Abstract: Considering the composite scenario of energy storage peak shaving(PS) and frequency regulation(FR) can help improve the economy and application effectiveness of energy storage in renewable energy grids. An optimized configuration method is proposed for energy storage capacity considering PS and FR data-driven modeling. Firstly, based on the mathematical description of traditional PS and FR models, mapping rules are constructed for multi time scale PS and FR scenarios. Then, leveraging the end-to-end nature of deep learning and excellent non-linear mapping capabilities, the long short term memory(LSTM)-Transformer model is used to mine the data features of the FM and FD scenarios to ensure the accuracy of the trend computation and accelerate the frequency computation speed. Finally, based on historical data and K-Medoids clustering, typical day sets are selected and data-driven models are embedded into the energy storage capacity optimization configuration model. With the economic efficiency of energy storage configuration as the objective function, a dual layer optimization model for energy storage configuration operation is established considering the operational constraints within the system. The calculation example shows that compared to a single scenario, the economy of multi scenario energy storage configuration has improved by 34.7%, and data-driven auxiliary energy storage optimization models can effectively improve calculation speed and accuracy.