| With the rapid development of national economy and the gradual formation ofelectric power market, power quality problems in many countries have been widelyconcerned about by electric power department and users. Harmonic is an importantaspect of power quality problems. On the one hand, the power industry of our countrygrows vigorously, the electrical loads increase sharply, the wide application of powerelectronic devices and a large number of nonlinear and impact loads make the gridinjected a lot of harmonic component and lead to severe distortion of voltage andcurrent waveform. On the other hand, with the rapid development of modern society,highly automated and intelligent industrial electrical equipments for power quality havebecome increasingly demanding. Therefore, in order to improve power quality, harnessharmonics and reduce harmonic pollution effectively and improve electrical gridefficiency and environmental, harmonic sources identification is used to determine theinjected harmonic current.Harmonic source identification methods at home and abroad are mainly dividedinto two types: based on multi-point measurement and based on single-pointmeasurement. Based on the multi-point measurement methods mainly include neuralnetworks and harmonic state estimation, etc. Method for a neural network model ismore dependent on training samples and the weight matrix is susceptible to the gridoperation, so it is lack of flexibility and adaptability. Without the knowledge ofharmonic source, making use of harmonic measurements from phasor measurementunits on the bus and line, harmonic state estimation is to infer the whole power gridharmonic voltage and harmonic current status and identify the sources. This method isrelatively effective, but it requires the entire network electrical parameters and thetopology structure, and component models must be accurate, otherwise there is a greatimpact on the estimation results. The method based on single-point measurement ismainly to estimate the utility harmonic impedance. The fluctuation method and thelinear regression methods are the non-invasive approach. The fluctuation methodestimates the utility harmonic impedance by the harmonic fluctuation ration of voltageto current and requires high precision measurement. The linear regression methodassesses the harmonic emission level based on voltage and current equations at PCC.Partial least squares(PLS) regression method can overcome the influence of the correlation between variables, but it does not consider the large error in measurement orother abnormal values.According to the analysis of the advantages and disadvantages of each method,harmonic identification method based on blind source separation for multi-pointmeasurement and harmonic emission level assessment method based on robust partialleast squares are proposed. For multi-point measurement, blind source separation can beused to recover each of the original signal when the source signal and hybrid systemscharacteristics are unknown. Due to ICA algorithm can remove the low and high ordercorrelation and obtain several mutually independent components, the fast ICA techniqueis used to realize blind source separation. Only using the measured data on the bus, fastICA can identify harmonic sources and determine harmonic current when the powernetwork topology structure and electrical parameters are unknown. As the mixed matrixestimated by Fast ICA represents the relationship of harmonic voltage in concernednodes and harmonic sources current, it is used to calculate harmonic contributions ofeach harmonic source. For single-point measurement, robust partial least squares iscombining fast MCD with partial least squares and improve the resistance to outliers inmodeling data using robust covariance matrix in fast MCD. The customer harmonicemission level is calculated with system harmonic impedance estimated by robustpartial least squares regression.The Matlab simulation of harmonic identification of IEEE14nodes test systembased on blind source separation was performed. Three harmonic sources and threemeasuring points are included in this text system, each harmonic source contains5th,7th,11thharmonic component and the measurement points are the concerned nodes.According to the simulation results, the separated results by FastICA are very similarwith the actual harmonic sources waveforms. Error and the correlation coefficientanalysis were carried out on the waveform data, we can see every harmonic error isaround3%, and the correlation coefficient is around0.9. This shows that the ICAalgorithm can estimate the power system harmonics. Finally the harmonic emissionlevel of every harmonic source in concerned nodes is calculated according to theestimated harmonic current.The computer simulation evaluates the harmonic voltage emission level using theMatlab software. The results of the utility harmnic impedance are estimated in threedifferent cases. In case1, the estimation results are computed based on PLS regression.In case2, some outliers are added in sample voltage and current points artificially. The results are estimated with these data based on PLS again. In case3, the results areestimated based on robust PLS with outliers. The simulation results show the sensibilityto outliers of PLS and effectiveness and robustness of robust PLS. In addition, extensivefield measurements on35kV substation in Jiangjin are used to assess the harmonicemission level of a steel. The example calculation results further verify the accuracy andfeasibility of the method. |
PDF Full Text Request