Pattern Moving Based Modeling,Analysis And Control For A Class Of Complex Production Processes

In recent years,many scholars have deeply studied the pattern-moving-based description method of system dynamics for a class of complex production processes with statistical moving laws.Although some research results have been obtained for the class of production processes,there are many unresolved problems,e.g.system stability,the influence of class partition of working condition patterns on system performance and estimation of an initial control(prediction)model structure.In this paper,the pattern class variable is measured by the center of a working condition class,and the influence of partition of the working condition patterns on the stability and regulation performance is analyzed and built.At last,according to the characteristics of the dynamics description method based on pattern moving,a pattern-classification-based method of system modeling and control is put forward.The main research results are as follows:?In order to study the influence of class partition of working condition patterns on system stability,a class-partition feature is defined,and the relationship between the feature and system stability is studied.At first,from an initial control(prediction)model of the concerned system,a nonlinear state space model is derived and established in Euclidean space,and the class-partition feature of working condition patterns is defined based on a method of class partition.Then,the pattern-moving state space is defined,and system stability is defined and analyzed in the space.Thus,the relationship between system stability and the class-partition feature is built.At last,a state feedback controller is designed for systems with and systems without input delay.? In order to study the relationship between parameters of working condition clustering and system regulation performance,a method of maximum entropy clustering based on the particle swarm optimization is proposed,and the relationship between parameters of the clustering method and system regulation performance indexes is analyzed and built.At first,the method of maximum entropy clustering is proposed to resolve the problem that it is difficult to determine initial clustering centers and their number in the k-means or ISODATA algorithms.Furthermore,the probability distribution of its clustering results is closer to the real distribution,and enough system dynamic information is contained in the clustering results.Then,system regulation performance indexes,a dynamic regulation performance index and a product quality regulation performance index,are defined and extracted.At last,a new constructive classification neural network is proposed based on the covering algorithm,and it is used to establish the relationship between clustering parameters and system regulation performance indexes.?In order to resolve the problem that the structure of an initial control(prediction)model and model parameters are hardly determined in modeling of complex processes,from the characteristics of a pattern-moving-based system dynamics description method,a new method of system modeling based on pattern classification is proposed.At first,input and output data are respectively partitioned into classes,and input and output orders of the system model are obtained by analyzing the conditional entropy of output classes.Thus,the pattern describing the current running state of the system(system pattern for short)is constructed.Then,an output class of the current system pattern is regarded as its class label,and a one-step prediction model(classification model)is established by using the covering algorithm.Moreover,a control method of the system is given.Thus,the proposed modeling method transforms the problem of system modeling into the problem of pattern recognition.Finally,t he constructive classification neural network is used to classify the system pattern and to realize the prediction of system output.