SOLUTIONS

Energy Demand Forecasting

The challenge

Electricity demand has been changing significantly in the last decade, influenced by many factors across sectors. These factors include the growth of populations in cities, economic development within cities, and many others. Electricity companies spend huge amounts of money to understand these changes and deploy models that aim to understand the overall status of electricity demand and determine how much expansion is needed to cover this increasing demand. Major investments are put in place to adhere to changes that impact the electric infrastructure and the daily operation of the power system. Moreover, optimized operations require short-term forecasting of the energy demand of different regions (24 hours ahead). This is necessary because the short-term power forecast plays a major role in the day-to-day operation of power systems, where it affects the unit commitment and economic dispatch.

What we did

A wide range of expertise is needed to tackle such a complex problem. The proposed solution involved the development of highly reliable machine learning and statistical models that forecast the energy demand of different regions and/or cities, each on different temporal scales (e.g., one hour ahead, one day ahead, one year ahead). The solution maps to our wide range of expertise that included advanced mathematical modeling, optimization, machine learning, and big data analytics techniques. In addition, a web-based tool was developed to incorporate all the knowledge and insights that were extracted from the data. The solution relied on the collection and cleaning of data, as some of the data feeds may have significant amounts of missing information. A variety of models were fitted to the data; in addition, combinations of machine learning models were implemented to guarantee high prediction accuracy. All developed algorithms were tested and evaluated by following the best practices and standards of commercial grade tools.

Our Impact

The forecasting system achieved a mean absolute percentage error of less than 1% on data for different spatial areas and with various temporal scales. The developed algorithms consider the unique characteristics of each spatial area. This allows the decision-makers to easily cover more cities because the tool is developed to be universal. The web-based tool improves the efficiency of the daily operation process. It provides accurate forecasting of the demand, allowing the operation team to provide exactly the needed amount of energy for a city. Additionally, the tool provides a convenient user experience and allows users to explore and update the load forecast. The accurate predictions of this tool save millions of dollars per year while ensuring energy-grid security.