A Method to Assess Power Grid Operation Risk Based on Improved Multi-Layer Perceptron

Along with the scale expansion of power grid traditional operation risk assessment methods gradually can not satisfy real-time requirement, and in existing risk assessment methods based on machine learning technology the imbalanced sample in the real system has not been considered. For this reason, based on improved multi-layer perceptron (MLP) a method to assess power grid operation risk was proposed. Based on IEEE-RTS7 the risk data sample was generated, and according to four aspects, i.e., voltage out-of-limit, power flow overload, loss of load probability and power flow transferring, an index system, which could characterize current operating state of power grid and the influence of relative state change, was established to quantize the risk of power grid operation and according to the value-at-risk the risk data sample was labeled to construct power grid risk data set. Considering the sample unbalance in real power grid, multi sample balance methods were led in and by means of feature selection and principal component analysis (PCA) the data dimension reduction was performed. Finally, the sample was trained by the improved MLP model to obtain power grid operating risk assessment and calculation model. Using the obtained model, while the training speed was accelerated, the representational ability for the nolinear rule in power data was intensified, thus, the result of risk assessment could be obtained rapidly.