Deep Learning Models for Multi-Energy Prediction of Combined Electrical, Heat and Gas network systems

Abstract

Effective methods for predicting power consumptions of electrical, heat or gas systems are important for knowing in advance the state of the systems. Deep Learning (DL) models have been used extensively in predicting the electrical power consumptions and more recently for predicting heat or gas power consumptions. This paper presents Convolutional Neural Networks (CNNs) applied to individual prediction of multi-energy power consumptions of combined electrical, heat and gas systems. The CNNs are chosen because of their previous success and the designed models include three convolutional layers, with subsequent pooling and a fully connected layer for regression. The CNNs are trained and tested on real datasets of combined electrical, heat and gas systems. The predictions results are mainly evaluated in terms of the Signal to Noise Ratio (SNR) and the Normalized Root Mean Square Error (NRMSE). Single and combined input variables are being investigated and high correlation is identified between the present and the previous 24 hours power consumptions for any of the three studied systems electrical, heat or gas. The results demonstrate that the multi-energy loads can be predicted well for the testing datasets that is the electrical (SNR=34.63dB, NRMSE=0.0085), the heat (SNR=17.42dB, NRMSE=0.061) and the gas (SNR=8.86 dB, NRMSE=0.14) systems.