Research On Load Characteristic Classification And Parameter Identification For Load Modeling

Power system modeling is the typical scientific problem and core technology in areas such as power system analysis, design, operation and control. Load modeling has become a critical problem that is urgent to be solved in power system modeling. It is of great importance to build an accurate load model for power system digital simulation. Thus, this thesis mainly focus on two aspects. One is the load characteristic classification, the other is the parameter identification of load models. Main works of this thesis are listed as below.(1) Further research on load characteristic classification of substations. SOM neural network is applied for classification of feature vectors constructed using load composition data. Then typical substations of each class have been chosen according to the classification results. Further research has been done for classification of newly added substations. The accuracy and precision of classification results using the SOM neural network has also been proved through simulation analysis.(2) Research on load dynamic characteristic classification. Common feature vectors of dynamic load characteristics are analyzed and a feature extraction method based on lifting wavelet packet transform is proposed for load dynamic characteristic classification in this dissertation. The practical data of load current response is decomposed and reconstructed, then the wavelet packet coefficients can be extracted to construct energy moment based feature vector. The validity and practicality of the proposed method have been proved by feature extraction and classification test using simulation data and field measurement data. Compared with traditional wavelet packet transform, the lifting wavelet packet transform has shown advantages both in computational speed and reconstruction error, and can improve the accuracy of load dynamic characteristic classification.(3) Considering the global search ability of the quantum delta-potential-well-based particle swarm optimization (QDPSO) algorithm and the local search ability of the chaotic optimization algorithm (COA), a hybrid optimization algorithm that combines the both algorithms is proposed to identify the parameters of load models. Then, exponential+difference equation based load model is selected and practical data collected by power fault recorder are used to identify the parameters of the selected load model. Thus, the validity of the proposed method have been proved. Compared with the QDPSO algorithm and the PSO algorithm, the hybrid algorithm has shown advantages both in convergence speed and convergence precision.Load characteristic classification of substations based on the SOM neural network, load dynamic characteristic feature extraction and classification based on lifting wavelet packet transform can avoid simulation errors caused by adopting the same load model structure or parameters. Application of the hybrid optimization that combines QDPSO algorithm and COA for load model parameter identification can improve the precise of load model parameter identification. The research is of importance both in theory and practice to improve the accuracy of load models.