[Objecttives] To enhance the stability of renewable energy systems and achieve accurate photovoltaic power prediction

this study proposes a hybrid prediction model integrating a multi-scale temporal convolutional network (MS-TCN)

a gated position-aware transformer encoder (GPATE)

and a bidirectional gated recurrent unit (BiGRU). [Methods]　The MS-TCN module is first employed to extract local features from multi-dimensional time series data. Then

the GPATE module is introduced

in which the traditional feedforward network is replaced by a position-aware multilayer perceptron

augmented with a gating mechanism to suppress noise and enhance the model's ability to capture long-term dependencies and global contextual information. Finally

the feature representation of MS-TCN and GPATE is fused

and BiGRU is used to construct bidirectional temporal dependencies to further strengthen the modeling ability of complex temporal dynamic features. [Results]　Compared with the Transformer model

the proposed model achieves significant improvements of 42.87% and 29.37% in RMSE and MAE

respectively. [Conclusions]　Through structural innovation

the proposed model effectively improves prediction accuracy and provides reliable support for power system scheduling under high penetration of renewable energy.