Energy Demand Forecasting Research Articles

A flexible potential-flow model based high resolution spatiotemporal energy demand forecasting framework

Understanding urban demand profiles is an important determinant for energy dispatch and the optimization of the electric energy supply. For the design of the energy supply system, an important consideration is, to express the characteristics of urban household energy demand as a function of space and time. However, the focus of most research activities is only on the modeling of time series data. High-resolution forecasting models for the spatial–temporal distribution of energy were rarely reported in current literature. In this paper, we propose a spatio-temporal forecasting model based on potential-flow for urban energy demand forecasting. Compared with previous studies, potential-flow can describe energy migration in space with a high resolution. Based on the orientation of vectors, our model can predict the direction and intensity of spatial migrations in energy demand and identify energy transfer events. Extensive experiments on real-world data sets verify that our approach can achieve a better prediction accuracy compared with traditional methods. In further empirical studies, we find that the temporal electricity demand flow shows locally concentrating behavior for different regions of the city. In addition to temporal factors such as seasons, peaks and valleys, such clustering behavior also depend on local populations and major industries (financial, commercial, residential, etc.). Finally, we use entropy to quantitatively describe the intensity of this clustering phenomenon and explore its relationship with meteorological factors. Our research demonstrates a unified visual prediction approach to support exploratory demand analysis. We anticipate that the process will be expanded to support more forms of energy in the future.

Dual-stage attention-based long-short-term memory neural networks for energy demand prediction

Forecasting energy demand of residential buildings plays an important role in the operation of smart cities, as it forms the basis for decision-making in the planning and operation of urban energy systems. Deep learning algorithms are commonly used to reliably predict potential energy usage since they can overcome the issue of dependency on long-distance data in energy forecasting relative to the standard regression model. However, there are still two problems to be solved for energy forecasting, including the encoding of categorical characteristics and adaptive extraction of the most relevant characteristics for the use in predictions. To address the problems, we proposed a sequential forecasting model for medium- and long-term energy demand forecasting based on an embedding mechanism and a two-stage attention-based long-term memory neural network. An empirical study was conducted on three years of daily electricity consumption data from the residential buildings of the Pudong district of Shanghai to evaluate the model. The results show that the model can effectively extract the key features that are highly correlated with energy consumption dynamics by employing long-term dependencies in time series. In addition, the hybrid model outperforms others in terms of long-term forecasting capability. This paper also discusses future research directions and the possibilities for applying deep learning techniques in the energy sector.

Data centre day-ahead energy demand prediction and energy dispatch with solar PV integration

This paper presents a novel Marine Predators Algorithm for both training an Artificial Neural Network model used for predicting the energy demand and for solving a dynamic combined economic and emission dispatch of a data centre. The MPA is proposed for first finding the optimal weights and biases of the neural network based on a Mean Squared Error and Mean Absolute Error minimization objective function. Real life dataset obtained from an anonymous data centre operator in Cape Town, South Africa was used for the model implementation. The dataset was made up of a total of 564 samples and was split into training and testing set using an 80:20 ratio. The input variables contained in the dataset are the data centre’s ambient temperature, ambient relative humidity, chiller output temperature and Computer Room Air Conditioning supply temperature while the energy demand is the target variable. The optimal weights of the neural network model were analysed using a weights-based approach to determine the level of influence each input parameter of the model has on the data centre’s energy demand. Then based on the predicted energy demand of the data centre, a dynamic economic and emission dispatch problem is solved for the building while considering thermal and solar photovoltaic generations. A spinning reserve is also incorporated in the energy dispatch model to cater for any shortfall that may exist between the predicted and actual energy demand of the data centre due to possible inaccuracies in the energy demand prediction model. Results for the energy demand prediction task showed that the proposed method outperformed the state-of-the-art by producing the least Mean Squared Error, Root Mean Squared Error, Mean Absolute Error, Mean Absolute Percentage Error and highest prediction accuracy for the training and testing sets. Further analyses also highlighted that the data centre’s ambient temperature has the highest influence of about 37.63% on its energy demand pattern. For the energy dispatch task, the proposed method also identified solar photovoltaic as the preferred energy source over conventional thermal generators in fulfilling the objective function, depending on its availability. Overall, the findings presented in this study emphasize the robustness of the proposed method in solving the problems considered and its potential application towards solving even more complex problems.

