Prediction-based dynamic multi-objective optimization algorithms (DMOEAs) are widely used to explore the relationships of Pareto-optimal solutions (POSs) under continuous time steps, aiming to tackle dynamic multi-objective optimization problems (DMOPs). However, DMOPs with irregular POS shapes pose significant challenges to the quality of predicted solutions owing to the misaligned binding of solutions. To bridge this gap, this paper proposes a learning algorithm based on similarity identification and knowledge transfer, called SIKT-DMOEA, which comprises the following three steps. Firstly, a cluster centers-driven feedforward neural network (CCD-FNN) with global optimal binding assignment is constructed, aiming to learn the regional POS dynamics between adjacent environments. Secondly, a similarity identification technique archives valuable knowledge in historical environments and transfers it to the current environment for evolutionary acceleration. Finally, a population reconstruction strategy is presented for adaptive guidance according to the dominant property of each solution, which approximates the new POS with superior convergence and distribution. Comprehensive experiments demonstrate that SIKT-DMOEA manifests competitiveness when addressing DF test problems and one real-world application problem compared to state-of-the-art DMOEAs. SIKT-DMOEA has corroborated its capability of effectively reducing the loss of population convergence and diversity facing different patterns of environmental changes.

In recent years, prediction-based algorithms have attracted much attention for solving dynamic multiobjective optimization (DMO) problems in the evolutionary computing community. However, this class of algorithms still has potential for further improvements by enhancing the historical information extraction approach to balance convergence and diversity. In this article, we propose a DMO algorithm based on a recurrent neural network (RNN) to balance the population’s convergence and diversity in dynamic environments. The RNN model in the proposed algorithm employs online learning in order to constantly improve according to the increasing evolutionary information. Meanwhile, differing from most existing prediction-based algorithms, the learning machine is not limited by assumptions, such as linear or nonlinear correlation, when it predicts new solutions for future evolutionary environments. Besides, an auxiliary strategy is performed, which adaptively introduces the random or mutated solutions according to the error losses between the prediction solutions and the optimal solutions in the whole optimization process. The experimental results show that the proposed algorithm is more effective for handling DMO problems than several recent algorithms.

Multitype satellite observation, including optical observation satellites, synthetic aperture radar (SAR) satellites, and electromagnetic satellites, has become an important direction in integrated satellite applications due to its ability to cope with various complex situations. In the multitype satellite observation scheduling problem (MTSOSP), the constraints involved in different types of satellites make the problem challenging. This article proposes a mixed-integer programming model and a generalized profit representation method in the model to effectively cope with the situation of multiple types of satellite observations. To obtain a suitable observation plan, a deep reinforcement learning-based genetic algorithm (DRL-GA) is proposed by combining the learning method and genetic algorithm. The DRL-GA adopts a solution generation method to obtain the initial population and assist with local search. In this method, a set of statistical indicators that consider resource utilization and task arrangement performance are regarded as states. By using deep neural networks to estimate the$Q$value of each action, this method can determine the preferred order of task scheduling. An individual update strategy and an elite strategy are used to enhance the search performance of DRL-GA. Simulation results verify that DRL-GA can effectively solve the MTSOSP and outperforms the state-of-the-art algorithms in several aspects. This work reveals the advantages of the proposed generalized model and scheduling method, which exhibit good scalability for various types of observation satellite scheduling problems.