Dynamic Load Modeling Based on Extreme Learning Machine

摘要: The dynamic load characteristics have significant impact on the power flow, transient stability computation, voltage stability calculation of the power system, and so on. Noticing that traditional mechanism loads model has difficulty in precisely describing the dynamic characteristics of synthetic load, this paper presents a non-mechanism dynamic load model based on Extreme Learning Machine (ELM). The Power Fault Recorder and Measurement System (PFRMS) is used to obtain data for load modeling. Take voltage and real/reactive power with different time delay as inputs, and take real/reactive power as output, train the ELM using the samples formed by fault data, the real power model and reactive power model are established respectively. The number of hidden layer nodes which has impact on the ELM model is also discussed. Dynamic simulation experiment is conducted at power system dynamic simulation laboratory. The simulation result shows that the ELM load model is simple and flexible, its parameters are easy to be identified. The ELM load model can describe the dynamic load characteristics accurately.