Narrowing uncertainties in forecasting urban building energy demand through an optimal archetyping method

This paper aims at indicating and certifying the implemented framework for forecasting buildings' energy demand of the city of Bologna, Italy. The method is developed through an automated calibration and is based on 7 known, physics-based building parameters and 6 unknown, and highly uncertain variables. The proposed method focuses on reducing computing time while keeping the accuracy of the output by narrowing the uncertainties in predicting unknown parameters. To accomplish this task, 11 archetypes are defined which are representatives of the buildings in a specific neighborhood in Bologna, Italy. For every defined archetype, the most informative unknown variables are recognized and the Gaussian Process (GP) is employed to emulate the variable-to-data map. A wide sampling of the GP outputs is then applied by No-U-Turn Sampler (NUTS). The methodology is validated for 1156 Italian urban buildings based on the city database. The level of evaluation metrics demonstrates no bias in the output of the long-term forecasting while it accelerated the prediction of building energy demand and calibration on the city scale. The method is flexible for application in other contexts and various available urban datasets.

Scenario Analysis of the Low Emission Energy System in Pakistan Using Integrated Energy Demand-Supply Modeling Approach

Pakistan’s dependence on energy imports, inefficient power generation and distribution, and lack of planned investment have made the country’s economy vulnerable. Low carbon and resilient climate development in Pakistan can help to ensure climate action and reduce the chronic energy deficit ailing the country’s economy, society, and environment. This study focuses on developing and applying an integrated energy supply-demand modeling framework based on a combination of microeconomics and system integration theories, which can be used to address policies that could dramatically change the future course of Pakistan toward a low emission energy system. The methodology involves medium-term forecasting of energy demand using an integration of top-down and bottom-up modeling approaches. The demand-side model is interlinked with a bottom-up technology assessment supply model. The objective of the supply-side model is to identify the optimal combination of resources and technologies, subject to satisfying technical, institutional, environmental, and economic constraints, using the cost minimization approach. The proposed integrated model is applied to enable a complete perspective to achieve overall reductions in energy consumption and generation and better analyze the effects of different scenarios on both energy demand and supply sides in Pakistan. The results revealed that, in the baseline case, the energy demand is expected to increase from 8.70 Mtoe [106.7 TWh] to 24.19 Mtoe [297.2 TWh] with an annual average growth rate of 6.60%. Increasing the share of renewable energy power generation by 2030 can help to reduce emissions by 24%, which is accompanied by a 13% increase in the total cost of power generation.

Modelling the Future of the Baltic Energy Systems: A Green Scenario

Abstract The electricity sector in Europe and in the world is undergoing rapid and profound changes. There is a sharp increase in the capacity of renewable energy sources, coal and nuclear power plants are being closed and new technologies are being introduced. Especially rapid changes are taking place in the energy systems of the Baltic States. Under these conditions, there is an emerging need for new planning tools particularly for the analysis of the power system properties in a long-term perspective. The main contribution of this article lies in the formulation and solution of optimization problems that arise when planning the development of power systems in the Baltic States. To solve this problem, it is necessary to use models of various power plants and make a number of assumptions, the justification of which requires the following actions: to briefly review the current situation of the production and demand of energy in the Baltic power systems; to conduct an overview of the Baltic interconnections and their development; to make forecasts of energy prices, water inflow, energy production and demand; to set and solve the problems of optimization of power plant operation modes; to demonstrate the possibility and limitations of the developed tools on the basis of real-life and forecast data. In this paper, a case study is performed using the main components of the overall modelling framework being developed. It focuses on the Baltic power systems in 2050 under the conditions of significant expansion in the installed capacity of renewable energy sources (RESs) and diminished fossil fuel power plant activity. The resulting electricity generation mix and trade balance with neighbouring countries is assessed, showing that even with significant RES expansion, the Baltic countries remain net importers and because of the intermittency of RESs, there are hours within the year when the demand cannot be met.

DEVELOPMENT OF A SHORT-TERM PREDICTION SYSTEM FOR ELECTRICITY DEMAND

This article describes the development of a prediction method for the demand for electrical energy of a marketer's customer portfolio. The project is motivated by the economic benefit produced when the entity has accurate estimates of energy demand when buying energy in an electricity auction. The developed system is based on time series analysis and machine learning. As this system was part of a real-world project with data from a real environment, the article focuses on practical aspects of the design and development of system of these characteristics, such as the heterogeneity of data sources, and the delay in data availability. The predictions obtained by the developed system are compared with the results of a simple method used in practice. Keywords: energy demand prediction, electric power, machine learning, data-driven prediction.

