论文部分内容阅读
局部连接神经网络简化了网络结构,提升了网络收敛速度和减少了网络训练复杂度,可用于函数逼近和系统建模.为了采用直观的建模方式对实际系统网络拓扑逼近,对此文章提出一种新型的局部连接BP网络模型(local BP neural network,LBPNN).该模型的网络结构可以模拟任意前馈型网络拓扑结构,其网络模型中的连接权和神经元与被模拟的网络拓扑中的边和节点一一对应.传统带约束的非线性规划和智能优化算法,其参数辨识受限条件多和算法代价较大,同时提出了与LBPNN模型相应的一种新型的带约束的随机梯度下降法(constrained stochastic gradient descent,CSGD)对其权值参数进行训练.通过算例仿真验证了CSGD训练算法的有效性,稳定性和鲁棒性.
The local connection neural network simplifies the network structure, improves the network convergence speed and reduces the network training complexity, and can be used in function approximation and system modeling.In order to adopt an intuitive modeling approach to the actual system network topology, this article proposes a A new type of local BP neural network (LBPNN) is proposed.The network structure of this model can simulate an arbitrary feedforward network topology, in which the connection weights and neurons in the network model are similar to those in the simulated network topology Edge and node one by one.The traditional constrained nonlinear programming and intelligent optimization algorithms, the parameters of the identification of many constraints and the algorithm costly, while the LBPNN model corresponding to a new type of constrained stochastic gradient descent (CSGD) is used to train its weight parameters.The effectiveness, stability and robustness of the CSGD training algorithm are verified by simulation results.