[Objective] To address the challenges of high photovoltaic （PV） curtailment rate and supply-demand mismatch faced by energy storage-free PV direct-driven AC systems under dynamic electricity prices， this paper proposes an optimal control strategy integrating price response with physical constraints to improve economic efficiency. [Methods] First， a convolutional neural network- long short-term memory （CNN-LSTM） forecasting model incorporating PV panel parameters and building thermal capacity constraints is constructed. Second， a threshold optimization model minimizing dynamic operational cost is designed. Finally， energy-storage-free coordinated switching is achieved using optimized thresholds and a hysteresis comparator. [Results] Taking a library in Northwest China as a case study for simulation， the results demonstrate that by leveraging the building’s thermal inertia for precooling， the load peak during high electricity price periods is reduced by 18%， and the system’s average daily operating cost is lowered by 21.6%. [Conclusions] The proposed closed-loop “forecasting-optimization-control” framework effectively enhances the system’s economy and flexibility in a market-oriented environment， providing a viable solution for energy conservation and carbon reduction in public buildings.