The LGO program offers a variety of internships looking at problems related to delivering and using energy. These internships provide a unique opportunity to pair a student’s engineering and MBA studies with an energy engineering and management concentration. For students looking to go even deeper, both the Mechanical Engineering and Civil Engineering departments have specializations in energy, and MIT’s Energy Initiative is a campus-wide organization focused on the topic.
Adam Chao (LGO ’16)
Company: Pacific Gas & Electric (PG&E)
Location: San Ramon, CA
Problem: PG&E owns and operates the natural gas system in northern and central California. Part of that system includes over 2,500 regulator stations that control the flow of gas, controlling a process to go from high-pressure transmission lines to lower-pressure distribution lines. Regulators can fail through a variety of causes. For example, sulfur contaminants or debris can build up inside the regulator. When regulators do fail, it can cause significant overpressures in the pipeline downstream. Therefore, regulator system safety is critical. PG&E has made significant efforts to link regulator stations to the supervisory control and data acquisition (SCADA) system for remote monitoring. By installing more SCADA sensor at regulator stations, PG&E can use predictive analytics of sensor data to identify safety issues before they occur.
Approach: Adam developed a set of algorithms to predict and identify potential unsafe conditions. He was also able to identify performance degradation in a regulator station. Detection of anomalies days or hours ahead of an overpressure event can allow the maintenance team to fix the problem before failure occurs.
Adam’s algorithms focused on time-series of pressure readings downstream of regulator stations. First, stations were clustered using k-means, thereby grouping stations with similar operating characteristics like volatility and pressure patterns. Second, Adam developed a series of algorithms to identify anomalous behavior in a given cluster. The algorithms use statistical process control techniques, CUSUM, and weighted moving averages to model long-term trends in the time-series and identify deviations from those trends. Local regression methods were also used to identify abnormal local trends.
Impact: Adam’s methods have been shown to detect buildups of sulfur and debris days before a corresponding overpressure event. PG&E was able to use Adam’s model to further the predictive analytics for shorter intraday time scales. Adam’s study also allowed PG&E to begin integrating other data sources such as temperature and maintenance into the model.
David Millard (LGO ’16)
Company: General Motors
Location: Detroit, MI
Problem: Vehicle manufacturers are developing technologies to increase fuel economy and reduce adverse vehicle emissions. Alternative propulsion technologies such as fuel cells are of tremendous interest, but factors including infrastructure requirements, technology costs, and market interest all significantly impact the viability of hydrogen fuel cell vehicles (HFCV). David quanified these HFCV adoption barriers and investigated strategies that may mitigate barriers to commercialization and long-term sustainability of an HFCV market.
Approach: David defined an existing system dynamics model based on the geography, demographics, and regulatory stance of California. He completed dozens of scenarios to assess variable impacts on the projected market share of vehicles based upon propulsion type.
Impact: David found several critical points that can be woven into HFCV commercialization strategies, including:
- Infrastructure: David identified optimal hydrogen infrastructure growth to support people buying HFCV cars and minimize required fueling stations. Furthermore, he outlined the conditions where external station support could give way to organic growth.
- HFCV ownership costs: David identified when vehicle price and government subsidies were optimized to speed HFCV adoption.
- Familiarity and consideration: David assessed how marketing investment needed to make customers familiar with HFCV while minimizing costs.
- Regulatory impact: David simulated the ZEV Action Plan impact and outlined measures that can be taken to meet regulatory requirements.
Arvind Simhadri (LGO ’15)
Company: National Grid
Location: Boston, MA
Problem: Massachusetts wants to have 1,600 megawatts (MW) of solar photovoltaic (PV) generation installed by 2020. National Grid expects that integrating solar power into the electricity network will require changes to the transmission system. Arvind developed a prediction methodology for distribution of solar PV in Massachusetts for use in a transmission system simulation framework. She then analyzed the impact on the transmission system.
Approach: To develop a prediction model for solar PV aggregate and spatial long-term distribution, Arvind gathered ten years of solar PV installation and electricity consumption data and added in population, land availability, average solar radiance, number of households, and other demographic data. He utilized data mining and machine learning methods to develop a model that gathers data on prediction variables. He then applied a regression model to forecast the capacity of solar PV generation per zip code.
Next, Arvind developed an electrically equivalent model to represent the predicted addition of solar PV on the distribution system and analyzed the impact of solar PV installations on steady-state voltage of the transmission system. He also conducted a sensitivity analysis on scenarios such as peak and light electricity consumption periods, different locations of PV, and voltage control methods to identify potential reliability concerns. Arvind tested system reliability in the event crucial transmission line outage.
Impact: The transmission system analysis with Arvind’s simulation framework identified that the transmission system in Massachusetts is robust and the voltage impact on transmission system is within +/- 5% of operating range. As National Grid acquires more information on solar installations, the new data will further improve the accuracy of solar PV capacity and location prediction. Also, the simulation framework developed in this project could be used to rerun analysis on other scenarios to identify potential problem areas on the transmission system.