Mismatches between forecasted and actual demand, for construction, repair and maintenance work in a regulated utility, is a growing risk for the performance of the supply chain. High target service levels and high levels of demand uncertainties necessitate Inventory Management to maintain a significant amount of safety stock to buffer against uncertainties. Furthermore, the increasing complex of supply chains make it difficult to anticipate possible effects changes, such as improved forecasting or policy changes. In this thesis we propose an innovative approach for demand forecasting by creating a predictive model based on identifed patterns of repair and maintenance projects underlying the demand data. We further present a unique approach to simulate an overall supply chain, using locally available data, giving the supply chain the ability to evaluate the implications of improved forecasting on the overall network. Through the improved methodology the supply chain can reduce the amount of noise in the data and create a forecast based on data that better represents the real demand. The proposed method improves on current forecasting methods by reducing forecasting noise, such as bullwhip and human error, by tying the forecast for material demand to the forecast of the source of the demand. To do so, we use unsupervised clustering K-means to identify similar consumption behaviors in the data. We further propose the use of a time-series analysis and hierarchical forecast aggregation for the creation of the final forecast, although this will not be the focus of this thesis. Although the results of the clustering process were inconclusive, we present data that supports the validity of out premise and propose alternative algorithms that could produce superior results. In addition we propose a supply chain network simulation to validate the model and valuate its affects. We use the model to emulate the possible effects of forecast improvements on the overall supply chain.;
展开▼
