Research On Model Predictive Control Techniques For Power Systems

Safety, reliability, economy and flexibility of the modern power system operation depend on efficient, stable, high-performance control strategies which are objectives pursued by researchers. Model Predictive Control, as one of the best advanced control theories, is proved to have advantages in model inclusiveness, nonlinear processing, constraints handling, control performance, flexibility, etc. Difficulties and problems encountered while Model Predictive Control applied to the power system are studied in this dissertation, and solutions are proposed.The dissertation firstly introduces the development of Model Predictive Control and its research progress, describes the characteristics of modern power systems and its control requirements. Model Predictive Control research and applications in the power system are introduced and the problems encountered are also discussed. Then the work of this dissertation is introduced.For a class of Lipschitz-continuity systems, its model uncertainties may cause the state variables violate the given limits. For this, Robust Strategy based on Pontryagin difference sets is employed. Because Pontryagin difference sets of this strategy are often too conservative, a dynamical process for determining the Lipschitz constants in the prediction horizon is proposed. Smaller Lipschitz constants can be obtained by this method, which effectively reduces the constraint sets’ conservative, while the amount of calculation increased is very small. A single machine infinite bus system for simulation is tested and the results show that dynamic determination of Lipschitz constants behaviors well in adaptability for uncertainties of the system model. At the same time, it greatly decreases the unfeasible probability of the optimal control problem constrained, and increases the robustness of control.For the large-scale, distributed power systems’feature, as well as the weak independence, difficult processing for associated variables between the subsystem controllers to which existing Distributed Model Predictive Control methods are applied, a Distributed Model Predictive Control algorithm based on Chebyshev polynomial approximation for related variables is proposed. The method can not only shorten the process of system dynamic transition, but also improves the independence between the subsystems controllers. The design and interdependence of operation are simplified between subsystems. It increases maintainability of subsystem controllers.For the problems of heavy calculation burden and bad real-time capability of Model Predictive Control, in this dissertation, a strategy based on parallel light weight threads in GPU is proposed to solve the optimal control problem. Up to tens of thousands of light weight threads in GPU can speed up the solving process of model predictive control related to the large-scale matrix operations and large-scale linear equations repeatedly solving. The speedup can be of8or more. This makes Model Predictive Control can be applied to larger systems or faster processes.Power system controllers are usually combined with other advanced control strategies. In order to allow Model Predictive Control to function properly with advanced control strategies and to maintain good compatibility with the most widely used PI control, a MPC-PI algorithm is proposed. This algorithm has not only high performance of Model Predictive Control, but improves the compatibility with the other existing policy. The combination also improves the reliability of the controller.Validation and simulation for the methods and strategies of Model Predictive Control above is conducted in power system load frequency control. Taking into account the electricity network load frequency control model, so that the control can reflect the impact of the nodes in the network parameters on the system frequency, the PI control law as the basis of this model will be integrated into the model predictive control. IEEE14-bus system is chosen as the simulation object. On this basis, the proposed Distributed Model Predictive Control algorithm based on Chebyshev approximation applied in constructed three-interconnected system, and results show that this method can effectively achieves control goals in a good coordination. The response to factors such as the disturbance, impact load on the system, the controllers can quickly make it reach stability. At the same time, GPU-accelerated and non-accelerated calculation is also tested and results show that the GPU can effectively reduce the MPC optimization time.