介绍复合神经网络模型及特性，讨论了基于假逆矩阵的自学习算法，研究用现场可编程门电路（ＦＰＧＡ）实现复合神经网络的自学习过程，电路设计成半并行全数字式．实验结果表明其权值学习结果与计算机软件模拟结果一致，而其运算速度是软件模拟无法比拟的 The composite neural network model and its characteristics are introduced. The self - learning algorithm based on the false inverse matrix is ​​discussed. The self - learning process of the composite neural network is studied by field programmable gate circuit (FPGA). The circuit design is semi - parallel and all - digital. Experimental results show that the weighted learning results are consistent with the computer software simulation results, and its computing speed is unmatched by software simulation