Building energy flexibility is critical for integrating renewables and achieving efficient energy management. This study develops a multi-indicator assessment framework and a customized battery storage regulation strategy to improve flexibility in PV-integrated buildings. LSTM and Rolling-XGB models are employed to accurately predict building energy demand and PV generation, respectively. A data-driven battery model simulates system performance under three scenarios: standalone PV, uncoordinated PV-BESS, and coordinated PV-BESS with forecasting. Validation is conducted on a near-zero-energy building in Ningbo Zhejiang. Results show that the coordinated strategy significantly enhances flexibility and reduces energy costs by nearly 40% compared to the PV-only case. The system achieves 62.75% local energy coverage and 34.77% energy surplus time percentage, demonstrating the effectiveness of the proposed approach in balancing supply and demand.