The traditional lithium-ion battery design approach relies on experimental trial-and-error methods, which are costly and inefficient. The second-generation, structure-performance relationship model based forward design method offers some cost and efficiency benefits but it relies heavily on experience, making it difficult to achieve optimal design. This work proposes a data-driven “retrieval + generation” battery design approach that aims to overcome the limitations of existing methods and further enhance the level of design. This work first extracts design variables and performance indicators from battery design scenarios and automotive power battery working conditions, respectively. Taking a prismatic wound lithium iron phosphate battery as a prototype, hierarchically using material balance calculation and simulation to obtain performance indicators under different design variables, a database from design to performance is built. Then, based on this database, explained the characteristics of fixed-size design and the correlation between design variables and performance indicators is quantified using Spearman’s correlation coefficient and the variation in performance indicators, which can provide a reference for battery design. Finally, using the open-source Faiss library and radial basis functions, a “retrieval + generation” design is realized, and a design case is given. This case, without relying on simulation, enables rapid design within 5% error, demonstrating that the “retrieval + generation” approach, as a data-driven design method, is feasible and has great potential in improving efficiency and reducing costs.