Otomotiv yan sanayi firmasında yapay sinir ağları ile talep tahmini
Demand forecasting with artificial neural networks in an automotive supplier company
- Tez No: 496492
- Danışmanlar: PROF. DR. FERHAN ÇEBİ
- Tez Türü: Yüksek Lisans
- Konular: Endüstri ve Endüstri Mühendisliği, Industrial and Industrial Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2017
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: İşletme Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 99
Özet
Otomotiv sektörü tüm dünyada ekonominin önde gelen iktisadi faaliyet kollarından biridir. Her geçen gün üretim hacimleri artmaktadır. Teknolojiler ve üretim hacimleri geliştikçe de tedarik zincirine verilen önem artmaktadır. Tüm Dünyada daha önceki dönemlerde birincil önem verilen üretimin geliştirilmesi iken daha sonraki yıllarda bu önem pazarlama satış faaliyetlerine kaydırılmıştır. Günümüzde ise şirketlerin verimliliklerini, karlılıklarını arttırmak için önem verdikleri alan tedarik zinciri yönetimi olmuştur. Tedarik zinciri; üretim için gerekli olan hammadde, malzeme ihtiyaçlarının temininden müşteriye ürünün teslimine kadarki tüm süreçleri kapsamaktadır. Tüm bu süreçlerin verimli ve etkin bir şekilde yönetilmesi için ilk basamak talep öngörülerinin oluşturulmasıdır. Talep öngörüleri oluşturulduktan sonra talebin uzun, orta ya da kısa vadeli oluşuna bağlı olarak; stratejik planlar, yatırım planları, kapasite planları, üretim planları, malzeme-işgücü-hammadde ihtiyaç planları yapılmaktadır. Talep tahminlerinin vadesine göre ürün grubu bazında, ürün çeşidi bazında ya da ürün bazında tahminler yapılabilir. Talebin kapsadığı zaman aralığı azaldıkça ya da ürün çeşitliliği arttıkça duyarlılık artmaktadır. Tahminlerin her zaman bir hata payının olduğu unutulmamalıdır. Tahmin yöntemleri kalitatif ve kantitatif olarak iki ana gruba ayrılmaktadır. Kalitatif yöntemler sayısal olmayan, yoruma dayalı tahminler iken kantitatif tahminler sayısal veriler içeren çeşitli matematiksel işlemler sonucu elde edilen tahminlerdir. Tahmin yöntemlerinde teknolojinin gelişmesi ile birlikte yapay zeka uygulamaları da etkin bir şekilde kullanılmaya başlanmıştır. Çok çeşitli faktörleri dikkate alarak verimli sonuçlar verebilen yapay zeka uygulamalarından Yapay Sinir Ağları bu çalışmada kullanılmıştır. Öncelikli olarak talebi tahmin edilecek olan ürün, daha önceki satış çalışanlarının öngörülerine dayanan tahminlere göre sapması yüksek olan ve satışı toplam satışın %80ini oluşturan grup içinden seçilmiştir. Ürünün tahmin sapmaları MAPE değerlerine göre incelenmiştir. Yapay Sinir Ağlarında daha önceki literatür çalışmaları baz alınarak otomotiv satışlarını etkileyen geçmiş dönem otomobil satışları, Türkiyede otomobil üretimi, GSYH (Gayri Safi Yurtiçi Hasıla), Euro ve Dolar kuru, araç kredisi faiz oranlarının değişimi, Tüketici Güven Endeksi, Reel Kesim Güven Endeksi, Otomobil satışları gibi değerler girdi değişkeni olarak alınmıştır. Çıktı değişkeni olarak geçmiş dönem satışları alınmıştır. Tüm değişkenlerle YSA çalışmasından sonra, aynı datalarla regresyon çalışılmıştır. Regresyonda eşdoğrusallık değerlerine bakılarak modele katılan girdi değişkenleri TGE, Faiz oranı, Otomobil üretimi ve Euro kurları alınarak tekrar YSA çalışılmıştır. Regresyon ve ilk YSA sonucundaki MAPE değerleri ve ikinci YSA sonucundaki MAPE değerleri incelendiğinde tahmin performasının arttığı gözlenmiştir.
