Date of Award

11-2022

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering (MSEE)

Department

Electrical and Communication Engineering

First Advisor

Dr. Addy Wahyudie

Abstract

The main objective of this work is to develop a neural-network-based Reactive Control (RC) system for wave energy converters. The ability to maximize the power output of WEC while maintaining operation constraints, which can be physical or thermal, is crucial to the development of deployable control strategies. Having a control method that is robust, which means it handles uncertainty and noise very well, is one of the main performance criteria in evaluating the method. Therefore, this work starts by deriving an averaged WEC model to be simulated in MATLAB/Simulink. Additionally, the concepts of resistive loading control and reactive control (approximate conjugate control) are discussed. A solution to sea state estimation is developed and explained which poses a contribution the current WEC research. This novel technique uses recurrent neural networks (RNNs) with time-series data input to estimate the sea state in real-time. The technique fills the gap of estimating forces based on peak frequencies and also the problem of calculating sea states based on periodical averaged statistical analysis. To complete the methodology, an optimization technique using feed forward neural networks is improved to perform optimization that is proposed to optimize the power output with respect to the sea states. This is done by using the neural network as a cost function while using the physical limitations of the system as a constraint. The neural networks in this work are developed, trained and tested using MATLAB’s Deep Network Designer and Deep Learning Toolbox then imported as a Simulink block to complete the simulation. The results are evaluated for each of the section. First, initial logging of the performance metrics, such as mean power, is done prior to the addition of any neural networks. The accuracy and robustness of the sea state estimation RNN is then discussed. Finally, a comparison between traditional reactive Control optimized and reactive Control is conducted. To summarize the outcome, after experimenting with different datasets and architectures, the RNN is able to estimate sea states in real-time under different initial conditions.

