This paper proposes a renewable energy prediction method adapted to the climate characteristics of plateau mountains and oceans in the southern region. The proposed approach contains four parts: data clearing, time-period data clustering, modeling for renewable forecasts, and feature-prediction accuracy correlation matrix calculation. Therein, the forecast algorithm is improved mainly from two aspects: 1) Time-period clustering for data is implemented before training the forecast models. Different forecast models are trained with clustered data sets for achieving overall higher performance in forecasts compared with the one directly trained based on the whole data set. 2) We analyze the impact of complicated climate features, such as plateau mountains and ocean climate, on model prediction results, and select critical climate features for different prediction tasks based on the feature-prediction accuracy correlation matrix, indicating great improvements in renewable energy prediction.