Deep convolutional neural networks for short-term multi-energy demand prediction of integrated energy systems

Forecasting power consumptions of integrated electrical, heat and gas network systems is essential in order to operate more efficiently the multi-energy network system. Multi-energy systems are increasingly seen as a key component of future energy systems, and a valuable source of flexibility, which can significantly contribute to a cleaner and more sustainable integrated energy system. Therefore, there is a stringent need for developing novel and performant models for forecasting multi-energy demands of integrated energy systems, which to account for the different types of interacting energy vectors and of the coupling between them. Previous efforts in demand forecasting focused only on electrical power consumptions or, more recently, on the single heat or gas power consumptions. Therefore, in order to address the multi-energy demand forecasting problem, in this paper six novel prediction models based on Convolutional Neural Networks (CNNs) are developed, for either individual or joint prediction of multi-energy power consumptions: the single input/single output variable CNN model with determining the optimum number of epochs (CNN_1), the multiple input/single output variable CNN model (CNN_2), the single input/single output CNN model with training/validation/testing datasets (CNN_3), the joint prediction CNN model (CNN_4), the multiple-building input/output CNN model (CNN_5) and the federated learning CNN model (CNN_6). All six models are applied in a comprehensive manner on a novel integrated electrical, heat and gas network system, which only recently has started to be used for forecasting. The forecast horizon is short-term (i.e. next half an hour) and all the prediction results are evaluated in terms of the Signal to Noise Ratio (SNR) and the Normalized Root Mean Square Error (NRMSE), while the Mean Absolute Percentage Error (MAPE) is used for comparison purposes with other existent results from literature. The numerical results show that the single input/single output variable CNN model with training/validation/testing datasets (CNN_3) is able to equal the performances of the single input/single output variable CNN model with determining the optimum number of epochs (CNN_1), and to outperform the other four prediction models. The prediction accuracy of the multi-energy networks loads is shown to significantly depend on the level of non-linearity and scarcity existent in the input training dataset(s). Furthermore, this extensive multi-model study reveals that the characteristics (i.e. connections between the different networks, correlations between the different energy vectors) of the considered integrated energy system need to be explored when designing the CNNs prediction models.