A reliability forecasting method for distribution systems based on support vector machine with chaotic particle swarm optimization algorithm

In this paper, support vector machine (SVM) technique is applied to predict the reliability of power distribution system. To determine the SVM models´ optimal parameters for regression, particle swarm optimization algorithm is improved by combination with chaotic searching method (CPSO). The implementation approach of SVM for regression with CPSO (CPSO-SVR) is detailedly given. The CPSO-SVR models are first trained to learn the relationship between the influential factors of historical reliability and the corresponding reliability targets, and then future reliability can be predicted. In addition, a single but comprehensive index for distribution reliability is defined as IPSR. To examine the effectiveness of the proposed method, numerical experiments for the reliability forecasting of a city´s power distribution system in Southern China are conducted. The results reveal that CPSO-SVR outperforms the existing with higher forecasting accuracy and more robust performance. Hence, the proposed CPSO-SVR method is a proper alternative for forecasting power distribution system reliability. Furthermore, sensitivity analyses of input influential factors are demonstrated.