Power load prediction method based on LSTM and LGBM

Abstract

The invention discloses a power load prediction method based on LSTM and LGBM, and belongs to the technical field of power load prediction, and the method comprises the following steps: carrying out the preprocessing; constructing an LSTM network containing a plurality of layers of neural units to obtain an LSTM model prediction value; calculating a cost function of the LSTM model; constructing anLGBM decision tree to obtain an LGBM model prediction value; calculating a cost function of the LGBM model according to the cost function of the LSTM model; and taking the negative direction of the model loss function gradient as a search direction, iteratively solving a target value by utilizing a gradient descent method, and calculating an average absolute percentage error MAPE. According to the method, the power load prediction model based on the LSTM long short-term memory neural network and the LGBM decision tree gradient improvement method is constructed, so that better fitting of powerdata is achieved, potential distribution information in existing data is fully mined, and the purpose of power load prediction tasks considering weather factors is accurately achieved.

Description

technical field [0001] The invention relates to a load forecasting method, in particular to a power load forecasting method based on LSTM and LGBM, and belongs to the technical field of power load forecasting. Background technique [0002] Electric power is the resource base for building a country and a nation. Accurate power load forecasting is of great significance for maintaining the stable operation of the power grid and formulating power dispatching plans. Due to the influence of many factors such as production level, population density, residents' demand, and climate change, power load forecasting has the characteristics of high randomness and difficult modeling. In recent years, it has become a hot topic of research by Chinese and foreign scholars. [0003] Traditional power load forecasting methods achieve the purpose of fitting short-term power data by constructing regression equations. For medium and long-term power forecasting tasks, the forecasting accuracy is lo...

AI Technical Summary

Problems solved by technology

However, the trend extrapolation method often uses the second-order adaptive coefficient method to determine the optimal parameters of the model. The derivation process needs to assign initial values ​​to each parameter, and the setting of the initial values ​​greatly disturbs the final result. Therefore, this method is subject to subjective experience. The impact is relatively large, which is not conducive to the generalization of the method

Images