Improvement in the computational efficiency of a technique for assessing the reliability of electric power systems based on the Monte Carlo method

• Published in: Reliability engineering & system safety 2020-12, Vol.204, p.107171, Article 107171
• Main Authors: Krupenev, Dmitry, Boyarkin, Denis, Iakubovskii, Dmitrii
• Format: Article
• Language: English
• Subjects: Computer applications; Electric power; Electric power systems; Learning algorithms; Machine learning; Monte Carlo method; Monte Carlo simulation; Power consumption; Power system; Random sequences; Reliability; Reliability analysis; Reliability engineering; Shortages; Support vector machines; System effectiveness; System reliability
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2020.107171

•Reliability is an important property of power systems.•The Monte Carlo method is used to evaluate the reliability of large systems.•Sobol sequences increase computational efficiency in evaluating reliability.•Machine learning methods increase the speed of reliability assessment. The reliability of energy systems is assessed to control their operation and expansion. An effective method for reliability assessment is the Monte Carlo method. This process, however, is often time-consuming due to the large size of the power system. This interferes with subsequent control problems. The speed of reliability assessment and the accuracy of the result for the Monte Carlo method directly depend on the number of randomly generated states of the system, their quality and the complexity of the subproblem to be solved for each state. When solving such a subproblem for reliability assessment, random states can be defined as a shortage and shortage-free ones. To assess the reliability of power systems using the Monte Carlo method, one should analyze only the state of the system with a shortage. We suggest the use of machine learning methods to eliminate or sort the shortage and shortage-free states. The paper demonstrates the effectiveness of two methods: a support vector machine and a random forest. It also shows their performance when the Monte Carlo and quasi-Monte Carlo methods are used.