ANFIS rule base optimization algorithm for electroencephalograph signal characteristic classification

Abstract

The invention proposes an ANFIS rule base optimization algorithm for electroencephalograph signal characteristic classification, which relates to the classification algorithm of electroencephalograph signal characteristic parameters. Because of the obvious non-linear characteristics of the electroencephalograph signal, the nonlinear characteristic classification method is widely used in electroencephalograph signal classification. The adaptive neuro-fuzzy inference system (ANFIS) combines the advantages of a neural network and the fuzzy logic, which can not only performs effective learning to the samples, but can also perform good expression and extraction to the knowledge. When the ANFIS system performs the fuzzy logic, fuzzy rule bases need to be created, whose number is the product of the number of all parameter membership functions. At present, the number of the membership functions of each input parameter can be set only by experience, which quite often leads to poor training effect. The GA-ANFIS method used for the ANFIS system rule base optimization utilizes the genetic algorithm to properly set the membership functions of each input parameter, and at the same time can effectively control the number of the rule bases to ensure the system training and learning efficiency.

Description

technical field [0001] The invention relates to an optimization algorithm of an ANFIS rule base for feature classification of electroencephalogram signals. The invention relates to the technical field of classification algorithms of electroencephalographic signal feature parameters. Background technique [0002] Feature classification is the process of using pattern recognition to establish a mathematical model between feature parameters and corresponding feature states, and predicting the feature states of unknown samples through this model. To establish a good and robust pattern recognition model, it is necessary not only to optimize the selection of the characteristic parameters involved in the modeling, but also to make a reasonable selection of the samples involved in the modeling. An ideal modeling sample should include all state sets in the sample to be tested. For example, in the process of sleep stage detection based on EEG signals, an ideal sleep sample should inc...

AI Technical Summary

Problems solved by technology

At present, we can only rely on experience to set the number of membership functions for each input parameter, which often fails to achieve better training results.

Especially when there are many input parameters in the ANFIS system, it will lead to a sharp increase in the number of rule bases, which seriously affects the training and learning efficiency of the system. On the other hand, when there are many parameters, the setting of the membership function of each input parameter is only based on experience. The accuracy of classification results

[0005] To sum up, when ANFIS classifies feature parameters, the membership function setting of each input parameter and the establishment of the rule base lack reference standards, and setting only by experience often fails to achieve good training results

Images