114Any interruption in the electric service system could lead to major disruption of daily life and economic losses. Therefore, electric utility companies strive to maintain high customer service levels and try to make the best use of past power outage experiences. Increasing capabilities of analytics and improved accuracy of short-term weather forecasts present opportunities for electric utilities to leverage historic data to predict outages and proactively make decisions to minimize interruptions. As discussed in Chapter 3, analytics can be grouped into three types: descriptive analytics, predictive analytics, and prescriptive analytics. Electric utilities typically use descriptive analytics such as Pareto chart for descriptive statistics and data visualization to derive insights about most frequent cause of power interruptions. This chapter mainly focuses on predictive analytics that uses weather forecasts to predict power interruptions. In this chapter, prescriptive analytics is limited to formulating a multiple criteria mathematical programming model that minimizes staffing the cost and duration of power interruption. The objective of this chapter is to predict customer interruptions using historical data and hourly weather forecasts. Six different machine learning algorithms are evaluated and the best algorithm is chosen using multiple criteria ranking techniques to predict power interruptions. A case study using data of an electric utility covering 75 substations consisting of 19 fields and over 40,000 records collected over a 1-year period is used to illustrate the proposed methodology. Further, to illustrate the advantages of the prediction model, the prediction outputs are used as a key parameter to formulate a multicriteria mathematical programming model for workforce planning.
All Science Journal Classification (ASJC) codes
- Computer Science(all)