Scholarlink

<p>Accurate forecasts of electricity prices are crucial for energy plant managers, policymakers, investors, traders, and speculators. In this regard, machine learning is receiving growing attention in forecasting electricity prices across international markets. The main purpose of the current study is to implement and optimize various machines learning systems in the task of forecasting electricity price across five different international electricity markets, namely Australia, Denmark, Norway, South Korea, and the USA. Specifically, Bayesian optimization (BO) method is employed to tune the hyper-parameters of the Gaussian regression process (GRP), support vector regression (SVR), regression trees (RT), k-nearest neighbors (<i>k</i>NN), and deep feedforward neural networks (DFFNN). All the optimized predictive systems were validated on daily and intraday datasets. The simulations showed that on average, the SVR-BO yields the lowest forecasting error measured by root mean of squared errors. It is followed by the RT-BO, <i>k</i>NN-BO, GRP-BO, and DFNN-BO. Therefore, the SVR-BO is a promising machine learning system to predict international electricity markets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting International Electricity Market Prices by Using Optimized Machine Learning Systems

  • 期刊论文

  • 发表日期：

    2025年2月18日

查看全文
  • Salim Lahmiri

摘要

Accurate forecasts of electricity prices are crucial for energy plant managers, policymakers, investors, traders, and speculators. In this regard, machine learning is receiving growing attention in forecasting electricity prices across international markets. The main purpose of the current study is to implement and optimize various machines learning systems in the task of forecasting electricity price across five different international electricity markets, namely Australia, Denmark, Norway, South Korea, and the USA. Specifically, Bayesian optimization (BO) method is employed to tune the hyper-parameters of the Gaussian regression process (GRP), support vector regression (SVR), regression trees (RT), k-nearest neighbors (kNN), and deep feedforward neural networks (DFFNN). All the optimized predictive systems were validated on daily and intraday datasets. The simulations showed that on average, the SVR-BO yields the lowest forecasting error measured by root mean of squared errors. It is followed by the RT-BO, kNN-BO, GRP-BO, and DFNN-BO. Therefore, the SVR-BO is a promising machine learning system to predict international electricity markets.