Improvement of Maximum Power Demand Reduction Through Artificial Intelligence for Non-Linear Load

Koay, Kevin Chen Rong (2020) Improvement of Maximum Power Demand Reduction Through Artificial Intelligence for Non-Linear Load. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Kevin Koay Chen Rong.pdf Restricted to Registered users only Download (7MB)

Abstract

Maximum power demand always poses a challenge to industries that has a high power demand requirement. As the electricity is being generated and regulated by electric utilities, they will impose maximum power demand penalties to a company that did not manage well their electricity consumption. Therefore, naturally, the industry will search for a solution to reduce their penalties cost and improve their electrical bills. The uses of a non-linear load such as induction motor in the industry cannot be denied as they are used in various sectors such as manufacturing and production. But the challenges of using this type of machine is that they will draw in a huge amount of current, on top of that during the duration of energizing of an AC equipment will cause an inrush current, and these factors will be contributing to huge power consumption. Solutions were developed and introduced such as soft starter, variable frequency drive, etc. to solve this challenge and improve the maximum power demand but with the advent of industry 4.0, smarter and more intelligent system can be introduced that can act as a solution to the Maximum power demand problem. The implementing of artificial intelligence (AI) in the commercials and industrials sector will play a significant role in automation. During the simulation process, a PID controller is used as a reference model to compare with the AI controller. Linear regression supervised learning and DQN reinforcement learning are artificial intelligence that is developed in order to manage the non-linear load to improve the maximum demand condition. For supervised learning, its development is heavily supervised by the developer in order to ensure that the algorithm makes a proper prediction and providing correct analysis in the obtain data. As for reinforcement learning, its development will be the focus on the reward design model. The algorithm will develop its own set of policy in order to achieve the objective set by the developer. When comparing between supervised and reinforcement learning, reinforcement learning reduces the design process significantly, especially the time period while being able to achieve results that comparably better than the linear regression and PID controller. The implementation of the AI algorithm in hardware presented various challenge and limitation in hardware and software. For this project, the linear regression algorithm is only able to be implemented and perform as intended in its design in the hardware application while DQN algorithm is only able to perform in the simulation model. As the design methodology that has been proposed is implemented, and the project objective has achieved.