Sequence-to-sequence deep learning model for building energy consumption prediction with dynamic simulation modeling

This study used a more detailed dataset approach to address the limitations of existing building energy prediction methods and to predict building energy demand more accurately. The EnergyPlus dynamic simulation program closely modeled building envelope performance, zone division, and heating, ventilation, and air conditioning systems. Building energy simulation used actual weather data and generated occupancy data. The occupancy, lighting, and equipment schedules for each zone were generated in 5-min intervals using the Lawrence Berkeley National Laboratory occupancy simulator. Summer electric power consumption based on Internet of Things information from the testbed building validated the model. When applying generated occupancy, lighting, and equipment schedules, the mean bias error of simulation similarly improved to 4.73%, and the coefficient of variation of the root-mean-squared error (Cv(RMSE]) improved to 12.26%. Subsequently, a demand prediction model was constructed as a sequence-to-sequence (seq2seq) model using long short-term memory (LSTM) cells in recurrent neural network (RNN) algorithms, then its accuracy was evaluated. In the seq2seq model, the learning performance based on the EnergyPlus data exhibited an RMSE of 4.48% and a weighted average percentage error of 3.07%. As a result of applying prediction methods while changing climate scenarios, prediction performance also satisfied statistically significant levels. The occupancy information and solar radiation were determined to exert the greatest influence on the building energy demand prediction. • The occupant schedule for each zone was generated in 5-min intervals using the LBNL occupancy simulator. • When applying the generated internal schedules, the MBE improved to 4.73%, and the Cv(RMSE) improved to 12.26%. • In the seq2seq model, the learning performance based on the EnergyPlus data exhibited an RMSE of 4.48% and a WAPE of 3.07%. • The occupant information and solar radiation were determined to exert the greatest influence on energy demand prediction.

Predicting long-term monthly electricity demand under future climatic and socioeconomic changes using data-driven methods: A case study of Hong Kong

Data-driven methods, such as artificial neural networks (ANNs), support vector regression (SVM), Gaussian process regression (GPR), multiple linear regression (MLR), decision trees (DTs), and gradient boosting decision trees (GBDTs), are the most popular and advanced methods for energy demand prediction. However, these methods have not been cross compared to analyze their performances for long-term energy demand predictions. Therefore, this paper aims to identify the best method among these data-driven methods for quantifying the impacts of climatic and socioeconomic changes on future long-term monthly electricity demand in Hong Kong. First, historical 40-year climatic, socioeconomic, and electricity consumption data are used to train and validate these models. Second, different representation concentration pathway (RCP) scenarios and three percentiles of 24 global circulation model outputs are adopted as future climatic changes, while five shared socioeconomic pathways are considered for future socioeconomic uncertainties. The results show that the GBDT method provides the best accuracy, generalization ability, and time-series stability, while ANN method exhibits the lowest accuracy and lower generalization ability. The monthly electricity demands in Hong Kong under the RCP8.5–2090 s scenario are predicted to increase by up to 89.40 % and 54.34 % in the residential and commercial sectors, respectively, when compared with 2018 levels.

EUReCA: An open-source urban building energy modelling tool for the efficient evaluation of cities energy demand

Recently, the attention towards Urban Building Energy Modelling has been growing due to the large contribution of cities on the worldwide energy consumption rate. In fact, many models have been developed to simulate buildings and urban energy systems.This article presents a new open-source tool for city-scale simulations. The platform implements the electrical analogy to model buildings thermal behaviour through resistance-capacitance networks, thus predicting the energy demand of urban areas with a bottom-up approach and low computational resources. The model has been applied to two urban areas: a small and well-known neighbourhood and a broader urban district.Results show the accordance of simplified models with respect to single building detailed simulations in the evaluation of the seasonal energy demand. When comparing results at hourly resolution, the accuracy is increased by considering a thermal network that splits buildings’ surfaces into two capacitances. A model based on a single capacitance leads to a higher deviation, especially on the daily peak power. Regarding the district simulation, these simplified networks show a good prediction of the buildings energy demand according to relevant European references, thus demonstrating the reliability of the proposed approach despite their low computational effort.

