The Improvement of LDR Based Solar Tracker’s Action using Machine Learning

Solar energy is a clean, renewable and sustainable energy source that has the potential of becoming one of the major constituents of the total global energy mix. The achievement of maximum possible photovoltaic (PV) energy conversion requires that the sun's direct beam directly illuminates the photovoltaic panel's surface as long as possible. This can possibly be achieved through solar tracking techniques. Light Dependent Resistor (LDR) based tracking systems face multiple predicaments such as temporary clouding, fog, man-made blockage and others that cause impaired tracking. The objective of this research is to develop an LDR (Light Dependent Resistor) based solar tracker that has the ability to make accurate decisions and ignore faulty sensor readings. To satisfy this objective, Machine Learning techniques are implemented to produce a tracker that has the ability to learn and improve its action from daily interaction with the environment. A Machine Learning based algorithm is developed and implemented in a MATLAB environment. The algorithm has the functionality of continuously testing the consistency of input LDR sensor values, hence making the decision of either approval of the values and usage for further learning or alternatively, discarding of the values and replacing it by a more logical value from the previously learned information. The developed system is tested using Root Mean Square Error (RMSE). The system has showed a lowered and converging error response in varying environmental conditions (Sunny, Semi-Cloudy and Cloudy), thus the proposed system has proved its effectiveness in tackling the temporary blockage problem, resulting in a cheap, robust and noise tolerant solar tracking system.