Multiobjective formulation for synthesis of integral proportional observer
Proportional-Integral State Observer. Differential Evolution Algorithm. State feedback Control.
This work presents a synthesis formulation of Proportional-Integral (PI) state observers for multivariable linear time-invariant systems (LTI). Considering a bi-objective formulation, the purpose of the project is to minimize the state estimation errors, when the system is subjected to non-measurable disturbances, and the attenuation of the measurement noise effect. The performance specifications of the project are quantified in terms of the and norms of the closed-loop system matrices. The main contribution of this work is characterized by the formulation of the synthesis of PI observers as a multiobjective optimization problem that allows, through any multiobjective optimization algorithm, the generation of a set of efficient solutions with different trade-off between minimization of the state estimation error in the presence of disturbances and minimization of the effect from measurement noises. This technique aims at guaranteeing the null estimation error on stationary state for LTI systems in continuous time. A second contribution of this work is to compare the PI observer with Luenberger's classic observer to verify if, despite the greater complexity of the PI observer design and implementations, the benefits justify its use. To solve the problem, a Multiobjective Differential Evolution algorithm is applied. The set of solutions found in the multiobjective optimization problem form the approximate Pareto Frontier, with the optimized variables being the elements that compose the observer's gain matrices. A multicriteria decision-making method based on a fuzzy approach was applied providing three efficient solutions to the decision maker. The efficiency of the proposed sinthesys is verified by illustrative examples for different system orders. The results presented indicate the efficiency of the developed formulation, considering that through the PI observer it is possible to obtain smaller estimation errors with smaller gains than the classical observer, and still minimize the effect of disturbance and measurement noise.