A Machine Learning Approach for Understanding Power Distribution System Congestion

This study proposes a novel method for learning the congestion level of the power distribution system which designates the loading of a distribution feeder. A machine learning approach is proposed here to find a relating function between substation feeder power and local voltage. This model is, then, used to estimate real-time substation feeder power consumption using current local voltage measurements. This fully decentralized estimation of substation power consumption could facilitate more electric vehicle integration into the distribution grid without the need for real-time centralized control by a system aggregator. The concept is tested with real power loading data of a feeder located in the state of Alabama. The local voltage measurement data of a typical house residing in the downstream of the network of the same feeder is used to develop the learning algorithm.