Research on Photovoltaic Generation–Battery Energy Storage System Power Forecasting and Coordinated Scheduling Strategy Based on Transformer

With the increasing penetration of photovoltaic generation, accurate PV power forecasting and robust PV–battery energy storage system scheduling are essential for improving operational reliability and reducing grid electricity procurement. This paper proposes a hybrid framework named DCT–CA–TCN–Transformer, where the discrete cosine transform extracts multiscale frequency components, channel attention reweights informative features, temporal convolutional networks capture local‐to‐mid‐range dynamics via dilated causal convolutions, and a Transformer encoder models long‐range dependencies. On a real PV–BESS dataset, the proposed model achieves a coefficient of determination ( R 2 ) of 0.998141, an MAE of 12.115933, and an RMSE of 20.743818, reducing MAE and RMSE by 71.1% and 72.2% compared with the TCN‐Transformer baseline. Based on the forecasts, an uncertainty‐aware PV–storage dispatch strategy is developed by generating multiscenario PV trajectories via Monte Carlo sampling of forecast errors and enforcing BESS operational constraints. Compared with deterministic scheduling, UAPSD improves PV utilization to 94.6%, increases BESS utilization to 82.7%, and reduces total grid electricity procurement from 430 kWh to 340 kWh, demonstrating improved robustness and economic performance.