Research On Load Characteirstics Clustering And Model Identiifcation

With the gradual expansion of the current power grid size, its structure hasbecome more and more complex, and it is significantly important for power systemload modeling. Due to the particularity of the power load, establishing an accurateload model is considered to be a difficult problem in power system’s engineeringpractice and academic research. Firstly this paper focuses on the importance of theload modeling research, secondly deeply analyzes the basic theory and methods ofload modeling, finally the load characteristics clustering and model parametersidentification are mainly studied in two aspects, the specific contents are as follows:To establish a proper load model for substations, a comprehensive loadcharacteristics clustering algorithm based on ACO-PAM and SA-RS-K-means arepresented in this paper. The ACO-PAM algorithm applies PAM to make clusteringanalysis for the history optimal position of ACO, and then insteads the referencepoint for the new clustering center. The data can be clustered adaptively to theclassification. The SA-RS-K-means algorithm improves K-means based on simulatedannealing, and the K-means cluster numbers and initial cluster centers can beobtained dynamically with maximum and minimum principle, and boundaryobjectives are dealt with upper and lower approximation of rough set theory. In theend, the two comprehensive algorithms and a single FCM are applied for substationsclustering, and the inter-class distance and the intra-class distance are compared.The results verify the feasibility and validity of the proposed two synthetic methodsfor power system load characteristics clustering.The obtained clustering results and clustering center matrix can be appliedfosubstations classification, and the typical substation can be selected for eachclassification to install the load measuring device, then the proper load model basedon field data is established. The GA-dynamic modification method is presented forpower function model identification ignoring the frequency change. The simulationresults prove the established load model using the method can effectively reflect theactual static load characteristics. At the same time, the third-order electromechanicaltransient induction motor and constant impedance synthesis load model parametescan be identified combined with the PSO and GA algorithm. The simulation resultsverifiy the PSO-GA algorithm for load identification has a higher fitting precision and describes the characteristics of comprehensive load, which can providesignificant reference for practical load modeling.