期刊
出版社
ASSOC COMPUTING MACHINERY
DOI: 10.1145/3208903.3208923
关键词
Advanced metering infrastructure (AMI); k-means clustering; neural networks; random forest; support vector regression; smart meter; load profiles; electrical power load forecasting; short-term electricity forecasting
资金
- Engineering and Physical Sciences Research Council, Centre for Doctoral Training in Cloud Computing for Big Data [EP/L015358/1]
In order to reliably generate electricity to meet the demands of the customer base, it is essential to match supply with demand. Short-term load forecasting is utilised in both real-time scheduling of electricity, and load-frequency control. This paper aims to improve the accuracy of load-forecasting by using machine learning techniques to predict 30 minutes ahead using smart meter data. We utilised the k-means clustering algorithm to cluster similar individual consumers and fit distinct models per cluster. Public holidays were taken into consideration for changing customer behaviour, as was periodicity of the day, week and year. We evaluated a number of approaches for predicting future energy demands including; Random Forests, Neural Networks, Long Short-Term Memory Neural Networks and Support Vector Regression models. We found that Random Forests performed best at each clustering level, and that clustering similar consumers and aggregating their predictions outperformed a single model in each case. These findings suggest that clustering smart meter data prior to forecasting is an important step in improving accuracy when using machine learning techniques.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据