The military portable equipment power sources often employ multiple 18650-type lithium-ion batteries in series-parallel configurations, which possesses the different number of serially or parallel connected cells and heterogeneous battery types across different devices. Thus achieving the accurate identification and adaptive charging of heterogeneous lithium battery packs is significant in the increasingly complex and demanding battlefield environments. The traditional back propagation (BP) neural networks utilized for identifying lithium batteries suffer from the slow convergence, poor identification accuracy and local optima issues. To address these shortcomings, a system identification model based on the mind evolutionary algorithm optimized BP neural network is proposed to tackle the challenges posed by adaptive charging of military portable equipment power sources. This method first takes the widely used 18650-type lithium battery in portable equipment power sources as the research object, tests its charging data under different charging rates, and determines the key features of model construction through in-depth analysis of the lithium battery charging characteristic curve; Additionally in order to construct the system identification model, the mind evolutionary algorithm (MEA) algorithm is used to globally optimize the initial weights and thresholds of the BP neural network with battery type and state of charge (SOC) as output features during the network training period; Finally an experimental prototype based on a four switch Buck-Boost converter is developed, which collaboratively achieves the predictive identification and adaptive charging control of heterogeneous lithium battery packs by combing with the aforementioned adaptive charging strategy. The identification accuracy of the system identification model constructed by the MEA-BP neural network is compared with those of the BP model and the particle swarm optimization (PSO)-BP model. Experiments show that the proposed algorithm improves the identification and prediction accuracy of lithium battery type and SOC with the prediction errors controlled within 1%. It can effectively identify the lithium battery packs with different numbers of cells and SOC states, and achieve the adaptive charging control, demonstrating the superior estimation accuracy and robustness.