. There are three main sections: 1. Development of a steady state LV feeder modelling program allowing for flexible definition of connection imbalance and suitable for use with a supercomputer. 2. Development of a stochastic method to assess the combined influence (on voltages, currents and losses) of connection imbalance and photovoltaic generation. 3. Creation of an algorithm for the prediction of phase connections using Smart Meter Data, based on the GB smart metering proposals. The LV feeder model uses an unbalanced load flow based on network reduction and re-expansion with nodal analysis. It was validated using PSCAD. The feeder model uses a TNS earthing arrangement; this was shown to be equivalent to TN-C-S in normal operation, allowing for simpler modelling. A metric for connection imbalance was introduced – the highest proportion of houses connected to any phase. The model is capable of varying connection imbalance by changing the phase to which each house is connected. The connection imbalance was varied by randomly allocating houses to different phases. Demand profiles were created stochastically and PV generation was added to a varied proportion of houses (0 to 100% in 10% steps). More than 19 million unbalanced load flow calculations were performed using a supercomputer. It was found that, for a typical urban feeder serving residential properties, connection imbalance is not a significant problem for DNOs until it becomes severe (>60% of houses on one phase). The phase identification algorithm combines two methods found in the literature; voltage measurement clustering and solution of the subset sum problem. It uses vii smart meter voltage profiles and active power profiles with current measured at the supply substation. It correctly predicted the phase connection for 97% of smart meters, using simulated data representing a set 100 different connection configurations, across 6 different days (different sets of demand profiles) with a measurement averaging timeframe of 30 minutes