A data-driven dynamic reserve capacity allocation method considering multi-meteorological factors

To address the uncertainty of net load prediction errors influenced by multiple meteorological factors under high renewable energy penetration, this paper proposes a data-driven dynamic reserve capacity allocation method. First, by pre-filtering and efficiently selecting historical samples with similar meteorological conditions using a locality-sensitive hashing (LSH) algorithm, the probability distribution of net-load forecast errors driven by meteorological conditions is constructed. Subsequently, the probability distribution of prediction errors is constructed using non-parametric Kernel Density Estimation (KDE). Second, an optimization model aiming to minimize total system costs is formulated. By incorporating economic factors such as time-of-use prices, load shedding costs, and renewable energy curtailment penalties, the method achieves a balance among economic efficiency, reliability, and low-carbon objectives in reserve allocation. Furthermore, the computational efficiency of the optimization model is significantly enhanced through convex combination modeling and the introduction of screening variables. Simulation results demonstrate that the proposed method exhibits strong adaptability across various meteorological conditions and prediction error scenarios. Compared with conventional methods relying solely on load power or aggregated historical data, the proposed approach significantly reduces total system costs while maintaining a high level of reserve coverage reliability.