Hybrid Kolmogorov-Arnold Network and XGBoost Framework for Week-Ahead Price Forecasting in Australia’s National Electricity Market
Houxuan Zhou 1 , Sriram Prasad 1 , Chenghao Huang 1 , Jiajie Feng 2 , Hao Wang 1,3* · 2026
Read Hybrid Kolmogorov-Arnold Network and XGBoost Framework for Week-Ahead Price Forecasting in Australia’s National Electricity Market by Houxuan Zhou 1 , Sriram Prasad 1 , Chenghao Huang 1 , Jiajie Feng 2 , Hao Wang 1,3* (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-22387.txt.
Introduction
Zhou and colleagues propose a hybrid framework for week-ahead electricity price forecasting in Australia's National Electricity Market. It pairs Kolmogorov-Arnold Networks with XGBoost. The hybrid pairs the global nonlinear representation of Kolmogorov-Arnold Networks with the local robustness of XGBoost. It captures long-term dependencies and short-term price swings. Tests on real market data with an expanding window show it beats SARIMAX, LSTM, standalone KAN and XGBoost. Mean absolute error falls about 12 percent against XGBoost and over 50 percent against a naive baseline.
Catalog record
| CREATOR | Houxuan Zhou 1 , Sriram Prasad 1 , Chenghao Huang 1 , Jiajie Feng 2 , Hao Wang 1,3* |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27072 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.22387
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hybrid Kolmogorov-Arnold Network and XGBoost Framework for Week-Ahead Price Forecasting in Australia’s National Electricity Market *Corresponding author: Hao Wang ([email protected]). This work was supported in part by the Australian Research Council (ARC) Discovery Early Career Researcher Award (DECRA) under Grant DE230100046. Houxuan Zhou 1 , Sriram Prasad 1 , Chenghao Huang 1 , Jiajie Feng 2 , Hao Wang 1,3* Abstract Accurate electricity price forecasting (EPF) is essential for market participants to support operational planning and risk management, yet remains challenging due to strong volatility, nonlinear dynamics, and frequent extreme price spikes. These challenges are particularly pronounced in the Australian National Electricity Market (NEM), where high renewable penetration further increases uncertainty. This paper investigates week-ahead electricity price forecasting and proposes a hybrid KAN+XGBoost framework that integrates Kolmogorov–Arnold Networks (KAN) with tree-based learning. The proposed approach combines the global nonlinear representation capability of KAN with the local robustness of XGBoost to capture both long-term dependencies and short-term price fluctuations. Experiments are conducted on real-world NEM data using an expanding window evaluation strategy. The results demonstrate that the proposed model outperforms benchmark methods, including SARIMAX, Long Short-Term Memory (LSTM), standalone KAN, and XGBoost, reducing MAE by approximately 12% compared to XGBoost and by over 50% compared to a naive baseline. The results suggest that hybrid learning strategies provide an effective and robust solution for electricity price forecasting in highly dynamic electricity markets. I Introduction Electricity price forecasting (EPF) plays a critical role in modern industrial energy systems, where accurate price signals are essential for market participation, risk management, and operational planning [ 4 ] . Industrial stakeholders, including generators and large industrial consumers, rely on reliable forecasting to support resource scheduling and financial decision-making. Electricity prices are inherently volatile and influenced by complex supply–demand dynamics, including demand fluctuations, weather conditions, renewable generations, generation costs, and network constraints [ 11 ] . These challenges are particularly pronounced in Australia’s National Electricity Market (NEM), where prices are settled at five-minute resolution, leading to increased short-term volatility. Recent studies have also highlighted the growing importance of flexibility management and market coordination in modern electricity systems with high renewable penetration [ 19 ] . In addition, the growing penetration of variable renewable energy (VRE), such as wind and solar, further amplifies this complexity. VRE sources are dispatched first under the merit-order effect due to their near-zero marginal costs. However, their intermittent nature increases market uncertainty and contributes to nonlinear price dynamics and abrupt fluctuations under changing supply-demand conditions [ 10 , 8 ] . As a result, EPF in the NEM requires models capable of capturing both complex nonlinear dependencies and high market volatility. Although day-ahead forecasting has been widely studied in the NEM, week-ahead hourly forecasting remains relatively underexplored despite its practical importance. Compared with short-term forecasting, week-ahead prediction provides longer lead time for bidding, generation scheduling, and risk hedging, making it particularly valuable under increasing renewable penetration and medium-term operational planning [ 24 , 22 ] . Traditional statistical