基于ε-SVR的销量预测规划计算模型和算法研究

ε-SVR mathematical model and algorithm for sale prediction problems

设计了一个基于支持向量规划的香烟销量预测和销售趋势分析数学模型(LR_-εSVR和NLR_ε-SVR)及算法(-εSVR_SP)。企业的历史销售情况及企业外部的一些环境和条件作为算法的输入,输出未来一定时期的销量预测数据和销售趋势图。基于非线性核函数的学习算法降低了训练空间向量的维度,从而降低了计算复杂性,减少了对训练学习数据量的要求,提高了计算精度,降低了计算时间。通过模拟实验、实际数据集合实验及与神经网络算法的对比,验证了该算法的精确度和计算效率。

A mathematical model（LR_ε-SVR, NLR ε-SVR） and an algorithm （ ε- SVR_SP） were proposed on the basis of Support Vector Machine （SVM） regression, which could predict tobacco sale and sale tendency. Historical data and current context data were taken as the input, and such results as sale tendency in the future and the forecasting tobacco sales were outputted. By using kernel-based learning algorithm, the dimension of feature space in training process has been reduced, thus, the calculating complexity becomes lower and a small amount of training data will be sufficient. Simulated experiments, practical data set tests and comparative experiments with neural network show the advantages of the proposed approach in computation precision and efficiency.