Modeling For Multi-model Based On Immune Theory

Objects often have multiple variables, strong coupled, nonlinear and time-varying characteristics for complex industrial processes. It is difficult to accurately describe its characteristics by single model, model accuracy and generalization are also very poor and expensive. But because of its simple model structure and strong nonlinear approximation ability for complex industrial object modeling and control provides a feasible thinking. Modeling of Multi-model is considered to be treatment method for nonlinear system and technology.However, Modeling process of Multi-model model collection and the element model numbers and model related matching degree will directly influence the control of precision and performance. Disadvantages of multi-model controller is large amount of models, computing capacity. In order to ensure the control accuracy under the prerequisite of the optimization model collection, reduce the number of element model, can improve the calculation speed, meet industrial control requirement of real-time; Plus, because of actuator wear, and the change of external environment factors can be caused by the change of the object characteristics or the emergence of new condition, and get more recognition that the model system is hard to adaptive to reflect the non-linear characteristics of the object.In consideration of the actual system, at a certain time period, input tends to be qualified in a range, do not cover the entire input space. So as long as confirm some subspaces of a enter space, you can achieve the required accuracy, thereby saving time. Based on this, and combined with the immune system and the mechanism of antigen antibody recognition of the optimization of the excellent learning ability, this paper proposes a model based on immune theory modeling method. The method for the selection of model collection, optimization, update has practical significance, and this paper online identification method can reflect the adaptive nonlinear characteristics of the object.Focuses on complex nonlinear system, the paper propose a multi-model modeling method based on artificial immune theory. First, input space is divided with grid method. Input space domain is divided into multiple small input space domain. Second, under immune theory, antibody number of antibody library is adjusted dynamically by calculationing match index of antigen and antibody, with modeling datas as antigen and subspace as antibody, each antibody corresponding a local model of mutil-model, and taking method of the minimum squares to estimate linear parameter. Last, weight coefficient is getted by a method of membership function, and local models are fused by the way of weight fusion. This method has a simple, fast-learning, good real-time, and other characteristics, is suitable for online learning and structural adjustment of multiple-input systems.Modeling of multi-model based on immune theory for the model selection, the optimization, the set of update has practical significance, and this paper online identification method can reflect the adaptive nonlinear characteristics of the object.