models, such as Autoregressive Integrated Moving Average (ARIMA) [ 3 ] and Generalized Autoregressive Conditional Heteroskedasticity (GARCH) [ 2 ] , can capture seasonality and volatility patterns, but they often struggle with nonlinear dynamics and extreme price spikes [ 17 , 22 ] . Extensions, such as Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX) and Vector AutoRegression (VAR) models, further incorporate exogenous variables and interdependencies, improving modeling flexibility in electricity price forecasting [ 3 , 20 , 22 ] . More recently, machine learning and deep learning approaches have gained increasing attention. Neural network architectures, including Long Short-Term Memory (LSTM) [ 12 ] and Gated Recurrent Unit (GRU) [ 7 ] , have shown a strong capability in modeling nonlinear relationships and temporal dependencies, often outperforming traditional statistical models [ 13 , 14 , 9 ] . In particular, recurrent models are well-suited for capturing sequential patterns and long-term dependencies in time-series data. However, their performance remains sensitive to noise and instability when applied to longer forecasting horizons and highly volatile market conditions [ 13 , 14 ] . Furthermore, hybrid and ensemble approaches have been explored to improve forecasting performance by leveraging complementary model strengths [ 16 , 18 ] . While these approaches have shown promising results, they are often designed for short-term forecasting settings and may not fully address the challenges of capturing both global structures and local irregular variations in more complex environments such as the NEM. Consequently, existing approaches typically face a trade-off between global representation and local robustness, making them insufficient for reliable week-ahead forecasting in the NEM. This limitation highlights the need for more effective models that can capture both global and local price dynamics. Kolmogorov-Arnold Networks (KAN), based on the Kolmogorov-Arnold representation theorem [ 15 ] , provides a flexible functional approximation framework by learning univariate transformations. This design enables KAN to effectively capture complex nonlinear dependencies in electricity price series, making it particularly suitable for modeling global functional relationships in EPF. In addition, its use of univariate functions improves interpretability compared to conventional neural networks, and recent studies have demonstrated its competitive performance in time-series forecasting tasks [ 21 ] . In contrast, tree-based models such as XGBoost are inherently robust to local irregularities and outliers due to their piecewise partitioning mechanism, which allows them to effectively handle abrupt changes in data [ 6 ] . However, their piecewise nature may limit their ability to capture smooth global dependencies and long-term temporal patterns over extended forecasting horizons. The complementary characteristics of KAN and XGBoost highlight the limitations of using either model independently. KAN may struggle with localized irregularities and extreme price spikes, while XGBoost is less effective in capturing smooth global dependencies over longer horizons. Motivated by this challenge, this paper proposes a hybrid KAN+XGBoost framework for week-ahead electricity price forecasting in the NEM. From a functional decomposition perspective, KAN is used for learning global nonlinear structures, while XGBoost is effective in modeling residual local deviations, thus enabling a balanced representation of long-term dependencies and short-term irregular variations. • This study addresses the challenge of week-ahead electricity price forecasting in the NEM, where models must simultaneously handle long-term dependencies and short-term extreme price variations. The proposed framework provides longer-term foresight that can support more informed decision-making for market participants. • A hybrid KAN+XGBoost framework is proposed, which integrates functional representation learning with tree-based partitioning. This design enables the model to jointly capture global nonlinear structures and local irregular variations in electricity price dynamics. • Extensive experiments on real-world NEM data demonstrate that the proposed approach consistently outperforms statistical, machine learning, and deep learning baselines across multiple states. The model achieves approximately 10-15% improvement in MAE over strong benchmarks, while maintaining robust performance under diverse market conditions. The remainder of this paper is organized as follows. Section II introduces the methodology. Section III describes the experimental setup. Section IV presents and discusses the results, and Section V concludes the paper. II KAN–XGBoost Forecasting Framework This section presents the proposed framework for week-ahead electricity price forecasting in the NEM. We first define the forecasting problem, followed by data preparation and feature engineering procedures. Finally, the proposed hybrid KAN–XGBoost model is introduced. An overview of the proposed