go-v.orgdoi

Natural Gas Consumption Forecasting with Kolmogorov–Arnold Networks: A Comparison with MLP

Kürşad Arslan, Emrah Dönmez · Sakarya University Journal of Computer and Information Sciences · 2025

  • DOI 10.35377/saucis...1759966
  • Publisher OpenAlex
  • Type article
  • Issued 2025

Not cached yet. This record was fetched live for this request and is stored after the lookup. Reopening the page should serve the cached copy.

Sakarya University Journal of Computer and Information Sciences · OpenAlex · article · 2025

Abstract

Natural gas remains a vital resource for meeting residential heating energy needs, particularly during the winter months. Accurate demand forecasting is essential for maintaining supply-demand balance, optimizing operational costs, and supporting effective energy management. In this study, the natural gas consumption prediction performance of Kolmogorov-Arnold Networks (KAN), a new neural network architecture, was compared with that of the basic model, Multi-Layer Perceptrons (MLP). Both models were trained and tested on the same dataset using monthly consumption data.

While MLPs are diversified through the number of neurons, activation functions, and layer configuration, KAN models are configured by modifying B-spline parameters, grid size, and layer structure. The results show that the KAN model achieved the highest R2 value despite having fewer trained parameters. Although some versions of the MLP model yielded lower Mean Absolute Percentage Error (MAPE) values, they fell short of KAN in terms of overall fit. These findings demonstrate the superior ability of KAN to capture nonlinear patterns in energy demand forecasting, offering computational efficiency.

Open the resolved locator · 10.35377/saucis...1759966

Metadata licence: OpenAlex metadata (CC0)

Registry
OpenAlex DOI index + API
Fetched
2026-11-20 07:59 UTC
Cache state
not cached yet (bulk-index+openalex)
Locator
https://dergipark.org.tr/en/pub/saucis/article/1759966
Metadata licence
OpenAlex metadata (CC0)
Scope
Metadata as returned by the resolver; abstracts are never generated here.