Understanding the Drivers of Negative Electricity Price Using Decision Tree

Abstract

This paper proposes a hybrid approach, integrating Decision Trees (DT) and Artificial Neural Networks (ANN) for energy price classification in deregulated electricity market. The proposed model does not aim to predict future values of energy prices, but classify and explain the negative Locational Marginal Price (LMP) that are observed in the grid. The negative LMPs are grouped by the K-means technique and then taken as a target. Starting from a large set of potential variables that influence the price of energy, a feature selection technique is applied to identify the most relevant attributes in pricing. The C5.0 and ANN algorithms are used for classification. The California ISO market is used as a test system to demonstrate the proposed approach. The results show that an ensemble model of C5.0 and ANN produced high accuracy, and can be an interesting tool to explain the occurrence of negative prices.