Deep Learning-Based Short-Term Load Forecasting for Supporting Demand Response Program in Hybrid Energy System

Abstract

A novel method for short-term load forecasting (STLF) is proposed in this paper. The method utilizes both long and short data sequences which are fed to a wavenet based model that employs dilated causal residual convolutional neural network (CNN) and long short-term memory (LSTM) layer respectively to hourly forecast future load demand. This model is aimed to support the demand response program in hybrid energy systems, especially systems using renewable and fossil sources. In order to prove the generality of our model, two different datasets are used which are the ENTSO-E (European Network of Transmission System Operators for Electricity) dataset and ISO-NE (Independent System Operator New England) dataset. Moreover, two different ways of model testing are conducted. The first is testing with the dataset having identical distribution with validation data, while the second is testing with data having unknown distribution. The result shows that our proposed model outperforms other deep learning-based model in terms of root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In detail, our model achieves RMSE, MAE, and MAPE equal to 203.23, 142.23, and 2.02 for the ENTSO-E testing dataset 1 and 292.07, 196.95 and 3.1 for ENTSO-E dataset 2. Meanwhile, in the ISO-NE dataset, the RMSE, MAE, and MAPE equal to 85.12, 58.96, and 0.4 for ISO-NE testing dataset 1 and 85.31, 62.23, and 0.46 for ISO-NE dataset 2. Energies 2019, 12, 3359 2 of 16users' load demand. The major concern of this scheme is the cost to be borne by users once the renewable sources cannot supply adequate power to the grid in which they will pay an expensive fossil-based electric price. Moreover, fossil energy tends to gradually increase leading to economic conflict in society [2]. Therefore, in order to tackle this issue, an appropriate demand response scheme can be applied.Demand response is the change of electric usage by users due to change of electric price or maybe an incentive as a reward of lowering their power consumption [3]. Applying demand response to this hybrid system is very beneficial for shaving peak load demand [4,5], leading to the reduction of fossil energy consumption. Moreover, it can provide short-term impact and economic benefit for both consumer and utility.In order to support this demand response, short-term load forecasting (STLF) is very important for predicting whether the energy storage from renewable sources is able to handle the forthcoming power consumption or not. If the prediction states that the storage is not adequate to support the future load, then the electricity utility can announce this situation to the users, which eventually triggers them to reduce their electric usage, because users do not only want to pay more for conventional energy source but also want to get incentives from the authorities.Fortunately, with the help of developed infrastructures like smart meters equipped with a lot of sensors and the Internet of Things (IoT), a robust STLF method is feasible to...

Multiple Wavelet Convolutional Neural Network for Short-Term Load Forecasting

“…Our single model and integrated model are better than the other methods. Specifically, compared to the method proposed by Pramono et al [16], our single model improves the MAPE by 21.74%, the MAE by 21.84%, and the RMSE by 24.2%; our ensemble model improves the MAPE by 30.4%, the MAE by 29%, and the RMSE by 29.5%.…”

“…In [15], a short-term load forecasting method based on a deep residual network was proposed, and the generalization ability of the model was improved through a two-stage integration strategy. In [16], a WaveNet based on a dilated causal residual convolutional neural network (CNN) and long short-term memory (LSTM) layers was proposed for load prediction. In [18], a multilayer LSTM network was used for load prediction.…”

“…Specifically, compared to WT-ELM-MABC, our single model improves the MAPE by 20% the MAE by 16.81%, and our integrated model improves the MAPE by 26.7% and the MAE by 24.5%. We will also compare the proposed method with the methods of [38], [39], [16], where the training set ranges from January 1, 2004, to December 31, 2005, and the last month of the dataset is used as the validation set and the test set. The range is from May 1, 2006, to May 31, 2006 The experimental results are shown in Table 5.…”

“…Although many neural network-based methods are currently used for load prediction [11], [15], [16], these methods do not provide any explanation for the results of load prediction. In this use case, we use the integral gradient [25] to explain that our model relies on that input to make the final predictions.…”

“…At present, many neural network-based load prediction methods have been extracted and obtained excellent results [9], [10], [11]. Benefiting from the success of deep learning [12], [13], some researchers have achieved better prediction performance by adopting advanced network structures and deep networks [14], [15], [16].…”