A short-term power load forecasting method for industrial sectors based on an improved correlation analysis

Abstract

Accurate forecasting of short-term power load for industrial sectors plays a pivotal role in ensuring the safe and economic operation of regional power grids. To address this critical need， a short-term power load forecasting model in industries based on correlation analysis and convolutional neural network （CNN） is proposed. First， the k-means clustering algorithm is optimized， and the raw data of industry load and external influencing factors are clustered to improve the accuracy of the subsequent correlation analysis. Then， an improved correlation analysis method based on normalized mutual information （NMI） is proposed to quantitatively assess the correlation between various external influencing factors and industrial loads. Last， a load forecasting network considering the correlation of external influencing factors based on the CNN is designed， which can be used for the short-term power load forecasting of industrial sectors after training. The results of a controlled experiment demonstrate the superiority of the proposed model in terms of both accuracy and stability when it comes to short-term power load forecasting for industrial sectors.

Keywords

• load forecasting
• correlation analysis
• nmi
• k-means clustering
• cnn