Optimization of a renewable energy supply system using hard and soft computational control methodologies : a hybrid approach
As humanity becomes increasingly aware of the urgent need to re-address the delicate balance of nature, and as the Earth moves closer towards the natural global warming maximum of its current interglaciation, the reluctance of its decision makers to use the Earth’s diminishing supplies of fossil fuels in the traditional way is becoming more evident. To meet some of these changing objectives, research is gaining momentum into developing new forms of energy supply systems that are globally accessible, globally sustainable, and whose contribution to the current warming trends of our planet is negligible. Many of these real world energy systems have operating regions that exhibit varying degrees of non-linearity. An example of this is the significant variations in the dynamic characteristics of a distributed, solar-concentrating, parabolic collector field within a pilot solar thermal power plant situated in the Tabernas Desert, Almería, Spain. Here a hybrid controller was implemented, using a gain-scheduled controller with feed¬forward, to control the more linear operating regimes, while the natural task-orientated strengths of a fuzzy PI incremental controller were utilized to control the highly nonlinear operating region of the plant, i.e. below 5 litres per second. Removing the step-orientated nature of the problem from the fuzzy controller allows Multi-Objective Genetic Algorithm (MOGA)-tuning to use a greater variety of objective functions, improving its chances of finding better quality non-dominated solutions in a shorter time span. Bi-directional dynamic bumpless transfer was added to effect smooth transfer between the controllers. A unique way of improving the MOGA-tuning of the fuzzy logic controller was also employed by optimizing the number of input membership functions and their initial positions using fuzzy data clustering and adaptive neuro-fuzzy inference system data training techniques, and also by optimizing the number of generations required for convergence. Finally enhancements to the visualization properties of the MOGA’s graphical user interface (GUI) relating to its trade-off graph of parallel coordinates were implemented. These included percentage objective trade-off information in tabular matrix and bar chart form, and the incorporation of an evolving conflict or trade-off sensitivity mechanism to improve the visualization of the trade-off graph. These improvements were carried out to give the decision maker a better understanding of the system’s characteristics, and in doing so, to enhance the chances of a successful outcome when deciding between non-dominated solutions or potential fuzzy controller inference systems.