Household load forecasting based on CNN ensemble and non-uniform quantization
Multi-source uncertainty and unbalanced power distribution of load are the important factors re-stricting the accuracy of household load short-term forecasting.This paper proposed to use non-uniform quantization to solve the problem of large quantization errors caused by skewed power distribution and the problem of high-power approximation to"anomalous"samples;and proposed a combination of convolutional neural network(CNN)and ensemble learning to tackle the complex load patterns resulting from the uncer-tainty of multiple sources,and proposed a combination of CNN and ensemble learning to tackle the complex load patterns resulting from the uncertainty of multiple sources.Firstly,the framework of household load short-term forecasting was given.Subsequently,a μ-law is given to convert the raw load to approximately normally distributed,and multiple data affecting the load are interwoven into grayscale images in order to extract the deep nonlinear relationships implied between the feature data.Then,for short-term prediction of household load,CNN basic learner and ensemble learning algorithm of Adaboost coordinated multiple CNNs were designed in detail.The load forecasting tests on actual households in different temperature zones with 1 hour advance show that the indexes of MAPE,MSE,RMSE and MAE of the method in this paper are better than the existing advanced forecasting methods.This method can provide high-precision short-term energy consumption data for the scheduling management and optimization control of utility and users,steps are as follows.
万方数据
维普
