Performance Assessment And Application Of Multivariable Control System Based On Disturbance Restrain

Performance assessment is a young branch in control science, but it has received much attention from academia and industry over the20years of its development.In industrial production, most of the disturbances are random, so the performance of restraining random disturbance can be obtained from variance of the process output.For this reason, MVI and indexes based on it,like LQG, GMV index, become mainstream.The increasing complexity of industrial processes makes the actual system have more coupled loops, so the performance assessment of multivariable control system is also one of the hotspots of academia. The research topics include:1. This article presents a new approach to performance assessment based on initial closed-system and minimum variance index. When time delay of process becomes shorter, assessment based on minimum variance index may draw a wrong conclusion, while the new approach can avoid this disadvantage. By interactor matrix, generalized index can be valid for multivariable systems, and this article applies the approach to performance assessment of distillation columns process. Simulation result of distillation columns process shows the validity of this method, states that generalized index should be used for assessment when time delay of process becomes shorter.2. For the multivariable system, the calculation process of interactor matrix is complex. Xia and Huang’s research shows that the interactor matrix can be substituted by the matrix based on the time-delay information of the system to get the upper and lower limits of MV index. This article summarizes their algorithm and analyses the applicability of their algorithm. Calculating of interactor matrix essentially needs to know the Markov parameters of the system, and this means to know the complete information of the process model which is usually difficult to be achieved. But it is easier to obtain the time-delay information of the system which even can be obtain from daily operation data without test. Through simulation, it is found that the method is good when the order of interactor matrix is larger. 3. This article presents a method of LQG performance assessment based on subspace identification. Subspace methods to solving LQG problem, avoids solving algebraic Riccati equation. As the calculation is simple, it can be used for real-time computing, also can be applied to a large number of loops in industrial field. Although the method of subspace identification is used, but the specific model is not obtained, which means no truncation operation. So this method retains the original information as much as possible.4. This article presents a method of LQG controller design based on the performance benchmark function. At first get the function about weight matrix of LQG controller and variances of process input, output. When the variances of input, output are given, conversely solve the weight matrix by PSO and derivative-free method. At last complete controller design by weight matrix. The traditional LQG controller design is to artificially set weight matrix, trial and error. The method proposed in this paper is the first design for LQG controller when input and output variances are given.