Enhanced Multivariate Time Series Regression Prediction Using Temporal Convolution Networks

Multivariate time series data, which has more complex and multidimensional characteristics compared to univariate data, is used in fields such as urban transportation, power grids, and financial markets. Accurately capturing information and dependencies among multiple time series is crucial for future time variable regression prediction. In this paper, we propose a novel method that utilizes Temporal Convolution Network (TCN) model to fuse multiple convolutional layers and residual connections. Experimental results on the energy consumption dataset from the University of British Columbia (UBC) show that our method stable performance across datasets and outperforms LSTM and ARIMA in some cases.