Real-Time Power Consumption Monitoring and Forecasting Using Regression Techniques and Machine Learning Algorithms

• Main Authors: Arce, Jose Mari M, Macabebe, Erees Queen B
• Format: text
• Published: Archīum Ateneo 2019
• Subjects: Electricity Consumption; Energy Monitoring System; Forecasting; Machine Learning; Electrical and Computer Engineering
• Online Access: https://archium.ateneo.edu/ecce-faculty-pubs/16; https://ieeexplore.ieee.org/abstract/document/8980380
• Institution: Ateneo De Manila University

The demand for electricity in the Philippines has been steadily increasing with about one-third of the share going to the residential sector. Thus, there is a need to introduce energy management tools for residences to allow households to take control of their electricity consumption. This work presents a system which provides information on the power consumption of a residence through energy monitoring and forecasting. The system was deployed in a residential unit with a solar PV array and the electricity consumption was monitored for 28 days using an online cloud-server database. Moreover, different regression techniques and machine learning algorithms, such as linear and polynomial regression, support vector regression (SVR) and Random Forest, were trained and implemented to identify the model that gives the best accuracy in predicting the total electricity consumption of the residence at the end of the month. Results show that the linear and polynomial regressions produced large errors due to the nonlinear trend of the consumption data, which is attributed to the generated energy of the solar PV array. The support vector regression algorithm generated models with low percent errors in predicting the end of the month electricity consumption. Moreover, the random forest regression accurately predicted the next-day electricity consumption at 0.58 % error. However, the models generated using Random Forest are not suitable for long-term prediction.