In power systems incorporating solar photovoltaic (PV) generation, accurate prediction of solar irradiance is essential to reduce energy costs and ensure power quality. Conventional irradiance prediction methods mainly involve analyzing the time-series variation of historical irradiance or studying the relationship between irradiance and environmental factors. In this paper, a BiLSTM model based on atmospheric conditions and cloud distribution is proposed for global horizontal irradiance prediction. In this method, key features of cloud cover and atmosphere are selected as important variables affecting irradiance. In order to verify the superiority of the proposed model, its prediction results are analyzed in comparison with those of the LSTM network model using several metrics. The results show that the proposed solar irradiance prediction method exhibits high accuracy on different time scales and is more reliable than the traditional time series prediction method.