Importance of implementing smart renewable energy system using heuristic neural decision support system

Renewable energy sources are economically competitive choices for indigenous environmental solutions to produce conventional power, which offers high wind and solar resources to optimize resource management. Further, this resolution improves power storage and standby generation, leading to loss of revenue. Alternatively, better forecasts and demand for wind and solar energy allow advanced monitoring and optimization systems to replace costly equipment with lenient models. This research effort underway is consistent with the development of data-derived modeling through the use of machine learning technology. Furthermore, Hybrid plants will enhance the smart and sustainability of renewable energy systems economically and environmentally to meet energy demand. In this paper, the Heuristic Intelligent Neural Decision Support System (HINDSS) can be used to improve local energy production and forecast energy demand through optimized algorithmic approach on smart renewable energy systems. The experimental results show that the implementation of the smart system forecast on the renewable wind and solar energy production helps to reduce the demand for electricity at adequate time and increase the accuracy for an efficient replacement for alternative renewable energy via storage capacity.

Evolutionary Hybrid System for Energy Consumption Forecasting for Smart Meters

The usage of smart grids is growing steadily around the world. This technology has been proposed as a promising solution to enhance energy efficiency and improve consumption management in buildings. Such benefits are usually associated with the ability of accurately forecasting energy demand. However, the energy consumption series forecasting is a challenge for statistical linear and Machine Learning (ML) techniques due to temporal fluctuations and the presence of linear and non-linear patterns. Traditional statistical techniques are able to model linear patterns, while obtaining poor results in forecasting the non-linear component of the time series. ML techniques are data-driven and can model non-linear patterns, but their feature selection process and parameter specification are a complex task. This paper proposes an Evolutionary Hybrid System (EvoHyS) which combines statistical and ML techniques through error series modeling. EvoHyS is composed of three phases: (i) forecast of the linear and seasonal component of the time series using a Seasonal Autoregressive Integrated Moving Average (SARIMA) model, (ii) forecast of the error series using an ML technique, and (iii) combination of both linear and non-linear forecasts from (i) and (ii) using a a secondary ML model. EvoHyS employs a Genetic Algorithm (GA) for feature selection and hyperparameter optimization in phases (ii) and (iii) aiming to improve its accuracy. An experimental evaluation was conducted using consumption energy data of a smart grid in a one-step-ahead scenario. The proposed hybrid system reaches statistically significant improvements when compared to other statistical, hybrid, and ML approaches from the literature utilizing well known metrics, such as Mean Squared Error (MSE).

Modelling industry energy demand using multiple linear regression analysis based on consumed quantity of goods

Forecasting energy demand for the industrial sector is both interesting and difficult due to the difference in energy demand specific to each industrial sub-sector. For an accurate prediction of the future, Industry Energy Demand model was developed based on multiple linear regression method, using five macroeconomic independent variables. This model was tested by considering Morocco as a study case. Energy demand forecast is based on a bottom-up approach. It is built by piecing together consumed quantity of goods of each sub-sector to give rise to total energy demand. This model produces results comparable to those of the International Energy Agency. Regarding demand forecast, it was found that 8.27 MToe will be needed in 2050 to meet energy demand. It was also found that the adoption of energy efficiency measures allow an energy saving of 1 MToe in 2050. This model was also used to test the impact of variation in import and export on final energy demand. Regarding the potential of the production of biogas from Municipal Solid Waste, it was found that only 36.4% of total Liquefied Petroleum Gas demand could be replaced by biogas.

Prediction of transportation energy demand: Multivariate Adaptive Regression Splines

Energy usage in the transportation sector has been increasing in Turkey. Good management of energy is important as well as a reliable prediction of the energy demand in the transportation sector. The main objective of this research is to predict transport energy demand using Multivariate Adaptive Regression Splines (MARS) as a nonparametric regression technique. Transport energy demand was modeled for the period 1975–2019 based on a mix of factors including the gross domestic product (GDP), population, vehicle-km, ton-km, passenger-km and oil price. Five models were established and compared with real data collected from the Ministry of Energy and Natural Resources (MENR). Five MARS models including pairs of predictors, i.e. oil price-GDP, oil price-population, oil price-ton, oil price-vehicle and oil price-passenger, were evaluated comparatively in the prediction of energy demand. Among the candidate models, the third MARS model, which had the lowest RMSE, SD ratio, AICc values and the highest R2, Adjusted R2 and especially GR2 value, was selected as the best predictive model. In conclusion, it could be suggested that the third MARS model produced the highest predictive performance in the prediction of energy demand by two predictors, ton and oil price.

