• Energy Research
• 2025-2025close
• 7. Clean energy close
• 6. Clean water close

Background: Machine Learning (ML) techniques have successfully replaced traditional control algorithms in recent years due to their ability to carry out complicated tasks with significant efficiency and accuracy. Objective: The main objective of the current work is to investigate and compare the performances of different ML models in modeling Maximum Power Point Tracking (MPPT) control for a wind turbine system. The main advantage of the designed MPPT based on ML is that it does not require any detailed mathematical model or prior knowledge of the system, such as turbine parameters or aerodynamic properties, unlike traditional MPPT techniques. Methods: The ML models included in this study were Support Vector Machines, Regression Trees, and Ensemble Trees. Their design was performed through a training process, and their performances were evaluated based on various metrics. During the training phase, the ML models were selected to understand the basic concept of the control strategy and extract essential hidden connections between the inputs and the output of the system. Results: The effectiveness of the control method was investigated using MATLAB/Simulink. The findings of this study revealed that ML models were effective in modeling the MPPT for the studied wind power system, which provides an interesting and sophisticated alternative to classical control methods for wind systems. Conclusion: The ML models designed allow for optimal operation of the system with a simple structure that is independent of system parameters and wind speed measurement and is adaptable for any kind of system.

Background: In recent years, the integration of renewable energy sources into the grid has increased exponentially. However, one significant challenge in integrating these renewable sources into the grid is intermittency. Objective: To address this challenge, accurate PV power forecasting techniques are crucial for operations and maintenance and day-to-day operations monitoring in solar plants. Methods: In the present work, a hybrid approach that combines Deep Learning (DL) and Numerical Weather Prediction (NWP) with electrical models for PV power forecasting is proposed Results: The outcomes of the study involve evaluating the performance of the proposed model in comparison to a Physical model and a DL model for predicting solar PV power one day ahead and two days ahead. The results indicate that the prediction accuracy of PV power decreases and the error rates increase when forecasting two days ahead, as compared to one day ahead. Conclusion: The obtained results demonstrate that DL models combined with NWP and electrical models can improve PV Power forecasting compared to a Physical model and a DL model.

We present a modeling and optimization framework to design powertrains for a family of electric vehicles, focusing on the concurrent sizing of their motors and batteries. Whilst tailoring these component modules to each individual vehicle type can minimize energy consumption, it can result in high production costs due to the variety of component modules to be realized for the family of vehicles, driving the Total Costs of Ownership (TCO) high. Against this backdrop, we explore modularity and standardization strategies whereby we jointly design unique motor and battery modules to be installed in all the vehicles in the family, using a different number of these modules when needed. Such an approach results in higher production volumes of the same component module, entailing significantly lower manufacturing costs due to Economy-of-Scale (EoS) effects, and hence a potentially lower TCO for the family of vehicles. To solve the resulting one-size-fits-all problem, we instantiate a nested framework consisting of an inner convex optimization routine which jointly optimizes the modules' sizes and the powertrain operation of the entire family, for given driving cycles and modules' multiplicities. Likewise, we devise an outer loop comparing each configuration to identify the minimum-TCO solution with global optimality guarantees. Finally, we showcase our framework on a case study for the Tesla vehicle family in a benchmark design problem, considering the Model S, Model 3, Model X, and Model Y. Our results show that, compared to an individually tailored design, the application of our concurrent design optimization framework achieves a significant reduction of the production costs for a minimal increase in operational costs, ultimately lowering the family TCO in the benchmark design problem by 3.5\%. 17 pages, 17 figures, 7 tables

Global lakes hold about 87% of the freshwater. However, climate change has posed a severe threat to these freshwater resources. Evaporation (E) is a major water loss from lakes, and the strong coupling between lake E and changes in atmospheric conditions in a warming climate leads to temporal and spatial variability in water loss through E, making it challenging for water resource management. This dissertation examines such spatiotemporal variability in global lake E in response to climate change, investigates its environmental controls, and identifies regions with large sensitivities to climate changes. Using a state-of-science Lake, Ice, Snow, and Sediment Simulator (LISSS) that is a lake model within the Community Land Model (CLM), it is shown that the large spatial variability of global lake E is modulated by the vapor pressure difference (e_D) between lake surface and overlying air. The e_D also causes higher nighttime lake E, which contributes more to the spatial variability of global lake E than daytime lake E. The performance of the Penman method (PM) is also evaluated against observations and the LISSS modeling results in estimating global lake E. It is shown that the PM overestimates lake E due to a strong bias in the net radiation (Rn) and lake water heat storage (G). Using the LISSS simulated Rn and G in the PM, however, the PM performance is largely improved and the PM E becomes comparable to the LISSS E. The global lake E trend over 1951 - 1978 is analyzed, which shows a decreasing E trend. Such a declined global lake E was largely caused by the decreased downward shortwave solar radiation. The global lake E was switched from the decreased trend over 1951-1978 to an increased trend over 1981-2016 with an accelerated trend of 0.76 mm yr-1. The tropical, arid, and temperate climate regions lakes contribute 66% to the increasing trend despite covering only 38% of the global lake surface area. Such a change in the global lake E trend was attributed to the increased vapor pressure deficit in a warmer climate. The model projection indicates that the mean global lake E will increase by 13% by the end of the 21st century under the Representative Concentration Pathway (RCP) 8.5 emissions scenario, relative to the 1985-2000 mean global lake E. The changes in lake E are expected to be more pronounced in North America, equatorial South America, Africa, northern Europe, Siberia, and Southeast Asia due to increased interannual variability. The results in this dissertation indicate that the widespread but heterogeneous increase in the global lake E threatens the crucial socioeconomic benefits that lakes provide to human society.

