Performance evaluation of deep learning architectures for load and temperature forecasting under dataset size constraints and seasonality