Toward scalable stochastic unit commitment. Part 1: load scenario generation

Abstract

Unit commitment decisions made in the day-ahead market and during subsequent reliability assessments are critically based on forecasts of load. Tra-ditional, deterministic unit commitment is based on point or expectation-based load forecasts. In contrast, stochastic unit commitment relies on multiple load sce-narios, with associated probabilities, that in aggregate capture the range of likely load time-series. The shift from point-based to scenario-based forecasting necessi-tates a shift in forecasting technologies, to provide accurate inputs to stochastic unit commitment. In this paper, we discuss a novel scenario generation method-ology for load forecasting in stochastic unit commitment, with application to real data associated with the Independent System Operator for New England (ISO-NE). The accuracy of the expected scenario generated using our methodology is consistent with that of point forecasting methods. The resulting sets of realistic scenarios serve as input to rigorously test the scalability of stochastic unit com-mitment solvers, as described in the companion paper. The scenarios generated by our method are available as an online supplement to this paper, as part of a novel, publicly available large-scale stochastic unit commitment benchmark. Abstract Unit commitment decisions made in the day-ahead market and during subsequent reliability assessments are critically based on forecasts of load. Traditional, deterministic unit commitment is based on point or expectation-based load forecasts. In contrast, stochastic unit commitment relies on multiple load scenarios, with associated probabilities, that in aggregate capture the range of likely load time-series. The shift from point-based to scenario-based forecasting necessitates a shift in forecasting technologies, to provide accurate inputs to stochastic unit commitment. In this paper, we discuss a novel scenario generation methodology for load forecasting in stochastic unit commitment, with application to real data associated with the Independent System Operator for New England (ISO-NE). The accuracy of the expected scenario generated using our methodology is consistent with that of point forecasting methods. The resulting sets of realistic scenarios serve as input to rigorously test the scalability of stochastic unit commitment solvers, as described in the companion paper. The scenarios generated Disciplines Industrial Engineering | Systems Engineering

Solution sensitivity-based scenario reduction for stochastic unit commitment

“…In addition, the 10 highest probability wind energy scenarios are selected from 50 wind energy scenarios to cross with 8 load scenarios which have been generated as described in [47], forming 80 hourly net load scenarios. The net load scenarios were scaled down to match the reduced generation capacity in both case studies.…”

“…For details of this load scenario generation process refer to [47]. Hourly wind scenarios were obtained from a commercial vendor [49] according to an analogue method [50].…”

“…Stairwise start-up cost function is adopted and formulated in the first term of (47), where Kg = {k 1 , · · · , k Ng }, and gk i ≤ gk i+1 , i = 1, · · · , Ng − 1. Notice that, gk includes shutdown cost in the case study.…”

From Data to Assessments and Decisions: Epi-Spline Technology

“…Figure 15(a) shows temperature forecasts for five days in Boston in 2012, with corresponding actually observed electricity loads on those days shown in Figure 15(b). Although, the historical information may stretch back decades, the useful information is much more limited because the data need to be carefully segmented to ensure that only "similar" days are included; see Feng et al [13,14]. For example, data for Mondays should not be mixed with those for Saturdays because the electricity load is fundamentally different on the weekend, and data for hot days should not be mixed with those of cool days.…”

“…In the end, only about 15 to 25 days per year may remain on which the forecast must rely, making traditional techniques based on time-series or stochastic differential equations inaccurate. Here, we briefly describe the epi-spline-based construction of a stochastic process of the next day's electricity load as laid out in Feng et al [13,14]; see also Rios et al [16]. Epi-splines enter at three points.…”