A fuzzy neural network approach for power system evaluations

Every real-world dynamical system is nonlinear. Existing methods for solving a nonlinear problem entail linearizing the nonlinear problem and then using the different tools available for solving the linear system. These tools have been well understood for many decades. This research presents the application of Fuzzy Neural Network in reducing the large computational requirements associated with solving nonlinear systems. This approach utilizes fewer and faster steps for solving nonlinear problems. A practical use of this technique is power flow calculation where a large number of nonlinear equations are involved. The power flow problem is formulated as a nonlinear constrained optimization problem with the bus voltages, bus angles, real power injected into the buses, and reactive power injected into the buses as the problem variables. The equality and inequality constraints are appended to this objective function. Fuzzy rules based control is used to assist in choosing suitable penalty functions to form an augmented cost function. The linearized power flow equations at each iteration are translated to a scalar objective function of quadratic form. A neural network structure is given which implements the steepest descent method for minimizing the objective function.; This research also presents the application of Fuzzy Clustering to power systems. The technique of Fuzzy Clustering reduces large system states into a few representative clusters, which are sufficient for reliability analysis. The method is then shown for optimal network decomposition based on Fuzzy Clustering. Fuzzy Clustering presents a powerful, globally oriented optimization method, which exploits the mechanism of natural response to reach optima or near optima. The results for an IEEE 14-Bus test system are given and the Fuzzy Clustering algorithm approach is found to produce significantly better solution.