Abstract: To overcome the uncertainty of weather conditions and the nonlinear characteristics of high-dimensional photovoltaic power and environmental data, this paper proposes a combined photovoltaic power prediction model architecture based on TCN-Transformer-BiLSTM. Using high-dimensional photovoltaic and environmental data from the DKASC photovoltaic power plant in Australia for the year 2019 as the research subject, Pearson correlation analysis and normalization processing are first performed on the photovoltaic and environmental monitoring data. Then, a temporal-spatial convolutional network (TCN) is employed as the data spatial feature extraction layer, with the computation results fed into the Transformer module. Through its multi-head self-attention mechanism, the Transformer effectively captures global temporal-spatial relationships while enhancing the feature capture capability of the TCN; subsequently, a bidirectional long short-term memory network (BiLSTM) was used to calculate past and future information in the data to achieve the final photovoltaic power prediction. Through computational analysis, the results of the TCN-Transformer-BiLSTM model outperformed those of other combination models or single model structures.