提出一种自适应神经网络模型对可变比特率视频通信量进行非线性自适应预测 ,并采用基于递归最小方差的自适应学习及删剪算法对抽头延迟神经网络进行训练和结构优化 .仿真实验表明 ,该模型能够实现对复杂视频通信量序列的高精度预测 ,满足实时快速的预测要求 . An adaptive neural network model is proposed to adaptively predict the rate of variable bit rate video traffic, and the training and structure optimization of the tap delay neural network is performed by adaptive learning and pruning algorithm based on recursive minimum variance. It shows that this model can realize the high-precision prediction of complex video traffic sequence and meet the real-time and fast prediction requirements.