The evolutionary algorithm based on decomposition is an effective method in dealing with many-objective optimization problems. The traditional decomposition method relys on a set of uniformly distributed reference vectors, which decomposes the multi-objective optimization problem into a set of single-objective subproblems through aggregation functions, and then optimizes these subproblems simultaneously. However, these predefined reference vectors perform poorly in solving complex many-objective optimization problems because of the inconsistency of the distribution of reference vectors and the shape of the Pareto front. Aiming at the above problems, a many-objective evolutionary algorithm based on adaptive boosting learning (MaOEA-ABL) is proposed. The algorithm can be divided into two stages. In the first stage, an adaptive boosting learning algorithm is used to adjust the predefined reference vectors. In the learning process, useless vectors are deleted and new vectors are added. In the second stage, an unbiased decomposition method of Pareto shape is designed. Simulation has been conducted on the MaF test problems. The experimental results show that the IGD (inverted generational distance) mean value of MaOEA-ABL is better than that of the comparison algorithms in 67 % of the test functions, which indicates that the MaOEA-ABL performs well in many-objective optimization problems with complex Pareto front.