Forecasting-to-scheduling in virtual power plants: Optimization methods, EV/V2G integration, and demand-response orchestration — A strategic review

The rapid growth of variable renewable generation and electrified end uses is reshaping power-system operations and amplifying uncertainty across distribution-level portfolios. Virtual power plants (VPPs) address these challenges by coordinating distributed energy resources (DERs), storage, electric vehicles (EVs), and flexible demand to deliver market-facing energy and ancillary services. This review focuses on the methods that determine VPP performance forecasting, optimization, EV integration, and demand-response (DR) orchestration and maps these methods to service requirements and measurable outcomes (e.g., accuracy, latency, feasibility, robustness, and revenue). We first synthesize forecasting approaches for VPP decision support, covering load/renewable/price prediction using modern deep and hybrid models (e.g., N-HiTS, CNN–BiLSTM, KAN) alongside probabilistic and risk-aware forecasting. Particular attention is given to uncertainty representation (e.g., mixture-density models and recurrent probabilistic variants) and to practical trade-offs among data availability, compute budgets, and real-time latency. We then review optimization and scheduling frameworks, ranging from deterministic dispatch to stochastic and robust formulations, receding-horizon/model-predictive control, and two-stage day-ahead/real-time coordination. Scalability is addressed via decomposition and dual/Lagrangian methods for large, heterogeneous portfolios. Next, we examine DR and flexibility-market participation, including program targeting, baseline construction, and settlement, and provide a method-to-service mapping that links algorithm classes to products such as energy, reserves, fast frequency response, and Volt/VAR support. EV integration is treated as a central flexibility lever through V2G/G2V coordination, mobility-aware fleet modeling, charging-network constraints, and degradation-aware scheduling. Finally, we outline the cyber–data stack needed to deploy these methods (IoT/edge–cloud pipelines, streaming analytics, standardized telemetry/APIs, and cybersecurity), and propose benchmarking guidance and a comparative selection matrix. We conclude with open research directions including digital twins, degradation-aware valuation, multi-agent coordination, and trusted settlement mechanisms for emerging flexibility trading.