Kanerva的稀疏分布存储(SDM)模型存在一个编码过程把输入模式二值化,这易引起模式间不保序的问题,并且其学习规则使该模型不具有对函数的逼近能力。本文提出一种改进的SDM模型,输入输出空间是实值模式,不需任何编码,改进了学习规则,使该模型不仅具有模式识别的能力,且具有对函数的逼近能力。最后通过实验证实了该模型的有效性。 Kanerva’s sparsely distributed memory (SDM) model has a coding process to binarize the input mode, which can easily lead to the problem of non-conservative mode. And its learning rules make the model have no approximation function. In this paper, an improved SDM model is proposed. The input and output space is a real-valued mode. Without any coding, learning rules are improved. This model not only has the ability of pattern recognition, but also has the ability of approximating the function. Finally, the validity of this model is verified through experiments.