Support vector regression with asymmetric loss for optimal electric load forecasting

In energy demand forecasting, the objective function is often symmetric, implying that over-prediction errors and under-prediction errors have the same consequences. In practice, these two types of errors generally incur very different costs. To accommodate this, we propose a machine learning algorithm with a cost-oriented asymmetric loss function in the training procedure. Specifically, we develop a new support vector regression incorporating a linear-linear cost function and the insensitivity parameter for sufficient fitting. The electric load data from the state of New South Wales in Australia is used to show the superiority of our proposed framework. Compared with the basic support vector regression, our new asymmetric support vector regression framework for multi-step load forecasting results in a daily economic cost reduction ranging from 42.19% to 57.39%, depending on the actual cost ratio of the two types of errors.

Predicting Energy Demand in Semi-Remote Arctic Locations

Forecasting energy demand within a distribution network is essential for developing strategies to manage and optimize available energy resources and the associated infrastructure. In this study, we consider remote communities in the Arctic located at the end of the radial distribution network without alternative energy supply. Therefore, it is crucial to develop an accurate forecasting model to manage and optimize the limited energy resources available. We first compare the accuracy of several models that perform short-and medium-term load forecasts in rural areas, where a single industrial customer dominates the electricity consumption. We consider both statistical methods and machine learning models to predict energy demand. Then, we evaluate the transferability of each method to a geographical rural area different from the one considered for training. Our results indicate that statistical models achieve higher accuracy on longer forecast horizons relative to neural networks, while the machine-learning approaches perform better in predicting load at shorter time intervals. The machine learning models also exhibit good transferability, as they manage to predict well the load at new locations that were not accounted for during training. Our work will serve as a guide for selecting the appropriate prediction model and apply it to perform energy load forecasting in rural areas and in locations where historical consumption data may be limited or even not available.

Urban integrated energy demand forecasting method considering multi-station fusion

Under the background of energy reform, China’s energy-related industries actively innovate from the top to the bottom, vigorously promote the “Internet + Energy” strategy, and promote multi-station integration to deeply integrate energy, information communication and other pillar industries of the national economy to help China’s industrial upgrading and economic development. This paper proposes a method for forecasting urban comprehensive energy supply and demand that considers multi-station integration. Through the prediction and analysis of energy supply and demand under various scenarios of multi-station fusion, the comprehensive energy design scheme under different types of multi-station fusion scenarios is clarified to assist the early development of market developers and provide decision support for the promotion and application of comprehensive energy technology facing multi-station fusion.

Aggregated short-term load forecasting for heterogeneous buildings using machine learning with peak estimation

System operations and planning are crucial aspects of power system management. They aim to maintain the equilibrium of electricity supply and demand while ensuring reliable and secure power system operation. Consumers have to pay more for electricity during periods of high peak demand in various sectors. If consumers have knowledge about expected peak load ahead of time, such extra charges could potentially be avoided. Accurate energy demand forecasting, and therefore expected peak load information, not only will help to provide a reliable supply of electricity, but also can be useful in reducing the cost of electricity at the consumer level. In this paper, we develop a comparative study for aggregated short-term load forecasting using different data strategies and compare two prediction levels: predicting the aggregated load using a district-level data set, and performing predictions on a lower level and then aggregating them at the district level. After finding the best forecasting model and strategy, these accurate predictions will help to predict the percentage of peak over a certain subscribed power in the entire district. The results showed that the mean absolute percentage error is between 1.67% and 4.80% depending on the machine learning algorithm and the prediction horizon used.

Advanced Optimization Methods and Big Data Applications in Energy Demand Forecast

The use of data collectors in energy systems is growing more and more [...]

Probabilistic Prediction of Energy Demand and Driving Range for Electric Vehicles With Federated Learning

Today's drivers of battery electric vehicles must deal with limited driving range in a sparse charging infrastructure. An accurate prediction of energy demand and driving range is therefore important and enables reliable routing and charge planning applications. Predictions of energy demand entail uncertainty, which can be considered directly with the use of probabilistic prediction algorithms. Machine learning algorithms are frequently applied in this context, but data used to train these algorithms are often distributed over a fleet of connected vehicles. Federated learning can be applied in this setting, but predictive uncertainty is typically not considered. We apply an extension of the federated averaging algorithm to learn probabilistic neural networks and linear regression models in a communication-efficient and privacy-preserving manner. We demonstrate the performance advantage of probabilistic prediction models over deterministic prediction models using proper scoring rules. Furthermore, we show that federated learning can improve the standard, driver-individual learning. Using probabilistic predictions, variable safety margins based on destination attainability can be applied, leading to increased effective driving range and reduced travel time.