Artificial Neural Network Based Forecasting of Power Under Real Time Monitoring Environment

Abstract

While photovoltaic (PV) has grown in popularity as a viable alternative to traditional energy sources in recent decades, it still has to improve in some areas to become the preferred energy source. The degradation of output power owing to soiling is one of the areas that require attention as it results in inefficient operation. The performance of clean and soiled solar panels is investigated in this work. The prediction algorithm is designed to forecast the output power of a solar panel based on the input parameters. Forecasting PV power is essential as it reduces uncertainty and assists in developing an effective PV technology. This work utilized three phases of development and validation to achieve these goals. Firstly, a data acquisition system to collect solar panel parameter data is designed, built, data stored on the ThingsSentral™ cloud platform. Secondly, a prediction algorithm based on machine learning ANN is created to estimate the output power of a solar panel, and thirdly the algorithm is simulated and tested. Measurements from two solar panels were used to test and analyze the performance of the proposed technique. Clean solar panels have an RMSE of 1.328, while dusty solar panels have an RMSE of 1.272, indicating this system can reliably forecast based on the training and test datasets provided. Both solar panel conditions have an R2 of 0.999, indicating that the solar panel dataset used in this project precisely matches the ANN model and accuracy. In conclusion, the data acquisition system and prediction algorithm employed in this work successfully met the project objectives based on the results.