基于多维灰色模型与神经网络的销售预测模型研究

发布时间：2018-03-31 15:14

本文选题：销售预测　切入点：灰色关联度分析　出处：《浙江理工大学》2017年硕士论文

【摘要】：销售预测对企业至关重要,准确的销售预测可以帮助企业制定正确的营销策略以减小企业在经营过程的损失,提高企业盈利水平。但在销售领域,由于产品生命周期通常较短且产品销售情况受诸多因素影响,导致可用于预测的历史数据量少且波动大,极大了增加了准确预测的难度。本文针对线上销售的特点,分析可能影响产品销量的产品因素和外部环境因素,阐述比较了当前应用较广泛的EELM、ARIMA、神经网络和灰色理论系统中GM(1,1)和GM(1,N)等销售预测方法算法。针对GM(1,1)和GM(1,N)提出了改进的多维灰色模型算法(IGM(1,N))。IGM(1,N)算法可以使得销售序列不仅满足GM(1,N)对要预测序列要求的平滑和指数平滑条件,还将影响销量的相关因素纳入到预测考虑范围中,提高了预测的准确性。此外,本文还进一步将IGM(1,N)算法与神经网络算法相结合,提出混合智能算法。在利用混合算法对多组天猫销售数据进行预测时,首先对销售数据进行预处理,利用灰色关联度分析影响产品销售的多个因素对销量序列的关联度找出与销量关联度较大的因素并将其纳入预测过程。通过采用多种误差评定和预测曲线与实际销量曲线对比的方式,将混合智能算法预测结与EELM、ARIMA、GM(1,1)、IGM(1,N)等算法的预测结果进行比较,可以总结出混合智能算法对销售序列进行预测是可行的,其对销售序列销量预测的平均绝对百分误差MAPE误差始终保持在24%左右,预测准确性要优于EELM、ARIMA、GM(1,1)、IGM(1,N)算法,预测性能也更稳定。本文有如下的创新点:1)提出了IGM(1,N)算法,对系统特征序列加入控制因子,提高了原GM(1,N)模型的预测精度。2)IGM(1,N)对序列与处理过程中,加入附加因子,使得序列满足灰色预测方法对序列的要求条件。3)通过将IGM(1,N)算法和神经网络算法结合,从而提出了混合智能算法。该算法对销售序列的预测精度和表现要优于EELM、ARIMA、GM(1,1)、IGM(1,N)等算法。
[Abstract]:Sales forecast is very important to the enterprise. Accurate sales forecast can help the enterprise to formulate the correct marketing strategy to reduce the loss in the business process and improve the profit level of the enterprise. But in the sales field, Because the product life cycle is usually short and the sales situation is affected by many factors, the historical data quantity that can be used for prediction is small and fluctuating, which greatly increases the difficulty of accurate prediction. Analysis of product factors and external environmental factors that may affect product sales, This paper describes and compares the current widely used algorithms for sales prediction such as EELMN ARIMA, neural network and grey theory system, such as GM1) and GM1N). An improved Multidimensional Grey Model algorithm (IGM1N.IGM1NN) is proposed for GMM1) and GM1N). Only satisfying the smoothing and exponential smoothing conditions required by GM1N) for the prediction sequence, The related factors affecting the sales volume are also taken into account in the forecast, which improves the accuracy of the prediction. In addition, the IGM1N) algorithm is further combined with the neural network algorithm. A hybrid intelligent algorithm is proposed. When using the hybrid algorithm to predict the sales data of Tmall, the first step is to preprocess the sales data. By using grey correlation degree analysis, several factors affecting product sales are analyzed to find out the factors which have a large correlation with sales volume and to bring them into the forecasting process. The evaluation of multiple errors and the prediction curve and the actual situation are carried out through the use of various errors. The way the sales curve is contrasted, By comparing the prediction results of the hybrid intelligent algorithm with those of the EELMM-ARIMA GM1 / 1 GM1 / 1 IGM1N algorithm, it can be concluded that the hybrid intelligent algorithm is feasible to predict the sales sequence, and the results are as follows: (1) it is feasible to predict the sales sequence by using the hybrid intelligent algorithm (HIA). The average absolute percent error of the forecast of sales volume is always about 24%, and the accuracy of prediction is better than that of the EELMMAIMA / GM1 / 1 / 1IGM1N) algorithm, and the prediction performance is also more stable. This paper has the following innovations: 1) the IGM1N) algorithm is put forward in this paper. The control factor is added to the characteristic sequence of the system, which improves the prediction accuracy of the original GM1N) model. The additive factor is added to the sequence and processing process. By combining the IGM1N) algorithm with the neural network algorithm, a hybrid intelligent algorithm is proposed, which is superior to the EELMARIMAN (GM1N) and IGM (1N) algorithm in predicting the sales sequence.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;N941.5;TP183

【相似文献】