Multivariate and Multistep Forecasting of System Marginal Price Using a Modified WaveNet

EasyChair Preprint 13404

Abstract

This paper explores the application of a WaveNet deep learning model structure for forecasting of System Marginal Price (SMP) in the context of the electrical grid, focusing on the multivariate and multi-step prediction challenges inherent in real-world application. SMP, the real-time cost of balancing for supply and demand for electricity, it’s inherently dynamic and volatility nature in response to numerous factors make the SMP forecasting challenging. By applying a WaveNet model renowned for its proficiency in capturing temporal relationships in time series data, with modifications, we aim to navigate the complexities of SMP forecasting, providing insights and methods that could potentially benefit such dynamic times series applications. The effectiveness of the model is evaluated by comparing with three other state-of-the-art deep learning models: Artificial Neural Network (ANN), Convolutional Neural Network (CNN) and long short-term memory (LSTM) models. The results show that the modified WaveNet model is well-suited for the multivariate and multi-step regression-type problems.

Keyphrases: Multistep forecasting, System Marginal Price (SMP), WaveNet, multivariate, time series

@booklet{EasyChair:13404,
  author    = {Jaeyun Jung and Minkyu Lee and Daegun Ko and Jeonghoon Choi and Hyeseung Han and Bumsu Park and Nayeon Park and Kyoungjoo Kim and Hyunsup Kim and Sungkyu Kim},
  title     = {Multivariate and Multistep Forecasting of System Marginal Price Using a Modified WaveNet},
  howpublished = {EasyChair Preprint 13404},
  year      = {EasyChair, 2024}}