Abstract: To deal with the weakness of the BP neural network in learning speed, an Deviation Units Recurrence Neural Network model is presented based on the Jordan and Elman neural network. The weight regulatingmethod is developed based on BP algorithm. Simulations on fault diagnosis were performed with this neural network model. Experimental results show that the converging speed of this network model is faster than that of the traditional BP network and this model has a good practicability. 

Key words: prediction of protein contact maps, artificial neural network, deviation units recurrence neural network, hydrophobicity, secondary structure