Evolutionary Multi-objective Algorithms Based On Global Optimization And Local Learning

The optimization problem has always been recognized as one of the most difficult but important problems in scientific research and engineering practice. According to the number of optimized objectives, the optimization problem can be divided into two categories:the single-objective optimization problem and the multi-objective optimization problem (MOP). Inspired by the biological intelligence, evolutionary algorithms can gradually approximate the optimal solution by population evolution and effectively solve most of the optimization problems. However, they have some inherent limitations. For example, the convergent speed is relatively slow especially for complex optimization problems. It is a good way to improve the algorithm performance by integrating local learning strategies. The aim of the dissertation is to propose effective multi-objective evolutionary algorithms by combining local learning algorithms and to apply them to solve the practical problems. The main contribution of this dissertation can be summarized as follows:(1) Basic evolutionary multi-objective optimization algorithms have some shortcomings like slow convergence. Therefore, a new hybrid multi-objective optimization algorithm based on local search chains is presented. In the proposed algorithm, the parameters of the local search strategy, such as the search step and the success rate of local learning, are encoded into the individuals. In this way, after the execution of local search, these parameters are kept so as to improve the efficiency of local leaning further. To avoid the loss of diversity, we design a diversity based selection mechanism to select proper individuals to perform local learning. In addition, an adaptive computational resource assignment strategy is established to balance the global optimization and local learning. Results of comparative experiments show that the introduction of local search strategy can improve the performance of evolutionary multi-objective optimization algorithm.(2) Inspired by the multi-armed bandit problem in the game theory, we propose an adaptive multimeme algorithm for solving the single-objective flexible job shop schedule problem, which is one of the most difficult combinatorial optimization problems. Due to the complexity of the problem, traditional mathematic programming methods may not be effective or fail to work. Population-based algorithms may not generate satisfied solution in the limited time. Diverse hybrid algorithms are designed to solve this optimal problem. How to choose the suitable method from so many choices for the current problem is a difficult problem. The proposed algorithm provides an adaptive strategy that can select the search operator automatically. Specifically, the adaptive strategy first uses a sliding window with limited length to record the recent performance achieved by the search operators, and then guide the subsequent selection of search operators. Comparison results on several well-known sets of benchmark problems show that the proposed algorithm can achieve good performance.(3) Traditional recommendation techniques in recommender systems mainly focus on improving recommendation accuracy. However, personalized recommendation, which considers the multiple needs of users and can make both accurate and diverse recommendations, is more suitable for modern recommender systems. Based on this consideration, a new recommendation algorithm based on multi-objective optimization is proposed. In this algorithm, the task of personalized recommendation is modeled as a multi-objective optimization problem. A multi-objective recommendation model is proposed. The proposed model maximizes two conflicting performance metrics termed as accuracy and diversity. The accuracy is evaluated by the probabilistic spreading method, while the diversity is measured by recommendation coverage. The proposed recommendation method can simultaneously provide multiple recommendations for multiple users in only one run, by encoding all the recommendations for multiple users in one individual. To improve the algorithm performance, a clustering technique is used to split a large number of users into several relatively small subsets. Experimental results demonstrate the effectiveness of the proposed algorithm.(4) For the characteristic of the tag-aware system, a new recommendation algorithm based on deep neural network is proposed. In tag-aware systems, users can use any vocabulary as a tag to describe or evaluate items. Arbitrarily associated by users, tags may suffer from many problems, such as redundancy, and ambiguity. Thus, the process of tag information is difficult but import for the design of the algorithm. To solve this problem, the deep neural network is introduced to form a new recommendation algorithm. In the proposed algorithm, users’profiles are initially represented by tags and then a deep neural network model is used to extract the in-depth metadata from tag space layer by layer. In this way, representations of the raw data will become more abstract and advanced, and therefore the unique structure of tag space will be revealed automatically. Based on those extracted abstract metadata, users’profiles are updated and used for making recommendations. The experimental results demonstrate the usefulness of the proposed algorithm and show its superior performance over the clustering based recommendation algorithm. In addition, the impact of deep network’s depth on the algorithm performance is also investigated.