Recurrent neural network approach for stability analysis and special protection scheme of power systems with distributed generation

Power system stability and protection is important due to the complexity of power system, uncertainties in load, generation and integration of large number of renewable energy sources that forces the system to operate close to its stability limits. Voltage stability analysis (VSA) is a part of static stability analysis which involves performing power flow analysis (PFA). The Newton Raphson (NR) based PFA technique is conventionally used for VSA which requires formation and inversion of Jacobian matrix that increases the computational burden and requires large memory. Hence, a Jacobian less power flow technique using Recurrent Hopfield Neural Network (HNN) has been proposed for on-line contingency ranking (CR) and VSA. Furthermore, the potential of proposed Recurrent HNN is used for analyzing the frequeny stability of the power system by employing advanced controllers in automatic load frequency control (ALFC) application. The conventional design of gain parameters of proportional-integral-derivative (PID) controller has poor performance in case of large disturbanaces due to its static gain. By using the proposed Recurrent HNN method of tuning the PID controller, the gain values become self-adaptive to handle the system uncertainties and restore to steady state quickly. Moreover, to enhance the reliability and stability of the power system in case of large disturbances (like severe fault or contingencies) that leads to cascading failures or blackouts, a special protection scheme to detect the high impedance fault (HIF) has been proposed using Recurrent Long short term memory (LSTM) network as the conventional protection scheme fails to detect the HIF that occurs in the power network. The results obtained from the developed PFA technique reveal that the convergence time is improved by 32 % to 76 % than conventional approaches. In case of ALFC, the proposed h-HNN based PID controller is studied in single- and multi-loop (cascade) for multi-area power system. The results obtained prove that the proposed design of h-HNN based controller outperforms by 13.22 % to 98.55 %, 12 % to 99 %, and 18 % to 22 % in terms of steady state performance indices, transient performance indices, and control effort, respectively than other tuning methods. In terms of detection of HIF, the proposed Recurrent LSTM network method is validated in IEEE 13-bus power network integrated with solar photovoltaic system. The results obtained reveal that the proposed LSTM network gives the maximum classification accuracy of 91.21 % with a success rate of 92.42 % in identifying the HIF compared to other intelligence classifiers.

Text FK 2021 91 UPMIR.pdf Download (1MB)

Actions (login required)

View Item