Medium—And Long-Term Load Forecasting Method for Group Objects Based on the Image Representation Learning

Medium-and long-term load forecasting in the distribution network has important guiding significance for overload warning of distribution transformer, transformation of distribution network and other scenarios. However, there are many constraints in the forecasting process. For example, there are many predict objects, the data sample size of a single predict object is small, and the long term load trend is not obvious. The forecasting method based on neural network is difficult to model due to lack of data, and the forecasting method based on time sequence law commonly used in engineering is highly subjective, which is not effective. Aiming at the above problems, this paper takes distribution transformer as the research object and proposes a medium-and long-term load forecasting method for group objects based on Image Representation Learning (IRL). Firstly, the data of distribution transformer is preprocessed in order to restore the load variation in natural state. And then, the load forecasting process is decoupled into two parts: the load trend forecasting of the next year and numerical forecasting of the load change rate. Secondly, the load images covering annual and inter-annual data change information are constructed. Meanwhile, an Image Representation Learning forecasting model based on convolutional neural network, which will use to predict the load development trend, is obtained by using load images for training; And according to the data shape, the group classification of the data in different periods are carried out to train the corresponding group objects forecasting model of each group. Based on the forecasting data and the load trend forecasting result, the group forecasting model corresponding to the forecasting data can be selected to realize the numerical forecasting of load change rate. Due to the large number of predict objects, this paper introduces the evaluation index of group forecasting to measure the forecasting effect of different methods. Finally, the experimental results show that, compared with the existing distribution transformer forecasting methods, the method proposed in this paper has a better overall forecasting effect, and provides a new idea and solution for the medium-and long-term intelligent load forecasting of the distribution network.