论文部分内容阅读
[目的]本文旨在提出更有效的时间序列组合预测模型的构建方法,建立预测精度较高的时间序列组合预测模型。[方法]以1978—2013年新疆农业机械总动力为数据源,建立了源序列的曲线回归、自回归积分滑动平均、3次指数平滑和灰色模型,并构建了预测对象和预测模型的关系数据库。提出了基于百分误差的计算属性重要度方法,依据该方法计算单一模型在组合模型中的权重,构建了单一模型预测值及其权重为输入的组合预测模型,使输出结果中完整的涵盖了时间序列不同单一预测模型的输出值特征。以误差分布特征为指标,对组合预测模型和各单一模型的预测性能进行分析。以组合预测模型拟合优度和预测值平均绝对百分误差(MAPE)作为评价指标,对基于百分误差、粗糙集、Shapley和熵权法的组合预测模型构建方法进行定量分析。[结果]预测周期内提出的组合预测模型的最大及平均误差与各单一模型最优值相比,分别降低了27.35和6.43,误差平方和(SSE)减少了73%,平均绝对百分误差降低了1.56%。基于百分误差的组合预测模型的拟合优度与基于粗糙集、Shapley和熵权法的组合预测模型拟合优度相比,分别提高了2.40%、5.10%和2.27%,粗糙集、Shapley和熵权法的预测值的平均绝对百分误差分别为1.673 0、3.726 1和2.702 4,而本文提出的模型的平均绝对百分误差为1.298 4。[结论]基于百分误差的组合预测模型在农业机械总动力和类似时间序列预测分析中,降低预测误差波动幅度及提高预测精度方面与其他单一模型和组合模型相比具有显著优势。
[Objective] The purpose of this paper is to propose a more efficient time series combination forecasting model construction method, and to establish a forecasting model of time series with high prediction accuracy. [Method] With the total power of agricultural machinery in Xinjiang from 1978 to 2013 as the data source, the curve regression, autoregressive integral moving average, 3-exponential smoothing and gray model of source sequence were established and the relational database of prediction object and prediction model . A method of computing attribute importance based on percentage error is proposed. According to this method, the weight of single model in the combined model is calculated, and the combined forecasting model of single model predicted value and its weight is constructed, so that the output result completely covers Output Value Characteristics of Different Single Prediction Models in Time Series. Taking the error distribution as an index, the predictive performance of the combined forecasting model and each single model is analyzed. Based on the combined prediction model goodness-of-fit and the predicted average absolute percentage error (MAPE), the quantitative prediction method based on percentage error, rough set, Shapley and entropy method was constructed. [Results] The maximum and average errors of the combined forecasting model proposed in the prediction period were reduced by 27.35 and 6.43 respectively compared with the optimal values of each single model, the error square sum (SSE) was reduced by 73% and the average absolute percentage error was decreased 1.56%. Compared with the goodness of fit of the combined forecasting model based on the rough set, Shapley and entropy method, the goodness of fit of the combined forecasting model based on percentage error improved by 2.40%, 5.10% and 2.27%, respectively. The rough sets, Shapley And the average absolute percentage error of the forecast value of the entropy method are 1.673 0, 3.726 1 and 2.702 4 respectively, while the average absolute percentage error of the model proposed in this paper is 1.298 4. [Conclusion] The combined forecasting model based on percentage error has significant advantages compared with other single model and combined model in reducing the fluctuation range of forecast error and improving the prediction accuracy in the total power and similar time series forecasting of agricultural machinery.