The increasing wildfire activity in the past few years has been devastating, setting negative records in many states and regions around the world, especially in North America. Power systems have been impacted by wildfires in many ways, even in regions located hundreds of kilometers away from high-risk zones, depending on wind speed and direction conditions, the stemming smoke of wildfires may significantly impact the air quality and reduce the solar PV generation, and forcing several utilities to rely on PSPS programs to mitigate wildfire risks. Thus, power system operators must ensure reliability and resilience across power generation, transmission, and distribution while minimizing carbon emissions that can harm even more the air quality of the affected communities during wildfire events. Furthermore, a cost-effective power system expansion planning solution in regions with increased wildfire risk is achieved by placing ESSs and new transmission/distribution lines while taking into account their availability given the increasing number of PSPS events. This research aims to analyze the impact of wildfire activity on the electrical system's planning and operation, by analyzing the impact of the 2020 wildfire season on renewable energy in Washington state, focusing on variables that directly impact the wind and photovoltaic power. After that, efforts are made to approach the expansion planning of power transmission and distribution systems under wildfire risk, considering sitting and sizing of ESS as an alternative, with a compliance check on unbalanced power flow and system operating limits. The resulting models are a MILP optimization problem, and simulation experiments are performed to validate the effectiveness of the proposed formulation using different High Fire-Threat District Tier Zones based on real-world data from electric utilities in California.

Combustion occurring in porous media has various practical applications, such as in in-situ combustion processes in oil reservoirs, the combustion of biogas in sanitary landfills, and many others. A porous medium where combustion takes place can consist of layers with different physical properties. This study demonstrates that the initial value problem for a combustion model in a multi-layer porous medium has a unique solution, which is continuous with respect to the initial data and parameters in $\mathtt{L}^2(\mathbb{R})^n$. In summary, it establishes that the initial value problem is well-posed in $\mathtt{L}^2(\mathbb{R})^n$. The model is governed by a one-dimensional reaction-diffusion-convection system, where the unknowns are the temperatures in the layers. Previous studies have addressed the same problem in $\mathtt{H}^2(\mathbb{R})^n$. However, in this study, we solve the problem in a less restrictive space, namely $\mathtt{L}^2(\mathbb{R})^n$. The proof employs a novel approach to combustion problems in porous media, utilizing an evolution operator defined from the theory of semigroups in Hilbert space and Kato's theory for a well-posed associated initial value problem.

Machine learning's application in solar-thermal desalination is limited by data shortage and inconsistent analysis. This study develops an optimized dataset collection and analysis process for the representative solar still. By ultra-hydrophilic treatment on the condensation cover, the dataset collection process reduces the collection time by 83.3%. Over 1,000 datasets are collected, which is nearly one order of magnitude larger than up-to-date works. Then, a new interdisciplinary process flow is proposed. Some meaningful results are obtained that were not addressed by previous studies. It is found that Radom Forest might be a better choice for datasets larger than 1,000 due to both high accuracy and fast speed. Besides, the dataset range affects the quantified importance (weighted value) of factors significantly, with up to a 115% increment. Moreover, the results show that machine learning has a high accuracy on the extrapolation prediction of productivity, where the minimum mean relative prediction error is just around 4%. The results of this work not only show the necessity of the dataset characteristics' effect but also provide a standard process for studying solar-thermal desalination by machine learning, which would pave the way for interdisciplinary study.

With California's ambitious goal to achieve decarbonization of the electrical grid by the year 2045, significant challenges arise in power system investment planning. Existing modeling methods and software focus on computational efficiency, which is currently achieved by simplifying the associated unit commitment formulation. This may lead to unjustifiable inaccuracies in the cost and constraints of gas-fired generation operations, and may affect both the timing and the extent of investment in new resources, such as renewable energy and energy storage. To address this issue, this paper develops a more detailed and rigorous mixed-integer model, and more importantly, a solution methodology utilizing surrogate level-based Lagrangian relaxation to overcome the combinatorial complexity that results from the enhanced level of model detail. This allows us to optimize a model with approximately 12 million binary and 100 million total variables in under 48 hours. The investment plan is compared with those produced by E3's RESOLVE software, which is currently employed by the California Energy Commission and California Public Utilities Commission. Our model produces an investment plan that differs substantially from that of the existing method and saves California over 12 billion dollars over the investment horizon.

Optimally scheduling multi-energy flow is an effective method to utilize renewable energy sources (RES) and improve the stability and economy of integrated energy systems (IES). However, the stable demand-supply of IES faces challenges from uncertainties that arise from RES and loads, as well as the increasing impact of cyber-attacks with advanced information and communication technologies adoption. To address these challenges, this paper proposes an innovative model-free resilience scheduling method based on state-adversarial deep reinforcement learning (DRL) for integrated demand response (IDR)-enabled IES. The proposed method designs an IDR program to explore the interaction ability of electricity-gas-heat flexible loads. Additionally, the state-adversarial Markov decision process (SA-MDP) model characterizes the energy scheduling problem of IES under cyber-attack, incorporating cyber-attacks as adversaries directly into the scheduling process. The state-adversarial soft actor-critic (SA-SAC) algorithm is proposed to mitigate the impact of cyber-attacks on the scheduling strategy, integrating adversarial training into the learning process to against cyber-attacks. Simulation results demonstrate that our method is capable of adequately addressing the uncertainties resulting from RES and loads, mitigating the impact of cyber-attacks on the scheduling strategy, and ensuring a stable demand supply for various energy sources. Moreover, the proposed method demonstrates resilience against cyber-attacks. Compared to the original soft actor-critic (SAC) algorithm, it achieves a 10% improvement in economic performance under cyber-attack scenarios. Accepted by Applied Energy, Manuscript ID: APEN-D-24-03080