Our risk analytics include modeling of distribution tails that represent of extreme events. Such events could represent extreme (peak) power loads, extreme day-ahead forecast errors, extreme spot prices of electricity, or extreme weather events. Long tail distribution models are used in the practice of financial risk management and actuarial applications. We scale to these approaches to modeling of the entire probablity distributions and tail trends from the big data in the IIoT.

Our tools build multi-level extreme event models occurrences and magnitudes reflecting location, seasonality, and trends of the extreme event frequencies and magnitudes from the historical data. Our prescriptive analytics rely on the predictive models of the extreme events for accurate and comprehensive risk evaluation for specific applications.

One set of the probabilistic risk analytics applications is in the electric power system planning. The tools enable assessment of system resource adequacy using data-driven models of the peak events. The main difficulty of assessing generation and transmission resource adequacy is that future peak loads are unknown. The solution is to build probabilistic model of past peak loads and use it to predict the risk Loss of Load Probability (LOLP). NERC mandates that Loss of Load Expectation (LOLE) does not exceed 1 day in 10 years. This means the acceptable LOLP is no more than 0.027%. Our analytics using accurate data-driven models of distribution tails based on Extreme Value Theory (EVT). Our analytics accounts for the year-to year trends as well as statistical similarities between different service zones or assets. This allows significant improvement in the accuracy and reliability of the estimate, while automating its computation. The approach has been proven with utility data.

Another set of the probabilistic risk analytics applications is in operational risk assessment. For a given day-ahead commitment of the generators the analytics allows to compute the risk of having to shed a firm load. In this application, characterization of the forecast error distribution tail allows accurate automated computation of the risk. The approach has been proven in a project with the regional electrical system operator, ISO New England.