Corrected Support Vector Regression for intraday point forecasting of prices in the continuous power market

TL;DR

This paper introduces a corrected support vector regression method for very short-term intraday electricity price forecasting, demonstrating improved accuracy and speed over traditional models on German market data.

Contribution

The paper proposes a novel kernel correction in support vector regression for intraday price forecasting, enhancing accuracy and computational efficiency.

Findings

cSVR outperforms benchmarks in forecast accuracy

Highest improvements observed during morning and evening peaks

Averaging schemes further improve forecast performance

Abstract

In this paper, we develop a new approach to the very short-term point forecasting of electricity prices in the continuous market. It is based on the Support Vector Regression with a kernel correction built on additional forecast of dependent variable. We test the proposed approach on a dataset from the German intraday continuous market and compare its forecast accuracy with several benchmarks: classic SVR, the LASSO model, Random Forest and the na\"{i}ve forecast. The analysis is performed for different forecasting horizons, deliveries, and lead times. We train the models on three expert sets of explanatory variables and apply the forecast averaging schemes. Overall, the proposed cSVR approach with the averaging scheme yields the highest forecast accuracy, being at the same time the fastest from the considered benchmarks. The highest improvement in forecast accuracy is obtained for deliveries in the morning and evening peaks.