Özet (Çeviri)
The automotive sector is one of the leading economic sectors of the economy all over the world. Production volumes are increasing day by day. As technologies and production volumes evolve, importance attached to supply chain management is increasing. While production, across the whole world has been of primary importance in earlier periods, later, importance has been diverted to marketing and sales activities. In today's world, supply chain management has become the focus for boosting productivity and profitability. In today's world, supply chain management has become an area where companies value their productivity and profitability. Supply Chain Management's significance lies in the involvement of customers as a partner in the delivery of products and services which offers several advantages. Firstly, this integration provides the supplier information flow to the customer. removing ambiguities and allowing the supplier to better analyze the manufacturer's needs. It also enables the supplier to reduce the quantity of supply, shorten the supply period and thus reduce its costs. The customer also becomes able to respond faster. The supply chain covers all the processes from the procurement of raw material and material requirements needed for production up to the delivery of the customer's product. The motor vehicle is composed of about 5000 parts, which are different in material structure, production type, technology. It is not possible to produce all of these parts within the parent industry. Having different options for each part, product diversity, investment needs have paved the way for the birth of automotive supplier industry to meet the needs of the industry. The parent and the supplier industries are mutually dependent on each other with regard to commercial, technical and production capacity areas. The first step is to create demand forecasts for effective and efficient management of all Supply Chain processes. Demand forecasting is the most important business activity because it guides all other activities. Which market will be followed, how much inventory will be kept, what products will be produced, how many workers will be hired; all these decisions are derived from demand estimates. Forecasting and planning are two closely related actions. Planning is the process of choosing actions to be taken in the light of predictions. Poor forecasting leads to poor planning and poor planning leads to unwanted, unprepared situations like loss of sales and overstock. In order to make demand forecasting, internal and external factors affecting the product demand are identified first. Then, the estimation method is selected and primary estimates are made. The consistency of these estimates is monitored and the estimation method is developed. Forecasting methods are divided into two groups, qualitative and quantitative. A number of factors must be considered in the selection of the method such as the existing time to prepare the estimation, the long or short term of the decisions to be made, access to data, the quality of the data, the characteristics of the individuals who will apply the method, limits of variance. After the demand forecasts have been established, depending on the long, medium or short-term nature of the demand; Strategic plans, investment plans, capacity plans, production plans, material-labor-raw material requirements plans are made. Demand forecasts can be made on the basis of: product group, product type or product; depending on the timespan of the demand. Sensitivity increases as the covered timespan of the demand decreases or as the product variety increases. It should taken into consideration that estimates always have margins of error. Forecasting methods are divided into two main groups as qualitative and quantitative. Qualitative methods are estimations based on non-numerical interpretations, while quantitative estimates are obtained from various mathematical operations involving numerical data. Qualitative methods are used when the situation is not clear, when there is very little data (new product, new technology) or when intuition / experience is required. They are divided into four groups. These are managerial view, the salesperson's estimation, the Delphi method and consumer research. The managerial view is a method that takes the views of a small group of high-level managers. It is the most used method according to the other methods, except the field of product forecasting. Salesperson's estimation: Their opinion about the future of the market can be influential on demand forecasting. The Delphi method is applied to experts via a panel by addressing a number of questions about the future of the market or existing situations. Consumer research is the way companies run their research through telephone, personal interviews or questionnaires. Quantitative methods include in mixed (experimental) methods and time series. Mixed methods use the relationship between the data from the source and the data being targeted for estimation. In the time series method, it is predicted that the past values will continue similarly. Time series are simple methods, exponential correction, weighted average, moving average, moving weighted average, Box-Jenkins, trend analysis methods. Mixed methods are regression and artificial intelligence applications. Artificial intelligence applications are Genetic algorithms, Artificial Neural Networks. The accuracy of the estimates must be checked before the correct demand forecasting method can be selected. The prediction