Batch dispersion problem (BDP) restricts batch traceability in large-scale discrete production and negatively impacts batch recall costs. However, previous research has ignored the complexity of the BDP in their analyses. This paper investigates the BDP under the composed bill of materials (BOM) and develops a mathematical model for the BDP with the goal of minimizing the total batch dispersion by utilizing the batch dispersion as a measure of the degree of dispersed usage of part batches. BDP-GAVNS, a hybrid genetic algorithm with variable neighborhood search, is devised for the BDP based on the demonstration that the BDP is an NPC problem. In BDP-GAVNS, memory banks were introduced to increase the diversity of individuals performing crossover operations. Additionally, the encoding method and infeasible solution repair program are designed according to the characteristics of BDP. Numerical experiments validate the viability and effectiveness of BDP-GAVNS in solving BDP. They demonstrate that (1) the optimal combination occurs when the ratio of individuals produced by the three types of population initialization methods, namely global selection (GS), local selection (LS), and random selection (RS), to the population takes values of 0.30, 0.10, and 0.60, respectively; (2) The memory bank enriches the source of individuals required for crossover operations and improves the performance of crossover operations; and (3) The BDP-GAVNS is more effective than the other five heuristic algorithms including genetic algorithms in seeking the optimal solution of BDP.