Research Description

Master’s Project

Solar Net Energy Metering:

Net Energy Metering (NEM) schema has been recognized as an effective way to diversify energy resource mix, stimulate local economic growth, and encourage private investment in solar energy. However, several studies have shown that NEM policies are regressive onto low-income, disadvantaged communities as the revenue loss for utilities causes them to raise electricity rates in order to recover their costs. In this study, we

  1. conducted an Optimal Power Flow (OPF) analysis in Python to quantify how NEM schema can change grid power flow and influence grid congestion, LMP, and utility’s avoided cost.
  2. evaluated NEM policies by performing a cost-benefit analysis from the perspective of the NEM customers, non-NEM customers and the utilities, to demonstrate how NEM policies result in a total net cost.
  3. finally propose recommendations including using a time of use compensation rate, increasing fixed non-bypassable charges, and a transitional NEM credit system, etc.

Sample results:

(P1 to P4 represents increase in solar DGs penetration rate; while C1 to C4 indicates increase in solar capacity installed on rooftop)


Machine Learning application in the Energy field

OCR practice: Electricity meter recognition:

Automatic Meter Reading (AMR), which refers to automatically record the consumption of electric energy, gas and water for both monitoring and billing without human intervention, is an important application of OCR (Optical Character Recognition) technology. Specifically, operators don’t need to read meters manually on site, instead, taking a picture as reading proof and letting OCR to detect numbers online is cheaper in terms of human effort and time, and has high efficiency.

In this project, I implemented a state-of-art, CNN-based object detector called yolo for detecting counters in 800 electric meter images taken from a warehouse in Brazil. The model takes input images and learns the class possibilities with coordinates of the counter area, and after approx. 1700 times of iteration, the training model reaches an avg loss of 0.17, and mean average accuracy on training image is as high as 100%.

Left: Loss and mAP graph during training; Right: Example on testing image recognition


Renewable Energy Topics

Pathways toward carbon-free energy system: Utility in many regions across the U.S. have publicly stated emission reduction goals to reduce carbon footprint of local electric power system. How to plan in a region with both old fossil fuel plants and site-specific renewable alternatives is the key question for analysts. In a project to make capacity planning for a 3-bus electric power system, I lead a group of 4 to explore different renewable scenarios to enhance the proportion of green power in the grid by 10 years later. The purpose is to maximize aggregate carbon emission reduction, minimize total cost of new construction or retrofit, and meet annual system electricity demand. A capacity planning model based on Python (Pulp package) is built to examine plans that incorporate different resources at different times. New capacity under consideration includes residential roof-top solar, utility solar PV, large CSP facility, wind turbine and energy storage system. A last optimal plan is given by the model showing the construction schedule of new renewable capacities among 10 years and recommend-retired year for existing coal or natural gas plants.

Wind farm site evaluation in ArcGIS: Site assessment is one of the key factors in the successful installation, operation and performance of a new wind farm. Determining a properly-placed turbine usually requires a series of studies. In a research to assess 10 land parcels in western North Carolina as prospective sites for wind farm station, I used ArcGIS pro to evaluate site characteristics including: 1). its distance to flight path restriction areas and avian species inhabitants; 2). potential turbine generation given available site area and wind class; 3). total cost for each to cross terrain and be connected to existing power grids; and 4). compensation cost paid to nearby Parkway Tourist Network for remediating wind farm’s damage on scenic beauty (depending on the number of scenic overlooks fall within wind farm’s viewshed). Based on each criterion, the optimal land parcel, which can host a wind farm that can deliver the most power but also be built under specified budget constraints, is decided.

Energy time series data analysis and forecast: To predict future electricity usage, grid load, generation from intermittent resources, and many other energy-related indicators, time series model is a powerful tool. By fitting and analyzing past behavior of a variable, people can forecast future activities and plan for future operations. In a research aimed at studying patterns of Japan’s oil and gas import, I use R to fit time series models on O & G imports data from 1997 to 2019 and deploy the best-fit ones to forecast imports volume 5 years after. Each model’s performance is evaluated based on goodness of matching at inflection points, fitting residuals pattern, and accuracy measures such as MRE, RMSE and MAPE. In addition, analyzing how the imports pattern of both O & G have changed before and after Fukushima Event reflects the disruption of nuclear power loss on Japan’s energy system.


Environmental Science and Policy Topics

Supervised by Prof. Xu Zhao, Shandong University; Prof. Xin Zhang, University of Maryland Center for Environmental Science.

Related publications:

Global scarce water saving through International crop trade: Agricultural products trade is associated with virtual water transfer between trading partners. When the amount of virtual water transfer is lower than a hypothetical water use in importing countries to produce the same commodities, a global water saving is generated. When further taking domestic water scarcity into account, scarce water saving can also be estimated. In an undergraduate research, I used Python to calculate virtual scare water savings embodied in crop trade among 199 countries in 2015 and examine the relationship between a country’s water stress index and net virtual water imports. The result identified specific trade links that generate large amount of virtual water savings (see figure below) and exporter-importer pairs that significantly intensify water scarcity in exporting countries.

Managing nitrogen for sustainable agriculture: Large scale of croplands nourished more than half the world’s people, but also intensified the escape of reactive N from agricultural soils. Global trade enables one nitrogen-intensive country to import crops from a lightly-polluted country, thus reducing total amount of nitrogen leaking to the atmosphere or riverways. In an undergraduate research, I quantified the amount of nitrogen loss embedded in crop outflows and examined the role of 16 categories of agricultural products trade on alleviating global nitrogen pollution from 1986 to 2011, as shown in the figure below.

Evaluating water resource networks’ vulnerability in China: The water resource networks that provide water for urban consumption consists of both physical and virtual water supply. However, relying on external water footprints can influence the vulnerability of cities’ water networks. In an research to evaluate water vulnerability of China’s six megacities, we developed a vulnerability index and calculated physical / virtual water flows in six megacities’ networks based on a multi-region input-output model. The results enable urban water planners to internalize the risk of external water footprint and avoid importing water-intensive products from regions suffering from extreme water stress.