ترکیبی از تجزیه و تحلیل مولفه های مستقل و رو به رشد نقشه های خود سازماندهی سلسله مراتبی با رگرسیون بردار پشتیبانی در پیش بینی تقاضای محصول

In the evaluation of supply chain process improvements, the question of how to predict product demand quantity and prepare material flows in order to reduce cycle time has emerged as an important issue, especially in the 3C (computer, communication, and consumer electronic) market. This paper constructs a predicting model to deal with the product demand forecast problem with the aid of a growing hierarchical self-organizing maps and independent component analysis. Independent component analysis method is used to detect and remove the noise of data and further improve the performance of predicting model, then growing hierarchical self-organizing maps is used to classify the data, and after the classification, support vector regression is applied to construct the product demand forecasting model. In the experimental results, the model proposed in this paper can be successfully applied in the forecasting problem.

Companies excelling in customer’s demand know the importance of effective supply chain planning. Successful businesses need an accurate picture of demand to drive production, inventory, distribution, and buying plans across their operations. These challenges are intensified by the effects of seasonality, promotions, and product proliferation, not to mention growth through mergers and acquisitions. Customer demand forecasting can be implemented to manage a global demand plan with multiple channels to market. An accurate forecast, which, in turn, drives more responsive customer service with lower inventories and reduced obsolescence, especially in the 3C(computer, communication, and consumer electronic) market. According to the statistics provided by Taiwan’s Institute for Information Industry, the value of 3C related products was 1.3 trillion NT dollars in 2008 and 1.1 trillion NT dollars in 2009. For this reason, the 3C market plays an important role in Taiwan’s semiconductor manufacturing and, as a result, accurate forecast of product demand has become increasingly important. Accordingly, many researchers have begun to pay more attention to the integration of product demand into supply chain management models. Reiner and Fichtinger (2009) has developed an extended dynamic demand forecast and inventory model for a two stage supply chain process, assuming purchase decisions are made by rational actors. In addition, Dolguia and Pashkevich (2008) has proposed a new demand model for multiple slow-moving inventory items with short request histories and unequal demand variance. They both note the importance of demand forecast in the supply chain management model. With the recent development of artificial intelligence models, several methods have been found to work more effectively than traditional methods when applied to forecast models. However, many researchers (Chang et al., 2006 and Wang et al., 2009) have mentioned that no matter what kind of data, some noise may influence the forecast result a lot. It seems data preprocessing to be more and more important. An integrated demand forecast model will be developed in this research, in which noise detecting and removing task will be considered first and then all data will be clustered to increase the accuracy and the practicability of the model. The detailed methodologies introduction and literatures review applied in this research can be found in Section 2. The framework of proposed forecast model will be illustrated in Section 3. Section 4 will present the sales data set and results of our experiment, and Section 5 will discuss the contributions of our model.

This paper constructs a predicting model to mitigate the problem of product demand forecast. It is aided by the utilization ICA and GHSOM as a preprocessing tool before building a SVR product demand prediction model. As indicated by the experimental results, the ICA–GHSOM–SVR model proposed in this paper yields the many insights. In general, a forecasting model is developed to deal the general forecast problem. However, there are numerous data features that will influence the performance of forecast. Many researchers have mentioned data grouping is a better way to improve the accuracy of forecasting model. Data with similar features will be classified in the same group, hence the forecasting result for each group will more accurate. In this research, GHSOM is successfully applied for data grouping. However, there still exist many noises in the original data, and they will cause the bad performance of forecasting. The ICA model is integrated in the forecast problem to detect and remove the noise of data and further improve the performance of the forecasting model in this study. The current research has focused on the development of an effective demand forecasting model for a company which has many customers/branches. The clustering results of the GHSOM–SVR and the proposed hybrid forecasting model only provide the temporary clusters which are used to build more effective forecasting models. To improve the practicability of the proposed hybrid forecasting scheme, finding the meaning of the clustering results of the proposed hybrid demand forecasting scheme will be investigated in the further study.