【摘要】 针对税收收入预测的特点,提出了一种综合共轭梯度和自适应变步长的改进BP算法,并利用改进的BP算法建立了税收收入预测模型,通过与传统回归分析预测方法结果的比较,证明了该算法收敛速度快,学习精度高,而且有效避免了常规BP算法得局部极小值问题.更多还原

【Abstract】 In allusion to the characteristic of the tax revenue predicting, an optimal BP algorithm combined with conjugate gradient and self-adaptive variable step is proposed in this paper, and a predicting model is built. Compared with the traditional method - multiple linear regression, it proves that the algorithm can not only converge quickly and learn accurately, but also effectively overcome the local optimal solution which often occurs in general standard BP algorithm.更多还原