A Short Term Load Periodic Prediction Model Based on GBDT

Load forecasting is an important part of power system planning, and one of the important tasks of smart grid. It is of great significance for economically and rationally arranging the start and stop of internal grid generators, keeping the grid safe and stable. With the continuous development of power systems, short term load forecasting, especially the electricity consumption of independent power users, plays an increasingly important role in smart grid planning and operation. However, due to the high volatility and uncertainty of users' electricity consumption, it is quite difficult to predict the load of even an individual power user. The GBDT algorithm uses the addition model and the forward fractional algorithm to realize the learning optimization. It has good generalization ability and is suitable for solving regression problems similar to power load forecasting. For improving the load forecasting accuracy of independent power user, this paper proposes a model to automatically judge the load period and then perform GBDT fitting. Firstly, the user curve clustering is used to obtain different user types of power consumption and the corresponding user groups, and then the load period searching for each user group is used to obtain the minimum load period that can reflect the characteristics of the user's power consumption. Finally, the optimal parameters of GBDT regression are found according to the load period and then it trains the model. The experimental results show that this model has better results on short term load forecasting compared with the directly using of random forest and RBF kernel SVR model.