Short-term Energy Forecasting Research Articles

Short‐term energy forecasting using deep neural networks: Prospects and challenges

AbstractThis study presents an in‐depth overview of deep neural networks (DNN) and their hybrid applications for short‐term energy forecasting (STEF). It examines DNN‐based STEF from three perspectives: basics, challenges, and prospects. The study compares recent literature using metrics like mean absolute error (MAE), mean average percentage error (MAPE), and root mean square error (RMSE). Findings indicate that combining automated data‐driven models with enhanced DNNs effectively addresses forecasting challenges. It also highlights the role of DNNs in integrating energy prosumers, renewable energy systems, microgrids, big data, and smart grids to improve STEF.

A new approach to seasonal energy consumption forecasting using temporal convolutional networks

There has been a significant increase in the attention paid to resource management in smart grids, and several energy forecasting models have been published in the literature. It is well known that energy forecasting plays a crucial role in several applications in smart grids, including demand-side management, optimum dispatch, and load shedding. A significant challenge in smart grid models is managing forecasts efficiently while ensuring the slightest feasible prediction error. A type of artificial neural networks such as recurrent neural networks, are frequently used to forecast time series data. However, due to certain limitations like vanishing gradients and lack of memory retention of recurrent neural networks, sequential data should be modeled using convolutional networks. The reason is that they have strong capabilities to solve complex problems better than recurrent neural networks. In this research, a temporal convolutional network is proposed to handle seasonal short-term energy forecasting. The proposed temporal convolutional network computes outputs in parallel, reducing the computation time compared to the recurrent neural networks. Further performance comparison with the traditional long short-term memory in terms of MAD and sMAPE has proved that the proposed model has outperformed the recurrent neural network.

Short-Term Energy Load Forecasting Model with Sample Filtering

Short-term energy load forecasting is a crucial task in the power smart grid, which enables the power utilities to understand the future energy demands and plans to attain the demand and supply equilibrium, thereby optimizing power deployment and reducing power losses. Several techniques have been implemented to enhance energy load forecasting. However, the nonlinear nature of the data collected in the smart grid makes it difficult to attain 100% energy load forecasting accuracy. For instance, the Deep Feedforward Neural Networks model based on Input Attention Mechanism and Hidden Connection Mechanism has a mean absolute percentage error of 3.17%; model based on Sequence to Sequence Recurrent Neural Network with Attention had a mean absolute percentage error of 2.7%. The model based on Deep Recurrent Neural Networks with Levenberg–Marquardt backpropagation algorithm had a mean absolute percentage error of 0.58; and Deep Feedforward Neural Network with sample weights model had 3.22 % as root mean squared error. To improve energy load forecasting accuracy, this work proposed a model based on Deep Recurrent Neural Networks and sample filtering, which provides an exhaustive elucidation for modelling a sophisticated stochastic relationship between the input and output features. Deep Recurrent Neural Networks have proven to be good at modelling the nonlinearities in data of different fields and are mostly used in energy load forecasting to reduce forecasting error and a high degree of overfitting. Sample filtering is achieved through the use of K-Means clustering which determines the number of clusters to be used in the model. Findings from the study showed that by employing Deep Recurrent Neural Networks and sample filtering, the short-term energy load forecasting accuracy is improved in reference to mean absolute percentage error and root mean squared error of 0.31% and 1.014, respectively. As a result of the reduction in error, the energy demand and supply chain equilibrium are enhanced, thereby optimizing power deployment and reducing power losses.
 Keywords: Machine learning, Neural networks, Sample filtering, Smart grid, Short-term energy forecasting

A Novel Approach for Short-Term Energy Forecasting in Smart Buildings

Efficient energy management is required for optimal energy consumption. The building sector consumes 40% of the total global energy production and is expected to reach 50% by 2050. With the soaring price of electricity, buildings need economical and efficient energy management. Recent advances in artificial intelligence and the Internet of Things (IoT) have inspired researchers working in smart building management to harness the potential of these technologies for forecasting energy consumption in smart buildings. This article proposes a novel hybrid deep learning model consisting of convolutional neural network (CNN) and recurrent neural network (RNN) to predict hourly energy consumption for smart buildings. Experimental results demonstrate that the CNN-gated recurrent unit (GRU) model, with an accuracy of 97%, outperforms the state-of-the-art techniques.

