State-of-the-Art of Electricity Price Forecasting in a Grid Environment

TLDR: In this article, the authors classify forecasting techniques based on their objective, concept, time horizon, input-output specification, and level of accuracy, and demonstrate a hybrid forecasting system, which combines fuzzy inference system and least-squares estimation.

Abstract: The purpose of electricity price forecasting is to estimate future electricity prices, particularly locational marginal prices (LMP), with consideration to both security and capacity constraints in a grid environment. Electricity price forecasting is vital to both market participants and market operators in wholesale electricity markets. Electricity price forecasts are used to assist the decision making of market participants on bidding submissions, asset allocations, bilateral trades, transmission and distribution planning, and generation construction locations. Electricity price forecasts are also used by market operators to uncover possible market power. The inaccuracy of electricity price forecasting is due to problems associated with volatility of prices, interpretability of explanatory variables, and underlying impacts of power grid security. This study classifies forecasting techniques common in the literature based on their objective, concept, time horizon, input–output specification, and level of accuracy. Thus the state-of-the-art of electricity price forecasting is described in this study. This survey facilitates the validation, comparison, and improvements of specific or combined methods of price forecasting in competitive electricity markets. Moreover, this study demonstrates a hybrid forecasting system, which combines fuzzy inference system and least-squares estimation. The proposed mechanism is applied to the day-ahead electricity price forecasting of an actual security-constrained, wholesale electricity market. This hybrid forecasting system provides both accuracy and transparency to electricity price forecasts. The forecasting information is also interpretable with respect to the fuzzy representations of selected inputs.