A Novel Graph Based Sequence Forecasting Model for Electric Load of Campus

In this study, a new power prediction system named Pattern Similarity Graph based Model (PSGM) based on time sequential prediction algorithm and graphical neural network (GNN) structure is proposed. The proposed GNN architecture extracts the deep information of each power node and is used to predict the short-term power by using the sequential convolutional network (TCN) based on serial model. In the study, the electric junction is the high-tech campus. The number of enterprises in the electric campus, the total area of the campus, the time of operation of the campus and the daily average temperature data were selected as the inherent attributes of the electric campus. After constructing the graph model by taking the similarity of the power sequence of the campus as the correlation between the campuses, the feature vector of depth information of each node is combined with the feature vector of its sequence for the final prediction of short-term power consumption. The experiment compares the results of adding temperature conditions and obtains the influence degree of temperature on electricity consumption. The results are compared with baseline regression algorithm, LSTM, and ARIMA. The experimental results show that this method is superior to the traditional regression algorithm, and reveals the effective results with its competitive performance.