Electrical power and energy: residential power demand forecasting using AI technology

• Main Author: Tan, Wei Rong
• Other Authors: Xu Yan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/181507
• Institution: Nanyang Technological University

In recent years, accurate forecasting of electric power consumption has become increasingly vital due to the growing demand for electricity and the need for efficient energy management. This project focuses on developing robust models for predicting residential power consumption in Tetouan, Morocco, leveraging historical consumption data and various weather-related features. By utilizing advanced neural network techniques, such as basic Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Convolutional Neural Networks (CNNs). This project aims to discover the most effective model for time series power forecasting application. The dataset, sourced from Kaggle's Electric Power Consumption repository, contains electric power consumption records sampled every ten minutes for three distinct zones in Tetouan. The features include temperature, humidity, wind speed, general diffuse flows, and diffuse flows, with the target variable being the total average power consumption across the three zones. The data underwent preprocessing steps, including normalization and feature extraction, to enhance the model training process. The project employed a comprehensive experimental framework, splitting the data into training, validation, and test sets to evaluate the models' performance. Each model's architecture was carefully designed and trained using a window size of 72 for sequence generation. The LSTM model was further enhanced with feature engineering (LSTM-FE), incorporating additional temporal features such as hour, which day of the week, the month, the day, the day of the year, day of the month, to capture complex patterns within the dataset. Key metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) were used to evaluate the model’s performance. The LSTM-FE model, which incorporated additional temporal features, provided the best forecasting accuracy with a MAPE result of 6.07% as compared to the other models. The CNN model achieved an average MAPE result of 8.51%. The LSTM model produced an average MAPE result of 9.69% and the basic RNN model had the highest error metrics with an average MAPE of 14.67%. These results indicate that the LSTM-FE model is the most accurate for forecasting residential power consumption, followed by the CNN model, the standard LSTM model, and the basic RNN model in accurately forecasting residential power consumption. The insights gained from this research can aid in optimizing energy management strategies, ensuring a reliable power supply, and supporting sustainable energy practices in Tetouan and similar urban settings. Future work may explore integrating other advanced neural network architectures and additional external factors to further refine the predictive models.