Learning algorithm for the state feedback artificial neural network

摘要

Most research and application of recursive neural network is mostly the unit feedback recursive neural network, the dynamic process of the system is usually determined by the dynamic feedback, so it is difficult to control the dynamic process, which limits application. The recursive feedback factor implied the neural network dynamic performance, the different state feedback factor expressed the different dynamic characteristic, and therefore, the research dynamic characteristic and the learning strategy for state feedback neural network has extremely important theory significance and the application value. For this shortage, we proposed a kind of the state feedback dynamic evolved neuron model, as well as neural network which is composed by the state feedback neuron and learning algorithm. For this kind of neural network its static weight implies the neural network static behavior, the state feedback factor indicates the neural network dynamic behavior. Not only can the static weight be corrected through the learning from the static knowledge, but its dynamic state feedback factor also can be corrected through learning from the dynamic knowledge. Not only can it learn the static knowledge, but also the dynamic knowledge. Not only may it remember the static information, but also the dynamic information. It becomes truly dynamics characteristic neural network. Finally in this paper, we summarized the recursive neural network static weight and the dynamic recursive coefficient study algorithm by the theorem form.