On Data-driven Control Algorithm Based On Unfalsified Control

In recent years, with the development of science and technology, the scale of enterprises is growing faster, and the production process is more complex.Then, it becomes more and more difficult to control the production process based on the accurate mathematical model of the traditional control method. Because of the difficulty of establishing an accurate mathematical model of the plant, data-driven control comes into being, which uses a large number of on-line and off-line data to achieve optimal control of the plant. The American scholar Safonov proposed a data-driven control method, in1995, which is developed for identifying control laws with on-line data only. When parameters and structure of the system change, the current data reflect these changes, so that the controller can adapt these changes through feedback effects. This thesis is about data-driven control algorithm based on unfalsified control. The main work of this thesis is as follows:First of all, several issues about data-driven control theory and the development of unfalsified control theory are summarized. Meanwhile, the basics of the unfalsified control theory are introduced, as well as its stability.Secondly, the problem of inverted pendulum PID control based on unfalsified control is researched in this paper. Inverted pendulum is a typical multivariable, nonlinear, strong coupling and naturally unstable system, while the PID controller is the most widely used one in industry. A relative unfalsified controller is designed, and the Matlab simulation software is used to do the research on its simulation. The results show that the unfalsified controller can control the swinging rod angle of the inverted pendulum very well, without steady-state error. Besides, the controller can switch very fast, and it has good suspended performance. But the basic unfalsified control algorithm has the defects that the candidate controller set can not be infinite set.Thirdly, on account of the defect of the unfalsified control algorithm, the unfalsified control algorithm based on gradient method is researched. But it also has some defects that it’s easy to fall into the poles and it astringes slowly. For these problems, a modified algorithm is proposed. Adding momentum term into the adaptation rule can solve the problems very well. The following simulation study shows that the system performance is improved by the modified algorithm. And the gain of momentum term affects the system convergence rate. Furthermore, the range of gain is derived. At last, the thesis is summarized and the open problems for further research are also discussed.