An Accurate Dynamic Forecast of Photovoltaic Energy Generation

The accurate forecast of the photovoltaic generation (PVG) process is essential to develop optimum installation sizing and pragmatic energy planning and management. This paper proposes a PVG forecast model for a PVG/Battery installation. The forecasting strategy is built on a Medium-Term Energy Forecasting (MTEF) approach refined dynamically every hour (Dynamic Medium-Term Energy Forecasting (DMTEF)) and adjusted by means of a Short-Term Energy Forecasting (STEF) strategy. The MTEF predicts the generated energy for a day ahead based on the PVG of the last 15 days. As for STEF, it is a combination between PVG Short-Term (ST) forecasting and DMTEF methods obtained by selecting the least inaccurate PVG estimation every 15 minutes. The algorithm results are validated by measures taken on a 3 KWp standalone PVG/Battery installation. The proposed approaches have been integrated into a management algorithm in order to make a pragmatic decision to ensure load supply considering relevant constraints and priorities and guarantee the battery safety. Simulation results show that STEF provides accurate results compared to measures in stable and perturbed days. The NMBE (Normalized Mean Bias Error) is equal to −0.58% in stable days and 26.10% in perturbed days.

Novel short-term solar radiation hybrid model: Long short-term memory network integrated with robust local mean decomposition

Data-intelligent algorithms tailored for short-term energy forecasting can generate meaningful information on the future variability of solar energy developments. Traditional forecasting methods find it relatively difficult to obtain a reliable solar energy monitoring system because of the inherent nonlinearities in solar radiation and the related atmospheric input variables to any forecasting system. This paper proposes a new artificial intelligence-based hybrid model by employing the robust version of local mean decomposition (RLMD) and Long Short-term Memory (LSTM) network denoted as RLMD-LSTM. The objective model (i.e., RLMD-LSTM) is built near real-time, half-hourly ground-based solar radiation dataset for the solar rich, metropolitan study sites in Vietnam with all of the forecasting results being benchmarked through classical modelling approaches (i.e., Support Vector Regression SVR, Long Short-term Memory LSTM, Multivariate Adaptive Regression Spline MARS, Persistence) as well as the other alternative hybrid methods (i.e., RLMD-MARS, RLMD-Persistence and RLMD-SVR). Verified by statistical metrics and visual infographics, the present results demonstrate that the proposed model can generate satisfactory predictions, outperforming several counterpart methods. The predictive performance is stable for all study sites that the root-mean-square error remained profoundly lower for RLMD-LSTM (19–20%) compared with RLMD-MARS (20–21%), RLMD-SVR (29–35%), RLMD- Persistence (29–51%), LSTM (25–48%), MARS (21–51%) and SVR (23–85%), Persistence (29–51%). The Legates and McCabe’s Index, yielding a value of approximately 0.7988–0.9256 for RLMD-LSTM compared with 0.765–0.8142, 0.4917–0.5711, 0.6900–0.7482, 0.6914–0.7646, 0.4349–0.7170 respectively, for the RLMD-MARS, RLMD-SVR, RLMD-Persistence, LSTM, MARS, SVR, Persistence models, also confirms the outstanding performance of RLMD-LSTM model. Accordingly, the study ascertains that the newly designed approach can be a potential candidate for real-time energy management, renewable energy integration into a power grid and other decisions to optimise the overall system's scheduling and performance.

Ensemble Machine Learning Modelling for Medium to Long Term Energy Consumption Forecasting

Electricity demand forecasting is an important research area that is gaining popularity among researchers these days. Forecasting energy consumption is a critical task as it affects the overall functioning of the power industry. There are many types of research already done in the field of short-term energy forecasting, but there is a scarcity of researches in medium to long-term energy forecasting. This paper focuses on medium to long-term energy forecasting using machine learning and the ensemble approach. The machine learning methods include Linear Regression (LR), Random Forest (RF), Least Absolute Shrinkage and Selection Operator (LASSO), and Gradient Boosting Regressor (GBR). For the comparative analysis the performance metric selected are R2, Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE)..

Characterizing marine atmospheric boundary layer to support offshore wind energy research

The marine atmospheric boundary layer (MABL), under most wind conditions, is characterized by lower shear, less turbulence, and higher winds than its terrestrial counterpart, thereby making offshore wind farms attractive enough to offset higher costs of construction and maintenance. Significant questions, however, remain about the structure and characteristics of the MABL and its impact on wind turbines. This work characterizes the structure of the MABL in the coastal region south of Martha’s Vineyard where offshore wind turbines are planned to operate, using in situ and remotely sensed LIDAR measurements and numerical data generated by high resolution Weather Research and Forecasting (WRF) modeling. This work will benefit wind power estimates and short-term energy forecasting.

