Abstract:
The research investigates how to develop novel neural network vector control technology for Electric Power and Energy System Applications including grid-connected converters (GCC) and Electric Machines to overcome the drawback of conventional vector control methods and to improve the efficiency, reliability, stability, and power quality of electromechanical energy systems. The proposed neural network vector control was developed based on adaptive dynamic programming (ADP) principles to implement the optimal control. The new control approach utilizes mathematical optimal control theory and artificial intelligence, which is a new interdisciplinary research field. An examination of optimal control of a grid-connected converter (GCC) based on heuristic dynamic programming (HDP), which is a basic class of adaptive critic designs (ACDs), was conducted in this dissertation. The difficulty of training recurrent neural networks (RNNs) inspired the development of a novel training algorithm, that is, Levenberg-Marquardt ( LM) + Forward Accumulation Through Time (FATT). With the success of the new training algorithm, the difficulty of training a recurrent neural network has been solved to a large extent. The detailed neural network vector control structures were developed for different applications in power systems including three-phase LCL based grid-connected converters, single phase grid-connected converters with different filters, and in machine drive applications such as three phase squirrel-cage induction motors and doubly fed induction generators (DFIGs). Each of theseapplications has its own emphasis and features, e.g. , the resonance phenomenon associated with LCL filter, the rotor position estimation of induction motor and so on. Both simulations and hardware experiments demonstrated that the proposed ADP-based neural network control technologies produce superior performance to conventional vector control technology and approximates optimal control. Among all the advantages, one of most outstanding features of neural network control is that it can tolerate a wide range of system parameter changes, which is strongly needed in real applications. The proposed technologies provide the prospect to overcome the deficiencies of standard vector control technology and offers high performance control solutions for broad application areas in electric power and energy systems.
