Improved Fuzzy BP network in the ECG automatic identification and rule extraction

Electrocardiogram(ECG) is one of the necessary instruments used for clinic heart disease diagnosing. Researching and developing ECG auto- recognizing system can avoid most defects brought by reading ECG manually. It can also improve the accurate ratio of heart disease diagnosing. This system has important social significance and applied worthiness.In view of high dimension of ECG feature vector and complicated distribution of ECG patterns in feature space,we use fuzzy neural network to realize ECG pattern classification and recognition. Some alteration were made on structure and training algorithm for traditional serial fuzzy BP neural network. After alteration the structure became laconic,both training rate and correct recognition ratio were heightened. It became easier to extract rules from trained network basing on the characters of structure and knowledge distribution in the network. In addition we discuss the influence upon training efficiency and performance of this network generated by overlapping state of membership functions of fuzzy subsets. We bring forward modifying parameters of membership functions during samples learning and training. The problem on selecting and optimizing parameters was resolved effectively. .