Evolutionary optimization for sales forecasting and order scheduling in fashion supply chain management

Abstract

Facing with the increasingly fierce market competition, it is of paramount significance for apparel companies to establish effective and efficient fashion supply chains, which facilitate the reduction of costs, the improvement of service quality, and the enhancement of the competition ability of the companies. To build a competitive supply chain in fashion industry, it is necessary to improve its decision-making ability to manage problems, such as inventory problems, assembly line balancing problems, and so on. Among these problems, sales forecasting and order scheduling problems attract greater attention, because they largely influence the retailing and manufacturing in fashion supply chains. The primary purpose of this research is to solve sales forecasting and order scheduling problems in fashion supply chains via two hot branches of evolutionary optimization (multiobjective evolutionary optimization and robust evolutionary optimization) for the first time, and hence to establish a competitive and robust supply chain in fashion industry. The thesis consists of three main parts: algorithm development (Chapter 4), the application of the multiobjective optimization-based neural network model in fashion sales forecasting (Chapter 5), and the application of robust evolutionary optimization for fashion order scheduling (Chapter 6). The algorithms developed in this thesis are evolutionary algorithms and they belong to a new branch of artificial intelligence. The first algorithm in this thesis is called nondominated sorting adaptive differential evolution (NSJADE), it is a part of the forecasting model for addressing the fashion sales forecasting problems. The second algorithm, known as event-triggered impulsive control scheme based differential evolution (ETI-DE), is developed as the optimization tool to get the robust schedules in fashion order scheduling. In detail, NSJADE is a new multiobjective evolutionary algorithm (MOEA), which is developed based on a classic MOEA, i.e., nondominated sorting genetic algorithm II (NSGA-II). NSJADE replaces the search engine of NSGA-II with adaptive differential evolution (JADE), and the proposed NSJADE shows better performance on multimodal problems according to the experimental results. NSJADE is utilized for the fashion sales forecasting problems. In addition, a new scheme called ETI is presented in the framework of differential evolution (DE), and a powerful DE variant ETI-DE is obtained. The experimental results demonstrate that ETI can greatly enhance the performance of ten DE variants, and success-history based adaptive differential evolution with ETI (ETI-SHADE) has the best performance among all the variants. ETI-SHADE is modified to fit into the robust evolutionary optimization, and then to optimize the order scheduling problem in fashion supply chains after forecasting. A multiobjective optimization-based neural network model (MOONN) is developed to handle a short-term replenishment forecasting problem in fashion supply chains. The model employs a new MOEA called NSJADE to optimize the input weights and hidden biases of NN for the short-term replenishment forecasting problem, which acquires the forecasting accuracy while alleviating the overfitting effect at the same time. Furthermore, the MOONN model also selects the appropriate number of hidden nodes of NN in terms of different replenishment forecasting cases. Experimental results demonstrate that the presented MOONN model can handle the short-term replenishment forecasting problem effectively, and show much superior performance to several popular forecasting models. Robust ETI-SHADE is applied to develop robust order schedules in fashion supply chains. Unlike non-robust optimization, robust ETI-SHADE uses the mean effective objective value feff as the optimization objective. And the schedules obtained by ETI-SHADE are robust to the uncertain daily production quantity during the real production process. Experimental results show that schedules obtained by robust ETI-SHADE have uncertainty-tolerant ability, and benefit the real-world production in fashion supply chains. The results of this research demonstrate that the utilization of multiobjective evolutionary optimization can offer satisfactory performance for the fashion sales forecasting problems, and the introduction of robust evolutionary optimization can generate robust schedules for fashion order scheduling problems. It is revealed that these two branches of evolutionary optimization are of paramount significance to the establishment of an effective and efficient fashion supply chain.