Distributed Model Predictive Control of Load Frequency for Power Networks

In recent years, there has been an increase of interest in smart grid concept, to adapt the power grid to improve the reliability, efficiency and economics of the electricity production and distribution. One of the generator side problem in this is to meet the power requirement while not wasting unnecessary power, thus keeping the cost down, which must be done while the frequency is kept in a suitable range that will not damage any equipment connected to the power grid. It would theoretically be most logical to have a centralized controller that gathers the full networks data, calculates the control signals and adjusts the generators. However in practice this is not practical, mostly due to distance. The transmission of sensor data to the controller and the transmission of control signals to the generators would have to travel far, thus taking up to much time before the generators could act. This paper presents a distributed model predictive control based method to control the frequency of the power network. First, an augmented matrix model predictive controller is introduced and implemented on a two homogeneous subsystems network. Later the control method is changed to a state space model predictive controller and is then utilized on a four heterogeneous subsystems network. This controller implementation also includes state observers by Kalman filtering, constraints handler utilizing quadratic programming, and different connection topology setups to observe how the connectivity affects the outcome of the system. The effectiveness of the proposed distributed control method was compared against the corresponding centralized and decentralized controller implementation results. It is also compared to other control algorithms, specifically, an iterative gradient method, and a model predictive controller generated by the MATLAB MPC Toolbox. The results show that the usage of a distributed setup improves the outcome compared to the decentralized case, whilst keeping a more convenient setup than the centralized case. It it also shown that the level of connectivity for a chosen network topology matters for the outcome of the system, the results are improved when more connections exists