With the increasing penetration of wind power into modern power systems, wind energy-integrated systems exhibit pronounced probabilistic uncertainty under the combined effects of stochastic wind speed fluctuations and small-signal disturbance. Traditional small-signal stability analysis methods based on deterministic models are therefore insufficient to support real-time control and stability assurance of wind turbines. To address the limitations of conventional offline small-signal stability assessment in providing effective real-time control decisions under high wind power penetration, an integrated “analysis-decision-control” framework oriented toward controllable probabilistic stability was proposed. First, leveraging observations from PMUs (phasor measurement units), AKDE (adaptive kernel density estimation) approach was employed to construct probabilistic small-signal models of key system states—such as frequency, voltage, and current—under wind power integration, enabling a refined characterization of non-Gaussian disturbance features in stochastic environments. Second, based on the results of probabilistic stability analysis, a Markov decision process was formulated with the output power as the system state and pitch angle regulation as the control action. ABERL (adaptive bayesian ensemble reinforcement learning) control strategy was then developed, in which multiple policies were adaptively fused through posterior inference to achieve self-adaptive pitch angle control of wind turbines, thereby directly optimizing the probabilistic small-signal stability of the system. Finally, the proposed method was validated on the IEEE 10-machine and 39-node typical system and the real-world Southern Power Grid system under PMU-driven small-disturbance scenarios, and was compared with representative control strategies including Actor-Critic, Monte Carlo methods and Q-learning. Simulation results and case study demonstrate that the proposed approach achieves superior performance in terms of disturbance response speed, pitch angle control stability, and probabilistic stability metrics. The system-level probabilistic stability analysis effectively identifies potential instability risks, while precise pitch angle regulation of local wind turbines significantly reduces wind farm control costs and enhances stable operation capability. The proposed probabilistic-driven control framework thus provides a viable solution for the secure and stable operation of power systems with high wind power penetration.