| Summary: | 碩士 === 義守大學 === 管理科學研究所 === 87 === Distribution center is a kind of facility, it plays the function of both centralizing and distributing in the process of the entity- distributing of the merchandise. It also has the ability of the order-management, warehousing, circulating processing and picking, it can even expand its function to the range of searching customers, having the final marketing, purchasing, production-designing, the brand-developing, and etc.. For the aspect of business, the existence of the distribution center comes to assist the exploitation of the market and the collection of information, by that, it leads to the complete control of the market, the distribution and the warehousing. That makes the balance of product-sale. For the aim of the above, first of all, merchandise of marketing demand must be under forecasting. In the competitive and changeable market, the elements which effect the shipping amount of distribution center are a lot and complicated. Besides the general quantitative factors under the time series, there are some qualitative factors existing. These factors, mutually, affect the shipping model of distribution center.
For the establishment of the shipping forecasting model of distribution center, this research apply the neural network, fuzzy neural network, genetic algorithm and weight-elimination to construct the shipping forecasting model regarding both quantitative and qualitative factors. In the part of establishing the effect-degree upon shipping amount caused by qualitative factors, the way applied in this research is to build a fuzzy IF-THEN rule base which is constructed and formed by fuzzy neural networks and take the advantages of memory and learning part of fuzzy neural network. After learning process about the known rule base has been sufficiently done, the fuzzy neural network cannot only own the function of inference, but it can also infer the new or rebuild the fuzzy number of rule for the unknown or lost fuzzy rule base. By means of that, it can build a intellectual rule base. In order to reducing the error of fuzzy neural network, the genetic algorithm has been applied in the research. With the use of the genetic algorithm for the search of the initial fuzzy weights of fuzzy neural network, this research proves that genetic algorithm surely can avoid the network from sinking into the local optimal and reduce the network error. Same as the purpose of reducing the network error, we eliminate some smaller fuzzy weights while the fuzzy neural network is trained. During the period of simulating, it does prove that the network error would be reduced by means of eliminating some smaller fuzzy weights. The better fuzzy neural network will, then, be obtained after we go through the above several ways.
After gaining the intellectual neural fuzzy network, we may, then, apply the rule base to get the affection of shipping model when the qualitative factors ever take place. We hereby integrate the historical shipping amount of the quantitative factors for construct a integration forecasting model without ignoring both quantitative and qualitative factors. And the integration forecasting model is build by the neural network in this entire research.
|
|---|
