论文部分内容阅读
为了进一步减少管状双线性递归神经网络的计算复杂度,在管状双线性递归神经网络中采用了延时反向传播算法。延时反向传播算法使用了阶次微分,误差函数对权值微分进行后向计算。后向计算顺序降低了初始化要求,减弱了网络对初始化条件敏感性并降低了计算的复杂度。该网络采用了模块化设计,各个模块以并行的方式执行任务,改善了计算效率。基于管状双线性递归神经网络的结构与神经元的数学模型,提出了具体的延时反向传播算法实现方案。同时进行了仿真来评估滤波器在非线性系统辨识方面的性能。实验结果表明基于延时反向传播算法的管状双线性递归神经网络提供了相当好的性能。
In order to further reduce the computational complexity of tubular bilinear recurrent neural networks, delay backpropagation algorithm is adopted in the bilinear tubular recurrent neural network. The delay backpropagation algorithm uses order differentiation, and the error function performs a backward calculation of the weight differentiation. The backward calculation sequence reduces initialization requirements, reduces the sensitivity of the network to initialization conditions and reduces the computational complexity. The network uses a modular design, each module to perform tasks in parallel, improving the computational efficiency. Based on the structure and neuron mathematical model of tubular bilinear recurrent neural network, a specific implementation scheme of delay backpropagation algorithm is proposed. Simulations were performed at the same time to evaluate the performance of the filter in nonlinear system identification. The experimental results show that the tubular bilinear recurrent neural network based on the delayed backpropagation algorithm provides quite good performance.