The Problem: An electric utility must provide enough electricity to meet the demands of the customers in its service area. Sometimes it is cheaper to purchase electricity from adjacent utility companies than to ramp up generating assets. As power flows from one utility to another, it follows the paths of least resistance across all powerlines connecting the two utilities. If a powerline is overloaded, it can overheat and fail. When one powerline fails, all of the power it was transmitting diverts to the remaining powerlines, and the increased power load on the remaining lines can cause them to fail as well. This creates a cascading effect of powerline failure and can cause widespread blackouts.
The Solution: The network and capacity of all powerlines connecting two regions is known. Based on the total projected power transmission requirements, an algorithm can be developed to determine how much power will be flowing on each of the individual powerlines. These determinations are based on the physics of electricity and conductors. Risks to the reliability of the power grid can be assessed by comparing the projected power flow across each powerline against the powerline’s safe maximum capacity. Based on these determinations, it is possible to identify the critical powerline that is most at risk of failing, and power load decisions can be made based on these analyses to provide a safe margin for that critical powerline and ensure the reliability of the power grid.
The Problem: Electric utilities must provide a sufficient quantity of power to meet the expected power demand (or load) of its customer base at various times of the day. As additional power is required, the utility typically has a choice of either starting up a reserve power generating plant or purchasing the additional power from other utilities. The decision is based, in part, on economics. If the power can be purchased more cheaply than it can be generated, then the utility usually will enter into a contract to purchase the power. If a contract can be effected at least a day or more in advance, the utility can negotiate a much better price. The challenge is to know how much the power demand will be for each hour of the day at least a day or more in advance.
The Solution: The price of power on the wholesale energy market is based primarily on the temperature of the weather. Extremely hot (or cold) weather causes much more power consumption, thereby driving up the price. The solution was to develop and implement an algorithm that predicts the price of the wholesale energy based on the weather forecast. A non-linear regression analysis predictive algorithm is being used to predict the price of power days in advance. Based on the predicted price on the spot market, management can make decisions about which plants to activate and when to purchase wholesale power instead of generating it, saving millions of dollars in wholesale power costs.
The Problem: A financial institution makes sub-prime loans based on loan application data entered into a web page. The applicant must meet certain requirements to be approved for a loan, and the applicant’s data is verified via various web-based services. A percentage of the loans were being made based on fraudulent data that was not being detected. The financial institution needed a way to identify which loan applications were likely to be fraudulent so the loan requests could be denied while ensuring valid loan requests still were approved.
The Solution:The data from hundreds of thousands of loan requests were analyzed to identify characteristics associated with the loans that turned out to be fraudulent. Based on the results of these analyses, predictive characteristics were identified that subsequently were used to deny subsequent fraudulent loan requests. When the algorithm was first implemented, loan losses to fraud were reduced from about $250,000 per month to about $85,000. As the algorithm continued to learn, these losses were further reduced over time without denying any of the valid loan applications.
The Problem: One of the largest commercial floor manufacturing companies in North America needed a visualization tool to allow customers to see how their custom floor designs would look on completion before the beginning of the manufacturing process.
The Solution:The custom flooring generator tool required the development of innovative algorithms to render realistic images. It takes into account factors such as creating individual components from random samples of larger images, scale of each image, wood grain direction of each image, and alpha blending when composing the final image. This innovative visulation tool enables customers to experiment with various design patterns and flooring choices without having to resort to expensive trial production processes.
The Problem: A Support Help Desk fields hundreds of support calls each day. Service Level Agreements (SLAs) require that each call be handled within a specified timeframe. In order to meet the SLA requirements, the Support Help Desk must have a sufficient number of people to handle the work load. But additional people come at an additional cost. The problem was to ensure that there would be enough people on hand to handle the workload and meet the SLA requirements while not having the additional costs of having too many on hand.
The Solution: Develop and implement a non-linear regression predictive algorithm based on historical data to predict the workload for each hour of each day. Staffing decisions are based on these predictions, ensuring that there are enough people to meet the SLA while minimizing staffing costs.
Robert is an IT professional specializing in developing and implementing custom algorithms to solve difficult business problems. His solutions have been implemented in a variety of industries, including financial, energy, communications, and manufacturing environments.
When not solving business problems, Robert enjoys mentoring high school students in computer science and ACT Prep, and tinkering with his custom-built railroad.
He is currently participating in the Google FooBar Challenge, where he is has advanced to Level 5.
One of Robert's hobbies is trying to beat the market by picking stocks and options using statistical analysis to identify Technical Indicators for trade decisions. He continues work on his trading algorithm implemented in Python using Numpy and Pandas. The analysis currently considers the following technical indicators:
As part of these analyses, he has implemented an engine that captures each day's stock and option prices after the market closes. The code for this data capture can be found here: