To address the scheduling failures, power imbalances, and economic losses in virtual power plants (VPPs) caused by multisource uncertainties—including stochastic renewable generation, load fluctuations, and parameter deviations—this paper develops a multi-timescale adaptive dispatch framework incorporating multi-source uncertainty modeling and online parameter correction. The framework employs two-stage robust optimization for day-ahead scheduling to generate a robust pre-dispatch plan, and introduces a state-feedback mechanism in the intra-day stage, where an improved quantum-inspired genetic algorithm is used to recursively correct critical parameters, thereby forming a closed-loop dispatch structure. Simulation experiments validate the effectiveness of the proposed approach. Results show that, under significant forecasting errors in renewable generation and electro-thermal loads, the method improves actual operational revenue by approximately 3.2% compared to conventional deterministic dispatch. Moreover, the online parameter correction strategy reduces system balancing costs by nearly 90% during most time periods. The framework effectively balances robustness, economic efficiency, and adaptability, offering a viable technical pathway for the secure and economical operation of VPPs under high uncertainty