Dynamic cost forecasting model based on extreme learning machine - A case study in steel plant
Introduction
In competitive and complex markets, such as the cold rolling commodities in steel plants, having the ability to forecast major material cost and envisioning the behavior of changing prices in each of them constitute a competitive advantage for organizations. One of the most important cost-related decisions is to know how to produce good commodities at the lowest possible cost. This involves understanding the origin of costs in an integrated steel plant along with understanding and managing the process, which leads to the lowest possible costs (Troelsen, 2006). Since late 1960s, the purchasing issues related to raw materials have been tackled well by enterprises. Purchasing raw materials is not only the beginning of all production activities and linked to the connection between production operating activities, but also the major part of the production cost for an enterprise (Gao & Tang, 2003). By estimating, material cost accounts for 60–80% in total cost (Bender, Brown, Isaac, & Shapiro, 1985), especially in steel plant.
Making a time series forecasting model for raw material price assists in developing the dynamic cost system for cold rolling product in the steel plant. Actually, time series prediction techniques have been used in many real-world applications, such as financial market prediction, electric utility load forecasting, weather and environmental state prediction, and reliability forecasting. The existing traditional quantitative approaches include heuristic methods, such as time series decomposition and exponential smoothing as well as time series regression and autoregressive and integrated moving average (ARIMA) models that have formal statistical foundations (Chu & Zhang, 2003). Nevertheless, their forecasting ability is limited by their assumption of a linear behavior, and thus, is the results are not always satisfactory (Sapankevych and Sankar, 2009, Zhang, 2003). This paper considers establishing a dynamic cost forecasting model of the end products in a large-scale steel plant, studies the special cost forecasting model involved, proposes an extreme learning machine (ELM) algorithm to predict the price of raw materials, and construct a cost simulation system for cold rolling products.
The rest of this paper is organized as follows. Section 2 describes the related work. Section 3 introduces the characteristics of steel plants. Section 4 discusses the forecasting model and describes the features and behavior of the model in detail. Section 5 illustrates an example of the dynamic cost simulation model. Conclusions and further research directions are outlined in Section 6.
Section snippets
Literature review
There exist some studies in this field—such as the big data implementation framework (Dutta & Bose, 2015), business analytics (Dubey and Gunasekaran, 2015, Ji-fan Ren et al., 2016), supply chain management of production design (Li, Tao, Cheng, & Zhao, 2015), manufacturing (Dubey, Gunasekaran, Childe, Wamba, & Papadopoulos, 2015), servitization supply chain (Opresnik & Taisch, 2015). Big data leverages the information eco-system to understand customer behavior trends, to optimize supply chains,
Steel plant manufacturing process
The steel plant considered in this study produces approximately 15 million tons of steel-iron products (slabs, plates, wire, hot rolling coils, etc.) per year. Fig. 1 shows the manufacturing process of cold rolling products in the large-scale integrated steel plant in Asia. Consequently, a large quantity of raw materials is required in this plant every year. As illustrated in Fig. 1, the process of steel manufacturing from raw materials typically includes six phases: (1) Blast Furnace Plant
Research methodology
The following section presents the dynamic cost forecasting model based on ELM for cold rolling products in an integrated steel plant by integrating GRA and ELM. GRA computes the Grey Relation Grades (GRG) which are the influential degree of a compared series by relative distance. Subsequently, the data comprising these input and output pairs are divided into training, testing, and predicting data. All the datasets are normalized into a specific range [0, 1]. The extreme learning machine (ELM)
Experimental results and discussions
Using the ELM model to predict future price can increase accuracy in the proposed system. The procedures of the experiments and the results are described sequentially in the following subsections. We proposed the ELM forecasting model to predict the weekly and daily prices of iron ore and coking coal. Next, we provided those validated forecasting prices as the input data into the costing system to calculate the dynamic cost for cold rolling products.
Conclusions
In this study, we have used massive raw material prices as the input data for proposing a dynamic cost forecasting model. Additionally, we collected the production information, cost information, resource consumption rate, and production yield of all products in each processing stage as the production standard to assist in establishing the forecasting model. The product information has been collected for over fifteen years in the company. This study attempts to make use of the huge data to build
Acknowledgements
The authors would like to thank Professor Jen Teng Tsai for his positive comments during writing the manuscript. We also gratefully acknowledge the Ministry of Science and Technology, Taiwan, ROC, for support under contract MOST 104-2410-H-507-003.
References (29)
- et al.
Sales forecasting using time series and neural networks
Computers & Industrial Engineering
(1996) Insights from hashtag# supplychain and Twitter analytics: Considering Twitter and Twitter data for supply chain practice and research
International Journal of Production Economics
(2015)- et al.
Fuzzy Delphi and back-propagation model for sales forecasting in PCB industry
Expert Systems with Applications
(2006) - et al.
A comparative study of linear and nonlinear models for aggregate retail sales forecasting
International Journal of Production Economics
(2003) - et al.
Managing a big data project: The case of ramco cements limited
International Journal of Production Economics
(2015) - et al.
A multi-objective model for purchasing of bulk raw materials of a large-scale integrated steel plant
International Journal of Production Economics
(2003) - et al.
Data quality for data science, predictive analytics, and big data in supply chain management: An introduction to the problem and suggestions for research and applications
International Journal of Production Economics
(2014) A sales forecasting system based on fuzzy neural network with initial weights generated by genetic algorithm
European Journal of Operational Research
(2001)- et al.
Forecasting the NYSE composite index with technical analysis, pattern recognizer, neural network, and genetic algorithm: a case study in romantic decision support
Decision Support Systems
(2002) - et al.
Order-planning model and algorithm for manufacturing steel sheets
International Journal of Production Economics
(2006)