【目的】 针对高渗透率分布式光伏并网引发配电网运营成本增大和电压越限风险加剧的问题，提出一种模型-数据联合、集中-分散结合的主动配电网日前-日内两阶段运行成本优化与电压越限控制方法。【方法】 第1阶段为模型驱动的日前有功-无功协调优化，建立计及电压安全约束的运行成本最优模型，协调源-网-荷-储各类可调资源，在优化运营成本的同时提升小时级电压安全水平。第2阶段为数据驱动的实时电压/无功分散控制：基于日前功率预测生成面向分钟级功率波动的多运行场景，通过潮流与无功网损优化计算，构建光伏与储能调压样本集；设计基于知识嵌入的自适应网络模糊推理系统的光伏和储能（battery energy storage system，BESS）电压控制器，利用样本集进行控制器学习以兼顾全局网损优化；各控制器依据本地实时电压和功率信息自适应决策最优无功出力，共同实现电压越限实时校正与系统网损优化。【结果】 仿真对比多种控制方法表明，所提两阶段方法能够有效降低系统运营成本，显著抑制实时电压越限，在运行经济性和电压安全控制方面均优于传统方法。【结论】 模型驱动与数据驱动相结合、集中协调与分散控制协同的调控架构，可充分发挥源网荷储的调节潜力，提升高比例分布式光伏接入配电网的运行经济性和电压安全性。

[Objective] To address the increased operational costs and voltage violation risks in distribution networks caused by high-penetration distributed photovoltaic integration， a two-stage day-ahead and intra-day operational cost optimization and voltage violation control method for active distribution networks is proposed， combining model-driven and data-driven approaches with centralized and decentralized coordination. [Methods] The first stage is a model-driven day-ahead active-reactive power coordinated optimization. A minimum operational cost model incorporating voltage security constraints is established to coordinate adjustable resources including sources， networks， loads， and energy storage systems， thereby optimizing operational costs while enhancing hourly voltage security. The second stage is a data-driven real-time voltage/var decentralized control. Based on day-ahead power forecasts， multiple operating scenarios reflecting minute-level power fluctuations are generated. A voltage regulation training dataset for photovoltaic （PV） and battery energy storage system （BESS） units is constructed through power flow and reactive power loss optimization calculations. Knowledge-embedded adaptive network based fuzzy inference system based voltage controllers are designed for PV and BESS units and trained using the dataset to incorporate global network loss optimization. Each controller adaptively determines the optimal reactive power output using local real-time voltage and power measurements， collectively achieving real-time voltage violation correction and system loss optimization. [Results] Simulation comparisons with multiple control methods demonstrate that the proposed two-stage method effectively reduces system operational costs and significantly mitigates real-time voltage violations， outperforming traditional methods in both operational economy and voltage security control. [Conclusions] The control architecture integrating model-driven and data-driven approaches， along with centralized coordination and decentralized control， can fully exploit the regulation potential of sources， networks， loads， and storage systems， thereby enhancing the operational economy and voltage security of distribution networks with high-proportion distributed photovoltaic integration.