论文部分内容阅读
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.
Aiming at the characteristics of tax revenue forecasting, this paper proposes an improved BP algorithm with integrated conjugate gradient and adaptive variable step size. The tax revenue forecasting model is established by using improved BP algorithm. Compared with the results of traditional regression analysis, It proves that this algorithm has the advantages of fast convergence, high learning accuracy and effectively avoids the local minima of the conventional BP algorithm.