that gives the smallest error is the most successful one. None of the estimation methods give 100% correct results. The forecast error is the difference between the estimated value and the actual value. There are different methods of measuring prediction errors. These are mean error squares (MSE), mean absolute error (MAE), and mean absolute percent error (MAPE). Along with the development of technology in estimation methods, artificial intelligence applications have started to be used effectively. Artificial Neural Networks have been used in this study from artificial intelligence applications that can yield fruitful results taking into account a variety of factors. Primarily; the product group for which the demand is estimated, is selected from the main group whose deviation is the highest between previous estimates based on the predictions of salespeople and whose sales constitute 80% of the total sales. The estimated deviations of the product group are examined according to the MAPE values. Based on previous estimation studies with artificial neural network, it is seen that past automobile sales that affect past car sales, automobile production in Turkey, GDP (Gross Domestic Product), Euro and Dollar exchange rate, change in interest rates of vehicle loans, Consumer Confidence Index, Real Sector Confidence Index , gasoline prices sales were taken as input variables. In this study, as an input, the survey results of the probability of purchasing an automobile in the next 12 months of TUIK (Turkish Statistical Institute) were taken as an input. Past product sales have been received as output variable. 32-month data set was used. Multi-layer feed forward backpropagation algorithm, which is frequently used in estimation studies, is used. Momentum Proposed Back Propagation Algorithm (traingdm) is used as training function and learngdm function is used as learning function. Log-Sigmoid (logsig) and Hyperbolic tangent sigmoid (tansig) have been tested as an activation function. With the tansig estimates were continued because of its MAPE values were lower. Matlab R2011B is used as the software. Experiments were performed to select optimum values for learning coefficient, momentum coefficient, number of iterations, number of neurons, number of layers respectively. Since there is 32 months of data available, it is used for 25 months of data training. Firstly, the momentum coefficient for learning coefficient selection is fixed at 0.6. The appropriate learning coefficients were selected as 0.2 and 0.4. In the next step, we experimented with 0.2 and 0.4 learning coefficients and all momentum coefficients between 0.1-0.9. The appropriate combination of learning and momentum coefficients are provided in the 0.2-0.6 respectively. The appropriate number of iterations are tested with the learning coefficient is 0.2, the momentum coefficient is 0.6. The optimal number of iterations is 2250. In order to determine the number of neurons, experiments were carried out with 1-20 neuron counts. The optimum number of neurons were selected as 7. Later attempts were made to determine the number of layers. The appropriate number of layers was selected as 5. After the training of the artificial network, 7 months data for the test were presented to the network. The average MAPE value was 10.3 %. The last 6 months MAPE value was 12% in the product group selection. After all the variables were analyzed by ANN, the same data was used for regression. SPSS software was used in the regression. Based on the collinearity values, the input variables included in the model are Consumer confidence index, interest rate, car production and Euro currencies. The MAPE value has been 6.22 for the last 7 months. ANN was used again with the inputs used in the regression. All the steps in the first step are repeated in order. The appropriate learning and momentum coefficients are 0.8 and 0.3, the number of iterations is 4000, the number of neurons is 9, and the number of layers is 3. The MAPE value was 4.91% when tested with data from the last 7 months. Using more than one method improves the outcomes.When the regressions MAPE value and the first ANNs MAPE value are compared with the MAPE values of the second ANN, the estimation performance was observed to increase.
Benzer Tezler
- Yapay zeka tabanlı araç koltuğu tanıma sisteminin geliştirilmesi
Development of artificial intelligence based vehicle seat recognation system
ALİ İHSAN BADEM
Yüksek Lisans
Türkçe
2023
Mühendislik BilimleriBursa Teknik ÜniversitesiAkıllı Sistemler Mühendisliği Ana Bilim Dalı
DOÇ. DR. İZZET FATİH ŞENTÜRK
- Otomotiv yan sanayi firmasında optimum emniyet stoğu belirleme modeli
Optimum safety stoke detection model in automotive industry
SUZAN EREN
Yüksek Lisans
Türkçe
2018
Endüstri ve Endüstri MühendisliğiKocaeli ÜniversitesiEndüstri Mühendisliği Ana Bilim Dalı
PROF. DR. ALPASLAN FIĞLALI
- A Model for quality and productivity improvement in an automative supplier company a case study
Otomotiv yan sanayii için bir verimlilik ve kalite iyilieştirme modeli, uygulamalı bir çalışma
ERSAN MERİÇ
- Altı sigma metodolojisi ve otomotiv yansayisinde uygulaması
Six sigma methodology and an implementation in automotive supplier
NAZLI YİĞİT
Yüksek Lisans
Türkçe
2010
Endüstri ve Endüstri Mühendisliğiİstanbul Üniversitesiİşletme Bölümü
PROF. DR. NECDET ÖZÇAKAR
- Otomotiv sektöründe QS 9000 olgusu ve plastik enjeksiyonla kalıplama kuruluşuna uygulanması
The Fact of QS 9000 in the automotive sectors and a practise in the plastic injection moulding company
AHU GÜRKAVCI DEMİRLER
