%0 Journal Article
%T 时间序列组合预测模型研究:以农业机械总动力为例
%A 吐尔逊？买买提
%A 丁为民
%A 谢建华
%J 南京农业大学学报
%D 2016
%R 10.7685/jnau.201510042
%X [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1.6730、3.7261和2.7024,而本文提出的模型的平均绝对百分误差为1.2984。[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。</br>[Objectives] The combination model have emerged as a possible solution to the challenges associated with time series forecasting. In order to test this hypothesis,advantages and disadvantages of different methods used for combination model development were discussed in the present study. The models were assessed on the basis of prediction accuracy so that more effective approach for combination model for prediction is provided. Keeping in view the assessments,the prediction model for time series with higher accuracy was established.[Methods] The performance of combination prediction model was related with accuracy and weights of single model. According to their prediction accuracy on statistical data from 1978 to 2013 of total power used by agricultural machinery in Xinjiang,the cubic curve model,autoregressive integrated moving average model,cubic exponential smoothing model and grey model were used as member of combination model. Later the results of prediction models were used to develop a relational data model. On the basis of fitting error of every single model on prediction total power data,a new method of computing for importance degree of single model was developed by using the percentage error of each model,then a practicable way for calculating single model’s weight was developed by using the importance degree of each model. In the light of that method,percentage error based combination prediction model was used to predict time series data,i.e. the agricultural machinery total power of Xinjiang was fitted on percentage error based combination model. The accuracy of the proposed combination model and every single model was evaluated on the basis of minimum error,maximum error,average error,sum of the squared errors (SSE),mean absolute percent error (MAPE) and goodness of fit. Various combination prediction models on time series especially on the total power were already developed by
%K 农业机械总动力
%K 时间序列
%K 预测
%K 组合模型&lt
%K /br&gt
%K agricultural machinery total power
%K time series
%K forecasting
%K combination model
%U http://nauxb.njau.edu.cn/oa/darticle.aspx?type=view&id=201604023