The frequency nadir is both an indicator for situational awareness and a basis for emergency control, which consists of the maximum frequency deviation and the frequency nadir time, thus fast and accurate frequency nadir prediction is important for frequency stability of the modern power system. Considering the strengths and defects of physical-driven and data-driven methods, a physical-data integrated-driven method is proposed. As the physical-driven part, Frequency Nadir Prediction (FNP) model can solve the analytical solution of the frequency nadir and obtain the initial prediction results at high speed. As the data-driven part, Back Propagation Neural Network (BPNN) can correct the errors of the initial prediction results online to improve the accuracy. The serial integration approach is applied to integrate both models and obtain the final prediction results at both high accuracy and speed. Compared with the existing integrated-driven methods, FNP model can preserve more key influences, which greatly reduces the dependence of BPNN on sample data and feature dimensions. The case studies over the New England 39-bus system verify that the proposed FNP-BPNN integrated model can provide a more reliable indicator and basis for power system frequency stability analysis and control.