A Combined Methodology of Adaptive Neuro-Fuzzy Inference System and Genetic Algorithm for Short-term Energy Forecasting

1 Abstract—This document presents an energy forecast methodology using Adaptive Neuro-Fuzzy Inference System (ANFIS) and Genetic Algorithms (GA). The GA has been used for the selection of the training inputs of the ANFIS in order to minimize the training result error. The presented algorithm has been installed and it is being operating in an automotive manufacturing plant. It periodically communicates with the plant to obtain new information and update the database in order to improve its training results. Finally the obtained results of the algorithm are used in order to provide a short- term load forecasting for the different modeled consumption processes. I. INTRODUCTION With the continuously growing demand of the energy, it is getting more important to develop systems capable to optimize the energy use. Energy management is nowadays a subject of great importance because of the facing problems of the global warming and oil shortage. In the industrial sector, the energy management systems have focused so far on the monitoring and off-line management of energy as it is outlined in (1). The typical energy management systems are based on the collection of information about the plant's operation using energy meters. Those systems help to monitor the operation of the installations, collect data and generate reports to identify the possible critical points of the consumptions. However, intelligent systems can improve the operation of the energy management systems, offering further functionalities such as predictive maintenance, energy optimization, fault diagnosis and energy forecast. Different approaches for energy savings and energy prediction have been studied over the past years. Evolutionary algorithms such as Particle Swarms (PSO), Gravitational Search Algorithms (GSA) and Simulated Annealing (SA) have successfully implemented in optimization and control applications (2). An implementation of a model based on Artificial Neural Network (ANN) was presented in (3-4) and (5) in order to estimate the load forecast in an electrical distribution system while in (6) a comparison between ANN and Fuzzy Logic is made on applications of short-term and medium-term load forecasting. An application of Neuronal Networks (NN) is presented in (7) in which it faces Multi-Input-Multi-Output (MIMO) applications with single input and output (SISO) net works. An ANFIS implementation for energy prediction of regional electrical loads in Taiwan was presented in (8), comparing its performance with other similar techniques (i.e., regression models, ANN-based models, Genetic algorithms and hybrid ellipsoidal fuzzy systems). A cellular multi-grid genetic algorithm is presented in (9) to face balancing problems in assembling lines. Techniques based on cultural algorithms are presented in (10) to resolve complex mechanical design optimization problems in an efficient and effective method. Thi s document presents the modeling and prediction algorithms that were developed in order to generate customizable mathematical models for different consumptions, as a way to improve the operation of a general energy monitoring system. The paper is organized as follows: section II describes an overview of the algorithm that has been used for the model's training and the energy forecast while section III outlines a brief explanation about the Genetic Algorithm's operation. In section IV, the proposed methodology is presented explaining the combination of the two algorithms in order to develop a system capable to train the consumption models autonomously. In Section V, the implementation of the system in the pilot plant is explained, presenting the different results that have been obtained during the test and the evaluation of the system. Finally, section VI summarizes the paper and discusses the different conclusions.

Artificial neural networks for short-term energy forecasting: Accuracy and economic value

Sixteen electric utilities surveyed state that use of ANNs significantly reduced errors in daily electric load forecasts, while only three found otherwise. Data for five gas utilities reinforces this result: the mean absolute percentage error (MAPE) for ANN daily gas demand forecasts was 6.4%, a 1.9% improvement over previous methods. Yet ANNs were not always best, implying opportunities for further improvement. The economic value of error reduction for electric utilities was assessed by examining operating decisions. For 19 utilities surveyed, an average of $800 000/year per utility is estimated to be saved from use of ANN-based forecasts. Most benefits resulted from improved generating unit scheduling; the utilities estimated such benefits to be up to $143 annually per peak MW of demand for each 1% improvement in MAPE. This estimated worth of accuracy improvement (roughly 0.1% of annual generation O&M costs) is confirmed by solving generation scheduling and dispatch models under various levels of forecast accuracy.

Short-run forecasting of highway gasoline consumption in the United States

Forecasting highway consumption of gasoline can be expected to gain greater attention in future years as problems related to energy uncertainties and growing fuel tax revenue shortfalls continue to mount. Although much of the energy forecasting conducted to date has involved fairly large-scale econometric modeling, the immediacy of energy concerns warrants closer attention to near-term time series analyses of gasoline consumption patterns. This paper presents an autoregressive-integrated-moving average (ARIMA) model for forecasting monthly highway energy consumption in the United States using data from the 1977–83 period. Distinct secular and seasonal factors, revolving around 1- and 6-month cycles, emerged from the time series analysis that offer a practical basis for short-term energy forecasting. Efforts to incorporate the interventing effects of the 1979 oil supply disruption on U.S. highway gasoline consumption found that its impacts on the underlying secular, seasonal and stochastic processes behind usage patterns were fairly inconsequential. Comparable short-range models should be integrated into state-level planning programs to closely track possible energy futures at more disaggregate levels.