인공신경망을 활용한 선형재료절단문제의 휴리스틱 기법 선택

• Published in: 한국CDE학회 논문집 Vol. 25; no. 1; pp. 67 - 76
• Main Authors: 강민구(Minkoo Kang), 오지웅(Jiwoong Oh), 이유철(Youcheol Lee), 박기진(Kiejin Park), 박상철(Sangchul Park)
• Format: Journal Article
• Language: Korean
• Published: (사)한국CDE학회 01.03.2020; 한국CDE학회
• Subjects: 기계공학; Deep learning; One-dimensional cutting stock problem; Classification; Heuristic method; Machine learning; Artificial neural network (ANN); Combinatorial optimization
• ISSN: 2508-4003; 2508-402X
• DOI: 10.7315/CDE.2020.067

One-dimensional cutting stock problem (1DCSP) is a problem mainly applied in the pipes, cables, and paper rolls industries, and it is a problem of minimizing the trim loss of the stock while satisfying the demand of orders. 1DCSP can be solved by the integer linear programming to get an optimal solution. However, the computation time is exponentially increased depending on the number of types of orders and its quantity demanded. Although many heuristic methods have been proposed to solve the problem, it is difficult to develop a heuristic method that always provides a good solution to various problems due to the performance that is highly dependent on problem domain. In this paper, we propose a method to generate observations by producing various 1DCSPs, and then use the artificial neural network (ANN) algorithm to select the heuristic method that provides a good near-optimal solution for any 1DCSP. ANN models were implemented using the Sequential module of TensorFlow 2.0 Keras. According to the experimental results, the minimum value of root mean square error (RMSE), mean absolute error (MAE), accuracy, precision and recall are found with specific combination of parameters of batch size, epoch number and optimizer. KCI Citation Count: 0