An artificial-neural-network approach for the identification of saturated turbogenerator parameters based on a coupled finite-element/state-space modeling technique

This work centers on developing a methodology to accurately represent the saturation effects in cylindrical rotor synchronous machines (turbogenerator) and to identify, with a high degree of accuracy, the saturated turbogenerator parameters over a broad range of steady-state operating conditions. A modeling technique has been developed to predict the machine characteristics and accurately compute the saturated parameters using finite-element field solution in conjunction with state-space models in the abc frame of reference. This Coupled Finite-Element/State-Space (CFE-SS) modeling technique provides a powerful tool for the design and analysis of such turbogenerators. The CFE-SS technique is used in conjunction with an artificial neural network (ANN) to construct a representative training set of turbogenerator saturated parameters under different load conditions to perform the training of the ANN. This trained ANN can be used to interpolate between discrete finite-element based machine parameters and can provide initial data and parameters for use in power system stability studies.; In the CFE-SS modeling environment, the natural abc frame of reference is retained to represent armature winding currents which are embedded in their proper locations in the various stator slots according to the actual three-phase armature winding layout of any given turbogenerator. Two-dimensional finite-elements (2D-FE) are used to model the magnetic field profile over a complete ac cycle of steady-state operation. This method rigorously incorporates the full impact of space harmonics in the inductances caused by the turbogenerator geometry and the continuous relative rotor to stator motion, as well as the impact of magnetic saturation. Thus, the effects of rotor slotting and cylindrical rotor magnetic saliency, as well as armature slotting on space harmonics in the flux distribution, under saturated conditions, can be accurately accounted for in the turbogenerator model and resulting parameters.; The CFE-SS modeling technique is used to predict load characteristics and to compute the saturated parameters of a case-study 20-kV, 733-MVA, 0.85 pf (lagging) turbogenerator. The simulation results are in excellent agreement with the test results showing the strong validity of the CFE-SS based computed parameters. This modeling environment is then used to compute the machine saturated parameters at different judiciously chosen load points on the operating P-Q plane of the turbogenerator for three different levels of terminal voltage. These computed parameters constitute a data base available for training an ANN. A multilayer ANN structure has been developed and successfully trained using this data base. The back-propagation algorithm is used to perform the training of the neural net. The trained ANN is then tested for generalization by presenting it with arbitrary load points which are not in the data base. The ANN-computed parameters are cross-checked by recomputing them at the same load points using the CFE-SS technique. Results show that ANN-computed parameters are in excellent agreement with the CFE-SS-computed parameters for the same load points to within an error of less than 2%. (Abstract shortened by UMI.)...