The objective of this research is to develop intelligent vector control technology, intelligent wind power extraction and management strategy, and advanced simulation and testing mechanisms by using Adaptive Dynamic Programming neural control mechanisms. This will improve the efficiency and reliability of wind energy conversion systems and enhance wind power integration into the electric utility system for the 20% wind penetration vision of the United States in 2030.
INTELLECTUAL MERITS The intellectual merits of this research include development of: 1) Adaptive Dynamic Programming vector control technology to overcome the deficiency of conventional vector control technology, 2) intelligent wind power extraction and management techniques to improve wind power production efficiency and reliability, 3) intelligent control integration strategy under practical system constraints, 4) advanced simulation and testing mechanisms, and 5) a curriculum in Intelligent Sustainable Energy Systems to enhance student capability in multidisciplinary fields.
BROADER IMPACTS The broader impacts of this research are significant. The rapid development and increased complexities of sustainable energy systems make it increasingly urgent to develop intelligent and cyber-enabled technologies for research and education of sustainable energy system field. This research will enhance optimal and intelligent technology for future smart and sustainable energy systems and increase the participation of an EPSCOR state in advanced scientific research. The research should have direct or indirect impacts to the following mission areas of the United States: reduction in imported energy, reduction of energy-related emissions, improvements in energy efficiency, and a technological lead for the United States in advanced energy techniques.
The doubly-fed induction generator (DFIG) wind turbine is a variable speed wind turbine widely usedin the modern wind power industry. Presently, commercial DFIG wind turbines primarily usetechnology that was developed a decade ago. However, due to the rapid transition for managementand generation of wind power toward the smart grid, research is neededto enhance energy harvestingfrom the wind in more efficient and reliable ways due to the following reasons: 1) Recent studiesindicated that an existing power converter control technology used in a DFIG wind turbine has a competing control deficiency. 2) The present design of DFIG wind power control system does not support the needs for optimaland smart control of the system. 3) Recentresearch on DFIGcharacteristics showed that optimal and artificial intelligence technology is favorable for efficient DFIG control and integration. This project developedan optimal vector control technology based on adaptive dynamic programming (ADP) and artificial neural networks for a DFIG wind turbine, an advanced neural network training algorithm, and state-of-the-art simulation and hardware experiment facilities and mechanisms. This is the first-ever application of ADP to vector control of power electronic converters and induction generators. Compared to conventional vector control methods, the neural network control approach produces the fastest response time, lowest overshoot, and, in general, the closest to ideal performance. The specific outcomes and findings of the project include: 1) We have successfully completed the training and testing for both grid-side and rotor-side controllers for a DFIG wind turbine. 2) The neural network vector controller can track the reference d- and q-axis currents effectively even for testing trajectories and reference currents that are far away from the training data. 3) Compared to conventional vector control methods, the neural vector control approach produces the fastest response time, lowest overshoot, and, in general, the best performance. 4) In noisy, disturbance, and power converter switching environments, the neural network vector controller demonstrates strong capabilities in tracking the reference command while maintaining high power quality. 5) The neural network vector controller also showed excellent performance in more practical nested-loop control structure. 6) Under a fault in the grid system, the neural network controller for a DFIG wind turbine exhibits a strong short-circuit ride-through capability. 7) We integratedourneural network vector controller design with energy system principles and conventional control techniques. We found that this interdisciplinary neural network design approach can significantly enhance the performance and ensure the reliability and adaptive capability of neural network controllers under practical conditions and physical system constraints. We also found that it is very hard to train such a neural network, indicating that an advanced training strategy is needed. 8) We conducted research of nested-loop neural network controllers for electric machines. The results showed that the inner current-loop neural network controller gained the most benefit. The conventional outer-loop PI controller can still be used with an inner current-loop neural network controller to achieve good control performance. 9) We built a hardware experiment system and conducted a series of tests to examine the performance of the neural network vector controllers in challenging hardware experiment conditions. The experimental results were very encouraging. Even in highly noisy and distorted laboratory conditions, the neural network vector controller demonstrated excellent performance. Compared to the conventional controllers, the neural network controllers showed very fast tracking speed, low oscillations and harmonics, and strong stability and reliability. Beyond the proposed project scope, we conducted preliminary research in the following areas.i) Developed a new Forward Accumulation Through Time (FATT) algorithm and integrated FATT with Levenberg–Marquardt algorithm toimprove training of recurrent neural networks.ii) Investigated microgrid control using neural networks. We found that the neural network vector controller has the potential to overcome many challenges for smooth transition and control of a microgrid between grid-tied mode and islanding mode.iii) Investigated how to apply neural network vector control technology to high voltage dc transmission systems.iv) Conducted preliminary research of nested-loop neural network controllers for ac electric machines and drives. v) Investigated control of grid connected inverters with LC and LCL filtering schemes for single-phase applications. The performance of the neural networks was also very promising.vi) Investigated wind speed forecasting. vii) In controlling an electric machine in generator or motor mode, significant noise and vibration reduction was achieved by using aneural network controller, which will benefit energy systems and devices in a variety of aspects. A video from the project website (http://bama.ua.edu/~shli/Research/NSF_EPAS.html) shows that in controlling an ac electric machine in generator or motor mode, the machine runs much more quietly when using our neural network controllers, which means reduced oscillations, reduced harmonics, and faster response speed.