Bavly Hanna, Prof Guandong Xu, Dr Xianzhi Wang, Prof Jahangir Hossain
Missing values in electricity consumption data are a common problem faced by researchers and utility companies. Accurate predictions of missing values are crucial for various applications, such as load forecasting, energy management, and anomaly detection. There has been an increasing focus on developing and deploying smart grids to meet growing electrical demands in an effective and economical way while reducing glasshouse emissions.
RQ: How to better predict energy consumption missing values in a smart grid?
Objectives:
• Comparing various AI models for energy consumption missing values prediction.
• Using a hybrid of AI models to better predict energy consumption missing values.
We employ a diverse set of 30 distinct machine learning models and a hybrid approach incorporating three machine models to predict missing values in energy consumption. The hybrid approach, which combines the strengths of multiple models, outperforms individual models in terms of accuracy, robustness, and generalization capability. This underscores the potential benefits of leveraging ensemble techniques in addressing the complexities of energy consumption prediction of missing values. Our findings highlight the importance of model selection and ensemble design. Certain models within the ensemble exhibit complementary strengths, contributing significantly to the overall predictive power. This emphasizes the need for a thoughtful combination of models to maximize the predictive performance. The success of the hybrid approach showcases its potential for real-world applications where accurate energy consumption predictions are crucial for effective resource management and planning.