Short-Term Electricity Market Price Forecasting Based on De-Noised Wavelets and NARX Neural Network- Data Analytics Approach

Abstract

Electricity price data is often non-linear and highly volatile. Under a weather and climate disaster event, price forecasting represents a challenging task. Noise in electricity price data is commonly affected by several factors such as season, weekend or workday, critical event, etc. In this study, the proposed model uses a de-noised wavelet as a pre-processing algorithm to reduce price noise characteristics and a Non-linear Auto-Regression eXogenous (NARX) Neural Network (NN) for the data analytic approach. To test price forecasting, a seasonal week-ahead (168 hrs.) window is used. The forecasting models are evaluated using the Mean Absolute Percentage Error (MAPE). The model and methodology proposed show a remarkable improvement over standard methodologies, complemented by data visualization.