基于改进的GRNN的销量预测

根据灯具销量具有增长趋势和周期性二者合成的非线性特点,本文利用相空间重构理论确定GRNN的输入节点数,以加权误差代价函数代替传统误差代价函数为目标训练网络,采用粒子群优化算法优化平滑参数,并在此基础上将GRNN用于灯具销量的短期预测。与传统的BPNN相比,改进后的GRNN预测精度更高,抗噪声能力更强,模型更稳健。

Sales Prediction Based on Improved GRNN

According to the nonlinear characteristics caused by the increasing trend and the periodicity of lamps and lanterns sales sequence,this paper uses the phase space restructuring theory to decide the input neuron number of general regression neural networks,takes the weighted error cost function as the goal of network training to replace the traditional error cost function,uses the PSO algorithm to optimize the smooth parameter based on these improvements,and uses GRNN to predict short-term lamps and the lant...