http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
An Optimal Current Control of Interlink Converter Using an Explicit Model Predictive Control
Ismi Rosyiana Fitri,Jung-Su Kim 한국지능시스템학회 2018 INTERNATIONAL JOURNAL of FUZZY LOGIC and INTELLIGE Vol.18 No.4
An interlink converter (IC) presents in a hybrid microgrid allowing a bidirectional power exchange between AC and DC sub-grids. In this paper, a model predictive control (MPC) -based IC current controller is employed, taking the current constraint into account to ensure the safety of IC. To reduce online computation burden of the MPC, a tabular form of MPC named Explicit MPC is applied. The simulation result shows that the proposed IC controller successfully maintains the power supply-demand balance of the hybrid microgrid. Moreover, the proposed method keeps the power exchange value in the acceptable range, particularly under high loading condition
Computation of Feasible and Invariant Sets for Interpolation-based MPC
Ismi Rosyiana Fitri,Jung-Su Kim,Shuyou Yu,Young IL Lee 제어·로봇·시스템학회 2021 International Journal of Control, Automation, and Vol.19 No.10
The terminal invariant set plays a key role in the stabilizing MPC (Model Predictive Control) formulation. When control gains of the terminal local control laws and corresponding feasible and invariant sets are given, the existing interpolation methods unite them to enlarge the stabilizable region and enhance performance. In this paper, when an invariant set is given, an algorithm is proposed to find another invariant set such that their convex hull is maximized and also invariant. Numerical examples show that the set of the stabilizable initial state of the MPC is enlarged by the terminal constraint set computed by an interpolation-based approach.
한동기(Dong-Ki Han),Ismi Rosyiana Fitri,김정수(Jung-Su Kim) 제어로봇시스템학회 2018 제어·로봇·시스템학회 논문지 Vol.24 No.5
This paper presents a method of constructing an inverse model-based disturbance observer using two neural network methods: Multi-Layer Perceptron (MLP) and Recurrent Neural Network (RNN). Learning data is prepared for MLP and RNN by selecting input data such that it contains various signal shapes such as constant, step, sinusoidal and random number, and frequency components. This input data is injected into the nominal model of the system and the resulting state values are used as the measurement. The weights of MLP and RNN are optimized using these data, and how unknown disturbances are estimated is explained using the learned MLP and RNN. The simulation results show that the proposed method works well; in other words, the MLP- and RNN-based disturbance observer can reject both external disturbances and model uncertainties.