结合自适应谐振理论和域理论的优点 ,针对回归估计问题的特性 ,提出了一种新型神经网络回归估计算法 FTART3.该算法学习速度快、归纳能力强 ,不仅具有增量学习能力 ,还克服了 BP类算法需要人为设置隐层神经元的缺陷 .直线、正弦、二维墨西哥草帽、三维墨西哥草帽等 4个实验表明 ,FTART3在函数近似效果和训练时间代价上都优于目前常用于回归估计问题的 BP类算法 Combining with the advantages of adaptive resonance theory and domain theory, aiming at the characteristics of regression estimation problem, a novel neural network regression estimation algorithm FTART3 is proposed, which has the advantages of fast learning, strong inductive ability, not only incremental learning ability but also overcome BP algorithm needs to set artificial neurons in hidden layer.4 experiments on linear, sinusoidal, two-dimensional Mexican straw hat and three-dimensional Mexican straw hat show that FTART3 is superior to the current approximate regression estimation function in function approximation effect and training time cost BP algorithm