With the growing robustness of modern power systems, the occurrence of load curtailment events is becoming lower. Hence, the simulation of these events constitutes a challenge in adequacy indices assessment. Due to the rarity of the load curtailment events, the standard Monte Carlo simulation (MCS) estimator of adequacy indices is not practical. Therefore, a framework based on the enhanced cross-entropy-based importance sampling (ECE-IS) method is introduced in this paper for computing the adequacy indices. The framework comprises two stages. Using the proposed ECE-IS method, the first stage’s purpose is to identify the samples or states of the nodal generation and load that are greatly significant to the adequacy indices estimators. In the second stage, the density of the input variables’ conditional on the load curtailment domain obtained by the first stage are used to compute the nodal and system adequacy indices. The performance of the ECE-IS method is verified through a comparison with the standard MCS method and the recent techniques of rare events simulation in literature. The results confirm that the proposed method develops an accurate estimation for the nodal and system adequacy indices (loss of load probability (LOLP), expected power not supplied (EPNS)) with appropriate convergence value and low computation time.
The article addresses probabilistic and statistical methods for determining the total load losses of power and electricity in a power distribution network with distributed generation. The aim of the study is to compare the accuracy of existing and proposed models for the daily forecast of technical losses of power and electricity in a distribution network with partially determined information on load and generation schedules. Three analytical models are considered: the mode of mathematical expectations (ME) of nodal powers, consideration of the matrix of correlation moments of powers together with the MO of voltages, and allocation of a subinterval of maximum loads. It has been found that the estimation of losses based solely on the ME of loads yields a large error. In networks with distributed generation, in which the daily ME of the nodal powers may be close to zero, the classical methods based on the shape coefficient and the number of hours of highest losses are not applicable. For practical calculations, it is recommended to use methods based on the ME of loads with full or partial consideration of their correlation, as well as dividing the calculation interval into subintervals of characteristic modes (maximum and minimum loads, periods of conditionally constant load, etc.). Статья посвящена вероятностно-статистическим методам определения суммарных нагрузочных потерь мощности и электроэнергии в распределительной электрической сети с распределенной генерацией. Цель исследований – сопоставить по точности существующие и предлагаемые модели суточного прогноза технических потерь мощности и электроэнергии в распределительной электрической сети при частично определенной информации о графиках нагрузки и генерации. Рассмотрены три расчетные модели: режим математических ожиданий (МО) узловых мощностей; учет матрицы корреляционных моментов мощностей совместно с МО напряжений; выделение подынтервала максимальных нагрузок. Установлено, что оценка потерь исключительно на основе МО нагрузок дает большую погрешность. В сетях с распределенной генерацией, где суточное МО узловой мощности может быть близко к нулю, классические методы, опирающиеся на коэффициент формы и число часов наибольших потерь, неприменимы. Для практических расчетов рекомендовано использовать методы на основе МО нагрузок с полным или частичным учетом их корреляции, а также дробление расчетного интервала на подынтервалы характерных режимов (максимальных и минимальных нагрузок, периодов условно постоянной нагрузки и др.).