A Naïve Bayes Classification Approach for Short-Term Forecast of a Photovoltaic System

  1. Lookup NU author(s)
  2. Yang Quek
  3. Dr Wai Lok Woo
  4. Dr Thillainathan Logenthiran
Author(s)Quek YT, Woo WL, Logenthiran T
Publication type Conference Proceedings (inc. Abstract)
Conference Name6th Annual International Conference on Sustainable Energy & Environmental Sciences (SEES)
Conference LocationSingapore, South East Asia
Year of Conference2017
Source Publication Date
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
In the recent years, small scale photovoltaic (PV) power generation has become one of the mainstream options for building owners who seek to have their own isolated renewable power systems. Most of the PV output forecasting methods are high in complexity, resource intensive and use input parameters that are not readily available for small-scale users. This paper proposes a short-term forecasting method that applies Naïve Bayes Classification (NBC) machine learning technique on easily available input parameters such as instantaneous power, outdoor temperature, panel temperature, on-site irradiance and time of the day to forecast the energy that will be harvested by the PV system in the next 15-minute period. The forecasted results are classified into five easily-comprehensible categories, namely, Very Low, Low, Medium, High, and Very High. Historical test data of an existing PV system located in Singapore is used to evaluate the accuracy of the NBC forecasting method and the comparison demonstrates that the proposed method is able to achieve a forecasting accuracy of over 68 percent. The 15-minute period is sufficient for users or smart energy management system to start up alternative power supplies such as a diesel backup generator or reduce load demand by switching off non-critical loads. This information obtained through the low cost implementation will also be very useful in energy monitoring and management system in a Smart Home, particularly, in a small localized area where weather is very volatile—as such, mid-term forecasting being unreliable.