Artificial intelligent integrated into sun-tracking system to enhance the accuracy, reliability and long-term performance in solar energy harnessing
The solar energy collected by the sun-tracking system depends on the accuracy of the sun-tracking algorithm, which is different for each type of sun-tracking system. Furthermore, the solar image formed on the on-axis target can also be easily affected by gear backlash and wind load, and an absolute...
Saved in:
Main Author: | |
---|---|
Format: | Final Year Project / Dissertation / Thesis |
Published: |
2022
|
Subjects: | |
Online Access: | http://eprints.utar.edu.my/4947/1/PH_1906464_FYP_report_%2D_JUN_YOU_TAN.pdf http://eprints.utar.edu.my/4947/ |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | The solar energy collected by the sun-tracking system depends on the accuracy of the sun-tracking algorithm, which is different for each type of sun-tracking system. Furthermore, the solar image formed on the on-axis target can also be easily affected by gear backlash and wind load, and an absolute encoder is used to reduce the errors produced by the sun-tracking system. Therefore, a fully artificial intelligent (AI)-integrated sun-tracking algorithm is proposed and can be used in any type of sun-tracking systems such as concentrated photovoltaic (CPV), flat photovoltaic (PV) or heliostat systems. The proposed AI algorithm integrates two deep learning models which are object detection algorithm and reinforcement learning. YOLOv7 is chosen as the object detection algorithm to detect sun, while Q-learning is chosen as the algorithm for reinforcement learning to control the motors. A custom dataset of 300 images of sun and clouds is prepared for the training process of YOLOv7. The YOLOv7 is trained for 100 epochs at different batch sizes, where the batch size of 4 shows the highest mean average precision (mAP) of 0.768 and the lowest loss. A python script is created to integrate both YOLOv7 and Q-learning to execute the tasks simultaneously to adjust the position of the sun tracker based on the detected sun. Once the sun coordinate is obtained, Q-learning will determine the minimum steps taken for the centre point of camera to reach the midpoint of the Sun. The algorithm is also tested under different conditions such as during sunny day, sunset and also when the sun is partially blocked. It shows that the sun is detected under each condition and has a confidence level of over 90%. The Q-learning provides the minimum steps and the movement options of the agent to move from the centre point to the sun coordinate. It shows that the agent is successful in reaching its goal which is the coordinate of the Sun. The proposed AI algorithm also eliminates the use of encoders as the algorithm can constantly feedback the errors due to the misalignment of the sun tracker. Thus, the sun tracker is able to adjust and align itself with the sun at the central receiver. |
---|