Assessing behavior-conditioned building flexibility potential across building types for photovoltaic accommodation

Building flexibility is increasingly recognized as an important demand-side resource for renewable energy integration, yet its characterization in real buildings remains challenging because flexibility is strongly conditioned by operational behavior and is difficult to quantify without detailed system information. This study proposes a behavior-informed framework to assess behavior-conditioned flexibility potential across building types using historical operation data, and further examines its potential implications for photovoltaic (PV) accommodation under building-cluster scenarios. Using 15-min measured electricity data from multiple types of public buildings in Hangzhou, China, the framework first establishes baseline load behavior through quantile-based modeling, then identifies feasible operating ranges from historically comparable operating conditions, and finally estimates behavior-conditioned flexibility potential in terms of both instantaneous adjustable power and duration-based adjustable energy. XGBoost-SHAP analysis is further employed to interpret the statistical associations between input features and the estimated flexibility indicators, and an aggregated analysis is conducted to evaluate the value of flexibility potential for PV accommodation under different curtailment constraints. For the studied public-building dataset, the estimated flexibility potential shows clear heterogeneity across building types. Load level, hour of day, outdoor temperature, and daily load characteristics show the largest contributions to the model predictions, suggesting that the estimated flexibility potential is closely associated with energy-consuming behavior. At the cluster level, the results further reveal that high flexibility does not necessarily translate into high PV accommodation potential. Under a 15% curtailment threshold, the average PV capacity factor of the realistic mixed-building aggregation scenario increased from 0.288 to 0.369 in the flexibility-enhanced scenario, corresponding to an absolute increase of 0.08 and a relative improvement of 27.9%. Across different aggregation capacity levels, the relative improvement in PV accommodation ranged from 23% to 30%. These results indicate that the cluster-level value of flexibility depends strongly on the temporal match between flexible load patterns and PV generation. This study provides a scalable data-driven approach for characterizing building flexibility from historical operational data and offers new insights into the potential contribution of behavior-conditioned flexibility to renewable energy integration and building-cluster planning under specified scenarios.