Day ahead load forecasting using long and short term TCN

• 1、School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, 100876

Abstract：In recent years, the smart grid has developed rapidly and the marketization of power has continued. Short-term load forecasting has played an important role in applications such as real-time power price setting, contract power distribution, and system adequacy assessment. Load can be divided into individual household load and aggregate load of many users. Compared with the latter, the electricity consumption behavior pattern of individual household load is more unstable and more random, so it is more difficult to predict. This article focuses on a scenario of short-term personal load forecasting, day-ahead load forecasting. This paper proposes a novel prediction model based on deep neural network, which uses the popular temporal convolutional network instead of the traditional recurrent neural network. The long-term time series mode and the periodic short-term time series mode of the integrated sequence are used to extract the time series characteristics of the two respectively with TCN, and the prediction results are output after fusion. On the public data set, the effect of the model in this article and the comparison model such as LSTM and TCN are verified. The experimental results show that this model has achieved the best prediction effect.

Keywords： Time series, Neural Network, Residential Load Forecasting