Treffer: Predicting the accuracy of power consumption of electrical appliances using CNN vs auto MI.

The objective of this research is to conduct a comparative analysis of Convolutional Neural Networks (CNNs) and AutoML approaches for predicting electricity usage by household appliances. (Standards Association of Australia 1998)The primary goal is to determine the most accurate model architecture for different forecasting horizons and assess the effectiveness of automated machine learning techniques. Smart home appliance-level energy consumption data spanning 8 months is collected, encompassing refrigerators, ovens, and air conditioning systems. Python, with TensorFlow/Keras for CNNs,(Bertoldi, Ricci, and Wajer 1999) is used alongside an AutoML platform. The CNN model consists of 7 layers with regularization, pooling, and dropout techniques. The dataset is split into training (60%), validation (20%), and test (20%) sets using a time series split, and AutoML is employed to automate the model selection and hyperparameter tuning process. Evaluation on the test data reveals the performance of each model. The CNN model demonstrates competitive results overall, capturing complex spatial dependencies. The AutoML approach, leveraging ensemble methods and hyperparameter optimization, showcases comparable or superior results, indicating its effectiveness in automating the model development process. (OECD 2013) The results suggest that CNNs are effective for capturing spatial dependencies, especially in predicting near-term appliance consumption. Simultaneously, AutoML approaches provide a powerful alternative, automating the model selection and hyperparameter tuning process. Depending on the forecasting horizon, either approach may be preferable.(British Standards Institute Staff 1994) [ABSTRACT FROM AUTHOR]