Development of Embedded Optimal Control Platform for Efficient Energy Consumption and Growth in Fish Tank

Abstract

The fish farming industry has been receiving considerable attention and experiencing tremendous growth in the last decades around the world. Fish farming has the potential to overcome the requirement for the food because the world population is increasing day by day. Fish farm is one of the widely suggested fields for investing money because it supports a year-round production of fresh fish resources around 40% increased production rate in comparison to natural fish hunting. Ensuring healthy fish production involves elaborate monitoring and stable controlling of the fish farm; however, management of the resources inside fish tanks is a challenging task. It requires continuous monitoring and control, so energy consumption and labor cost are the central portions of the expenses. Using new advancement of technologies can support fish production improvement, cost reduction and automation of process. The Internet of Things (IoT) is one of the fast-growing technological areas which are influencing our daily life. Devices in our digital world are being furnished with various types of new types of technologies including microcontrollers, raspberry pies, sensors, transceivers, actuators, digital connectors, and Internet protocols. These technological advancements can give a wide range of opportunities for the development of interconnection among various devices and their users. The Internet of Things can support new types of services for companies, individuals, businesses, and governments with connecting various application scenarios to gather a number of parameters from the real world and using these data for future decisions and assessments by analyzing this data. The central key aspect of the embedded devices is a lightweight connectivity development for IoT devices and the development of fault tolerance interaction among them. These devices consist of light control actuators, video detector cameras, home automation tools, smart and autonomous vehicles, smart healthcare toolkits, intelligent actuators, and just to name a few. The installation of these smart devices can give various services to users, such as sensors collect the actual environmental information, and based on these gathered real data actuators are used to increase or decrease the environmental parameters via local network or Internet connectivity. Currently, IoT based applications are being widely utilized in various domains, including healthcare, self-driving transportation, aquaculture, agriculture, industrial and home automation, power management, traffic controlling, aerospace engineering, and numerous other fields. Machine Learning(ML) allows electronic technologies to learn autonomously from historical data and to utilize this knowledge to make predictions, decisions, and assessments independently. These types of applications are highly compute-intensive. As a result, these applications are conventionally executed on local servers, cloud servers, and personal computers. A new type of powerful embedded processors and advancements in algorithms, now machine learning algorithms can be performed directly on devices in the field Embedded Machine Learning. Embedded Machine Learning based applications can accomplish a number of achievements in the century of Industry 4.0. For instance, IoT sensor devices which measure optical or acoustic discrepancies and inconsistencies, then directly activate quality assurance functionality to the production or system state observation. Moreover, during the activation of cameras and microphones, these devices automatically monitor visual parameters and minimize soundwave errors based on contact, vibration, voltage, speed, temperature, and pressure sensor parameters. Then, these collected parameters also can be used for future improvement of the products. Embedded Machine Learning Algorithms has attracted many researchers to seek solutions for complex real-world problems. The highest percentage of the existing literature is paid attention to develop and run applications on a PC, local, or cloud server. However, these methods are not able to bring expected income to users. In this work, we attempt to take Embedded ML and IoT applications to the autonomous system development with deploying the to the Fish Tank, which is one of the most fast-growing industries. In this thesis, we propose an embedded optimal control platform based on ML and Optimization algorithms for efficient energy consumption and fish growth in Smart Fish Tank. We have developed the proposed embedded optimal control platform that integrates context-awareness, prediction, optimization, and control functionalities for controlling environmental parameters optimally in Fishtank. We have installed various IoT sensors and actuators to the fish tank and develop the context-awareness unit for the collection of sensing values from the real-fish tank environment. So the indoor environment data used in this work is real data which is collected from the fish tank during the three months. External environmental data collected from the physical fish tank. We have used the RNN-LSTM algorithm for prediction, the mathematical formulation for optimization, and the fuzzy logic controller for actuator control. A novel objective function for optimization is formulated and implemented for compute the optimal environmental parameters according to the predicted and user-desired environmental parameters data. In addition, we implemented the platform by considering various cases, firstly, we implement the platform based on actual fish tank environmental parameters without prediction model. Secondly, we have implemented prediction, optimization and control module using fish tank sensing data. Thirdly, we used outdoor environmental parameters to the prediction module. Fourthly, we consider the actuators control parameters in order to optimize energy consumption. Lastly, we implement the proposed platform by considering the power policy data. Besides, the control of the environmental parameters is tested with and without optimization schemes. Performance evaluation results prove that the optimization module with predicted values is 18% and 28.5 % effective in terms of environmental parameter optimization and energy consumption minimization compared to without optimization scheme. Furthermore, the proposed prediction-optimization based environment control energy consumption is 27%, 23.6%, and 11.8% effective in energy consumption compared with without prediction-without optimization, with prediction-without optimization, without prediction, and with optimization results. Also it spends 918 krw, 753 krw, and 423 krw less money for compared to other schemes.