Combining simulations and data with deep learning and uncertainty quantification for advanced energy modeling

Radaideh, Majdi I.; Kozłowski, Tomasz

doi:10.1002/er.4698

Cited by 27 publications

(11 citation statements)

References 77 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Like the development of the time series model, the nontime series model has attracted the attention of scholars. 27 Dong et al forecasted building energy consumption in the tropical region using mean data from monthly landlord utility bills by support vector machines (SVM). 21 Coefficient of variance of less than 3% and percentage errors (% error) under 4% were obtained.…”

Section: Energy Consumption Modelingmentioning

confidence: 99%

Multistep energy consumption forecasting by metaheuristic optimization of time‐series analysis and machine learning

Chou

Truong

2020

Int J Energy Res

View full text Add to dashboard Cite

Summary This study aims to develop a novel forecasting system that optimizes linear time‐series with nonlinear machine learning models to identify the historical pattern of regional energy consumption. The linear time‐series model, Seasonal AutoRegressive Integrated Moving Average (SARIMA), was applied to simulate the linear component, while the least squares support vector regression (LSSVR) model was used to capture the nonlinear component of time series energy data and combine the linear and nonlinear components. Several optimization algorithms were investigated using high‐dimension mathematical benchmark functions to compare their outcomes for applying in the proposed forecasting system. Wilcoxon rank‐sum test and convergence graphs for Jellyfish Search (JS) and other parameter‐less algorithms (ie, Teaching‐Learning‐Based Optimization (TLBO) and Symbiotic Organisms Search, SOS algorithms) indicate that the JS optimizer finds the best global solutions in mathematical tests. Then, three real energy time‐series datasets from a regional transmission organization that coordinates the movement of wholesale electricity were used to evaluate various forecasting methods. The analytical results confirm that the proposed system, JS‐LSSVR(SARIMA, LSSVR), can predict multi‐step ahead time series energy consumption with higher accuracy than the linear model (ie, SARIMA), nonlinear model (ie, LSSVR), hybrid model (ie, JS‐LSSVR), hybrid systems (ie, TLBO‐LSSVR(SARIMA, LSSVR) and TLBO‐LSSVR(SARIMA, LSSVR)), and prior studies. Numerical experiments show that the JS‐LSSVR(SARIMA, LSSVR) system can forecast energy consumption 1 week ahead efficiently (from 9.8 to 21.4 seconds on average). Notably, the proposed technique only requires 4 inputs vs 7 to 32 inputs of other models in the literature. Thus, a power company can apply the proposed system to forecast energy consumption to efficiently dispatch regional energy capacity and keep the electricity supply and demand in balance for residential buildings in sustainable cities.

show abstract

Section: Energy Consumption Modelingmentioning

confidence: 99%

Multistep energy consumption forecasting by metaheuristic optimization of time‐series analysis and machine learning

Chou

Truong

2020

Int J Energy Res

View full text Add to dashboard Cite

show abstract

“…Because of the generic nature of these methods, they can be applied directly to any fuel cell type or energy model, given that the input‐output relationship is well‐known to the analyst. Finally, additional improvements to the current methods can be performed by introducing more modern techniques of artificial intelligence and model‐form uncertainty evaluation at which real data is used to assess the fuel cell models.…”

Section: Resultsmentioning

confidence: 99%

Efficient analysis of parametric sensitivity and uncertainty of fuel cell models with application to SOFC

Radaideh

2020

Int J Energy Res

Self Cite

View full text Add to dashboard Cite

Summary Sensitivity analysis (SA) and uncertainty quantification (UQ) are used to assess and improve engineering models. SA of fuel cell models can be challenging because of the large number of design parameters to optimize. Following the regular approach of manually testing the fuel cell outputs under changing each design parameter one at a time can be tedious. In this work, a framework of SA and UQ methods is applied with a purpose of efficiently analysing fuel cell models. The SA and UQ methods are also compared to increase the confidence in the results. In this methodology, all design parameters and their effects can be analysed in an integrated form, where model variance, sensitivity, linearity/nonlinearity, parameter interactions, and importance ranking can be assessed. This paper highlights and compares between local SA (one‐at‐a‐time linear perturbation), parameter screening (Morris screening), variance decomposition (Sobol indices), and regression‐based SA. For UQ, stochastic methods (Monte Carlo sampling) and deterministic methods (using SA profiles) are used. All methods are applied to solid oxide fuel cell (SOFC) to analyse the power output and system efficiency under 21 uncertain design parameters. Using different methods, the uncertainty in the SOFC responses (maximum power and efficiency) is about 11%, when the current density is about 13 200 A m−2. In addition, analysis shows that operating temperature (T), length of grain contact (X), grain size (Ds), porosity (ϵ), and electrolyte thickness (Le) contribute to more than 97% of the SOFC's maximum power variance, with individual contributions as follows: 63.3%, 13.1%, 12.5%, 5.4%, and 3.6%, respectively. The previous conclusions are subjective to the simplified SOFC model used in this study to demonstrate the methods. Benchmarking the UQ methods in capturing the response uncertainty and the SA methods in raking the design parameters demonstrate very good agreement between them. The methods applied in this paper can be used to achieve a comprehensive mathematical understanding of more advanced fuel cell or energy models, which can lead to better performance.

show abstract

“…unprecedented results across various application domains 16,17,18,19,20 . In the task of unsupervised learning, generative models are one of the most promising technologies.…”

Section: Introductionmentioning

confidence: 99%

Day-ahead renewable scenario forecasts based on generative adversarial networks

Jiang¹,

Mao²,

Chai³

et al. 2021

Preprint

View full text Add to dashboard Cite

<p>With the increasing penetration of renewable resources such as wind and solar, especially in terms of large-scale integration, the operation and planning of power systems are faced with great risks due to the inherent stochasticity of natural resources. Although this uncertainty can be anticipated, the timing, magnitude, and duration of fluctuations cannot be predicted accurately. In addition, the outputs of renewable power sources are correlated in space and time, and this brings further challenges for predicting the characteristics of their future behavior. To address these issues, this paper describes an unsupervised distribution learning method for renewable scenario forecasts that considers spatiotemporal correlation based on generative adversarial network (GAN), which has been shown to generate realistic time series for stochastic processes. We first utilize an improved GAN to learn unknown data distributions and model the dynamic processes of renewable resources. We then generate a large number of forecasted scenarios using stochastic constrained optimization. For validation, we use power generation data from the National Renewable Energy Laboratory wind and solar integration datasets. The simulation results show that the generated trajectories not only reflect the future power generation dynamics, but also correctly capture the temporal, spatial, and fluctuant characteristics of the real power generation processes. The experimental comparisons verify the superiority of the proposed method and indicate that it can reduce at least 50% of the training iterations of the generative model for scenario forecasts.<br></p>

show abstract

Combining simulations and data with deep learning and uncertainty quantification for advanced energy modeling

Cited by 27 publications

References 77 publications

Multistep energy consumption forecasting by metaheuristic optimization of time‐series analysis and machine learning

Multistep energy consumption forecasting by metaheuristic optimization of time‐series analysis and machine learning

Efficient analysis of parametric sensitivity and uncertainty of fuel cell models with application to SOFC

Day-ahead renewable scenario forecasts based on generative adversarial networks

Contact Info

Product

Resources

About