Research On Adaptive Inference Modeling Method

Complex industrial process usually has the characteristics such as strong nonlinear, uncertainty and so on, so it is difficult to describe its dynamic characteristics by precise mathematical model. Fuzzy modeling of nonlinear systems and adaptive control imitate people's intelligent to effectively control complex uncertain systems, which has the ability to study from environment and adapt to the environment. Therefore in-depth study of nonlinear system fuzzy modeling and adaptive control has important theoretical significance and application value.Modeling based on fuzzy inference about control system calls fuzzy inference modeling. According to interpolation mechanism of fuzzy logic system, this method makes the mathematical model of control system as sectional interpolation function, thus it solves the difficulty of system modeling. Analysis of the one-dimensional and two-dimensional fuzzy inference model based on traditional triangular and Gaussian membership function, and from the interpolation point of view reveals that fuzzy inference model with different number of fuzzy partition and membership function has different truncation error characteristics. At the same time by analyzing the analytical structure of fuzzy controller using sub-domain method reveals the equivalence between fuzzy controller and PID controller, find the solution to select the membership functions which impact the controlling quality.For traditional fuzzy inference modeling methods, model parameters are selected based on experience, the model is also determined, which has the different dynamic tracking capability for different systems, so has bad generalization ability. Considering the limitations of conventional fuzzy inference modeling, here proposes a kind of Gaussian-type membership function, and proves that adaptive fuzzy inference system based on Gaussian-type membership can approximate nonlinear systems at arbitrary precision. Design of adaptive fuzzy inference system structure and parameters adjustment program, adopt gradient decent algorithm to study model parameters, and then combine with simulation experiments show its universal approximation. Finally, fuzzy inference model applies to nonlinear dynamic system identification and further verifies its feasibility.