Artificial intelligent integrated sun-tracking system with sun and cloud positions prediction

Huang, Dick Shen (2024) Artificial intelligent integrated sun-tracking system with sun and cloud positions prediction. Final Year Project, UTAR.

Preview

PDF Download (1769Kb) | Preview

Abstract

The purpose of this study is to develop an artificial intelligent (AI) -integrated sun-tracking system and cloud position prediction system. These systems are to be applied on concentrated photovoltaic (CPV) to improve the efficiency in solar power generation. In this study, YOLOv8 is chosen as object detection model to recognize and locate the sun position. After successfully tracked the sun’s coordinate, Q-learning will be applied to control the motor in order to follow the sun. In addition, YOLOv8 also be applied to recognize the position of clouds. The purpose of cloud tracking is to encounter the problem of lost tracking of sun when it is shaded by clouds during cloudy weather. YOLOv8 will first obtain the position of the cloud, then calculation will be made according to the cloud movement and speed to predict the cloud shading time. The cloud shading time will then be applied to calculate the predicted sun reappear position. After that, the CPV will turns toward and stand by at the predicted reappear sun position. This is important to shorten the response time after lost track of sun to increase the efficiency of power generated by CPV system. Furthermore, this project also using 180° fisheye lens for a wider view, so that the larger cloud image can be captured. In result, the YOLOv8 model trained have an accuracy of 0.69% on cloud detection, and 100% on sun detection. The Q-learning training result also shown that the agent is able to move towards the target in the end of 1,000,000 episodes. The fish-eye lens had improved the cloud detection by widening the field of view of the camera module. Furthermore, the solar irradiance results also proved the accuracy of the sun object detection model. While the sun position prediction result had shown the percentage error ranging from 11% to 25%, and 43% during rapid change of weather. Lastly, the implementation of artificial intelligence had improved the efficiency of concentrated photovoltaic system during cloudy day. For future improvement, wind speed sensor, real-time satellite forecast system can be implemented for higher accuracy in prediction.

Actions (login required)