Arabic Abstract

اﻟﮭﺪف اﻟﺮﺋﯿﺴﻲ ﻣﻦ ھﺬا اﻟﻌﻤﻞ ھﻮ ﺗﻄﻮﯾﺮ ﻧﻈﺎم اﻟﺘﺤﻜﻢ اﻟﺘﻔﺎﻋﻠﻲ اﻟﻘﺎﺋﻢ ﻋﻠﻰ اﻟﺸﺒﻜﺔ اﻟﻌﺼﺒﯿﺔ ﻟﻤﺤﻮﻻت طﺎﻗﺔ اﻷﻣﻮاج. ﺗﻌﺪ اﻟﻘﺪرة ﻋﻠﻰ ﺗﻌﻈﯿﻢ إﻧﺘﺎج اﻟﻄﺎﻗﺔ ﻣﻦ (WEC) ﻣﻊ اﻟﺤﻔﺎظ ﻋﻠﻰ ﻗﯿﻮد اﻟﺘﺸﻐﯿﻞ، واﻟﺘﻲ ﯾﻤﻜﻦ أن ﺗﻜﻮن ﻓﯿﺰﯾﺎﺋﯿﺔ أو ﺣﺮارﯾﺔ، أﻣراً ﺑﺎﻟﻎ اﻷھﻤﯿﺔ ﻟﺘﻄﻮﯾﺮ اﺳﺘﺮاﺗﯿﺠﯿﺎت اﻟﺘﺤﻜﻢ اﻟﻘﺎﺑﻠﺔ ﻟﻠﻨﺸﺮ. ﯾﻌﺪ وﺟﻮد طﺮﯾﻘﺔ ﺗﺤﻜﻢ ﻗﻮﯾﺔ، ﻣﻤﺎ ﯾﻌﻨﻲ أﻧﮭﺎ ﺗﺘﻌﺎﻣﻞ ﻣﻊ ﻋﺪم اﻟﯿﻘﯿﻦ واﻟﻀﻮﺿﺎء ﺟﯿﺪًا، أﺣﺪ ﻣﻌﺎﯾﯿﺮ اﻷداء اﻟﺮﺋﯿﺴﯿﺔ ﻓﻲ ﺗﻘﯿﯿﻢ اﻟﻄﺮﯾﻘﺔ. ﻟﺬﻟﻚ، ﯾﺒﺪأ ھﺬا اﻟﻌﻤﻞ ﺑﺎﺷﺘﻘﺎق ﻧﻤﻮذج (WEC) ﻣﺘﻮﺳﻂ ﻟﯿﺘﻢ ﻣﺤﺎﻛﺎﺗﮫ ﻓﻲ ﺳﯿﻤﯿﻮﻟﯿﻨﻚ وﻣﺎﺗﻼب ﺑﺎﻹﺿﺎﻓﺔ إﻟﻰ ذﻟﻚ، ﺗﻤﺖ ﻣﻨﺎﻗﺸﺔ ﻣﻔﺎھﯿﻢ اﻟﺘﺤﻜﻢ ﻓﻲ اﻟﺘﺤﻤﯿﻞ اﻟﻤﻘﺎوم واﻟﺘﺤﻜﻢ اﻟﺘﻔﺎﻋﻠﻲ (اﻟﺘﺤﻜﻢ اﻟﻤﺘﻘﺎرن اﻟﺘﻘﺮﯾﺒﻲ). ﺗﻢ ﺗﻄﻮﯾﺮ وﺷﺮح ﺣﻞ ﻟﺘﻘﺪﯾﺮ ﺣﺎﻟﺔ اﻟﺒﺤﺮ ﻣﻤﺎ ﯾﺸﻜﻞ ﻣﺴﺎھﻤﺔ ﻓﻲ أﺑﺤﺎث (WEC) الحالية. ﺗﺴﺘﺨﺪم ھﺬه اﻟﺘﻘﻨﯿﺔ اﻟﺠﺪﯾﺪة اﻟﺸﺒﻜﺎت اﻟﻌﺼﺒﯿﺔ اﻟﻤﺘﻜﺮرة ﻣﻊ إدﺧﺎل ﺑﯿﺎﻧﺎت اﻟﺴﻼﺳﻞ اﻟﺰﻣﻨﯿﺔ ﻟﺘﻘﺪﯾﺮ اﻟﺒﺤﺮ ﻓﻲ اﻟﻮﻗﺖ اﻟﻔﻌﻠﻲ. ﺗﻤﻸ ھﺬه اﻟﺘﻘﻨﯿﺔ ﻓﺠﻮة ﺗﻘﺪﯾﺮ اﻟﻘﻮى ﺑﻨﺎءً ﻋﻠﻰ ﺗﺮدد ﺣﺎﻻت اﻟﺬروة وأﯾًﻀﺎ ﻣﺸﻜﻠﺔ ﺣﺴﺎب ﺣﺎﻻت اﻟﺒﺤﺮ ﺑﻨﺎءً ﻋﻠﻰ اﻟﺘﺤﻠﯿﻞ اﻹﺣﺼﺎﺋﻲ اﻟﺪوري اﻟﻤﺘﻮﺳﻂ. ﻹﻛﻤﺎل اﻟﻤﻨﮭﺠﯿﺔ ﺗﻢ ﺗﺤﺴﯿﻦ ﺗﻘﻨﯿﺔ اﻟﺘﺤﺴﯿﻦ ﺑﺎﺳﺘﺨﺪام اﻟﺸﺒﻜﺎت اﻟﻌﺼﺒﯿﺔ (Feed Forward) ﻷداء اﻟﺘﺤﺴﯿﻦ اﻟﻤﻘﺘﺮح ﻟﺘﺤﺴﯿﻦ ﺧﺮج اﻟﻄﺎﻗﺔ ﻓﯿﻤﺎ ﯾﺘﻌﻠﻖ ﺑﺤﺎﻻت اﻟﺒﺤﺮ. ﯾﺘﻢ ذﻟﻚ ﺑﺎﺳﺘﺨﺪام اﻟﺸﺒﻜﺔ اﻟﻌﺼﺒﯿﺔ ﻛﺪاﻟﺔ ﺗﻜﻠﻔﺔ أﺛﻨﺎء اﺳﺘﺨﺪام اﻟﻘﯿﻮد اﻟﻤﺎدﯾﺔ ﻟﻠﻨﻈﺎم ﻛﻘﯿﺪ. ﺗﻢ ﺗﻄﻮﯾﺮ اﻟﺸﺒﻜﺎت اﻟﻌﺼﺒﯿﺔ ﻓﻲ ھﺬا اﻟﻌﻤﻞ وﺗﺪرﯾﺒﮭﺎ واﺧﺘﺒﺎرھﺎ ﺑﺎﺳﺘﺨﺪام ﻣﺼﻤﻢ اﻟﺸﺒﻜﺎت اﻟﻌﺼﺒﯿﺔ اﻟﻌﻤﯿﻘﺔ ﺛﻢ اﺳﺘﺨﺮاﺟﮭﺎ ﻛﻜﺘﻠﺔ (Simulink) ﻹﻛﻤﺎل اﻟﻤﺤﺎﻛﺎة. ﯾﺘﻢ ﺗﻘﯿﯿﻢ اﻟﻨﺘﺎﺋﺞ ﻟﻜﻞ ﻗﺴﻢ. أوﻻً، ﯾﺘﻢ اﻟﺘﺴﺠﯿﻞ اﻷوﻟﻲ ﻟﻤﻘﺎﯾﯿﺲ اﻷداء ﻣﺜﻞ ﻣﺘﻮﺳﻂ اﻟﻘﻮة، ﻗﺒﻞ إﺿﺎﻓﺔ أي ﺷﺒﻜﺎت ﻋﺼﺒﯿﺔ. ﺛﻢ ﺗﺘﻢ ﻣﻨﺎﻗﺸﺔ دﻗﺔ وﻣﺘﺎﻧﺔ ﺗﻘﺪﯾﺮ ﺣﺎﻟﺔ اﻟﺒﺤﺮ ﺑﺎﻟﺸﺒﻜﺔ اﻟﻌﺼﺒﯿﺔ اﻟﻤﺘﻜﺮرة. أﺧﯿﺮا، ﺗﻢ إﺟﺮاء ﻣﻘﺎرﻧﺔ ﺑﯿﻦ اﻟﺘﺤﻜﻢ اﻟﺘﻔﺎﻋﻠﻲ اﻟﺘﻘﻠﯿﺪي اﻟﻤﺤﺴﻦ ﻟﻠﺘﺤﻜﻢ اﻟﺘﻔﺎﻋﻠﻲ. ﻟﺘﻠﺨﯿﺺ اﻟﻨﺘﯿﺠﺔ، ﺑﻌﺪ ﺗﺠﺮﺑﺔ ﻣﺠﻤﻮﻋﺎت اﻟﺒﯿﺎﻧﺎت واﻟﺒﻨﻰ اﻟﻤﺨﺘﻠﻔﺔ، ﺗﺴﺘﻄﯿﻊ ﺑﺎﻟﺸﺒﻜﺔ اﻟﻌﺼﺒﯿﺔ اﻟﻤﺘﻜﺮرة ﺗﻘﺪﯾﺮ اﻟﺤﺎﻻت اﻟﺒﺤﺮﯾﺔ ﻓﻲ اﻟﻮﻗﺖ اﻟﻔﻌﻠﻲ ﻓﻲ ظﻞ ظﺮوف أوﻟﯿﺔ ﻣﺨﺘﻠﻔﺔ.

Download

Included in

Engineering Commons

COinS