Solar energy especially the photovoltaic (PV) system is one of “heat” renewable energy and widely adopted because mature development of PV technology lowers the threshold of application of PV all around the world. In Hong Kong, with different policies’ support, promotion of utilities and government, more and more commercial buildings, government buildings and even village houses installed or planned to install different scales of solar PV panels for the sake of environmentally friendly as well as cost saving. With the increasing penetration of PV in the electrical distribution system, an accurate PV output prediction in term of both short and long term is important for power companies, developers, and owners. This paper is to develop a PV forecasting model based on local weather data. T...[ Read more ]

Solar energy especially the photovoltaic (PV) system is one of “heat” renewable energy and widely adopted because mature development of PV technology lowers the threshold of application of PV all around the world. In Hong Kong, with different policies’ support, promotion of utilities and government, more and more commercial buildings, government buildings and even village houses installed or planned to install different scales of solar PV panels for the sake of environmentally friendly as well as cost saving. With the increasing penetration of PV in the electrical distribution system, an accurate PV output prediction in term of both short and long term is important for power companies, developers, and owners. This paper is to develop a PV forecasting model based on local weather data. The model is unique and Hong Kong specific because no similar forecasting model is being developed wholly based on Hong Kong’s local conditions before. In view of the large amount of data processing work, machine learning including artificial neural network (ANN) and long-short term memory (LSTM) are adopted in which the model predicts the PV output by learning the historical PV and weather data. This report illustrates the design of the algorithms and their performance, showing that the model is practical with sound accuracy in PV output forecasting.