Best practices in China Southern Power Grid competition of AI short-term load forecasting

China Southern Power Grid (CSG) hosted a competition of AI short-term load forecasting with real-world load data in Sept 2019. NARI Technology was invited for the 10 days closed match and got the 2
^nd
prize among 28 top players in China, such as research institutes, universities, and system suppliers. This paper highlights the best practices we did in the competition: feature selection of short-term load forecasting, outlier data preprocessing for both load data and meteorological data, multiple head long short term memory (LSTM) neural network innovation and data augmentation with the benchmark. opportunities for improvements are considered as well.