Machine Learning for Solar Power Forecasting

Abstract

The motivation for large scale renewable energy plants in the recent years has come along with the challenges and question of stability. Preferably, the outcome of solar panels and wind are highly intermittent, that it is difficult to rely too much on this system as a stand-alone one. When it comes to solar energy, the future of solar energy is not merely dependant on the instantaneous output. It needs to be well ascertained so that the system stability and optimum power output can be obtained. The availability of voluminous weather data and recent advancements in the computational power has enabled tremendous growth in machine learning algorithms to predict future of solar power. In the proposed paper, collection of datasets giving the information on average temperature, surface pressure, wind speed and humidity data are done. The fetched data is then used to train the model using machine learning to develop an Artificial Neural Network (ANN) Model. The data from the neural network is fed into the PV array and the output is fetched to the MPPT system so that the system sustains at maximum power, even if the load changes. Forecasting of the input parameters is then done using time series forecasting with the help of Long-Short Term Memory (LSTM). Thus, the proposed work involves data pre-processing, feature selection and training the dataset and optimization of the output data so that optimal forecasting of solar power can be achieved.