Short-term Bus Load Prediction Based on Similar Day Clustering of Hierarchical Analysis and Long Short Term Memory Network

With the rapid development of national economic marketization, the standard for judging the reliable operation of the power grid is getting higher and higher. As a real-time operation system, the particularity of massive storage, so it is particularly important to maintain the balance between supply and demand of the power system. When doing the short-term load prediction, we need to take into account that the load fluctuations within one day are affected by many factors. In addition, there is a certain periodicity in the total daily load per week.When making the short-term load prediction, we need to fully consider the factors such as temperature, humidity, rainfall and daily type. The load characteristic of the same bus bar is different on different dates. This paper presents a model for load prediction based on LSTM and hierarchical clustering algorithms. Then, based on the average daily load and clustering results of meteorological data, build datasets based on different day types. Finally, LSTM models are used to predict different types of daily type loads. The proposed method has higher accuracy than conventional methods in predicting load fluctuations in one day in high temperature or heavy rainfall weather.