Contents I. Introduction
State estimation is currently used in power systems at the fundamental frequency to provide accurate estimates of power flows [1] and [2]. State estimation allows us to obtain more accurate information from noisy, redundant measurements and values at non-metered buses. This same technique can also be applied to harmonic frequencies. Modern power systems have many harmonic sources and, as a result, it would be prohibitively expensive to monitor them all directly. Thus, state estimation for harmonics was introduced in [3], but the technique also required a large number of monitoring stations. Using artificial neural networks (ANN s) to provide pseudo-measurements for the state estimation was suggested in [4]. In this method, harmonics currents are monitored on various lines around the system. These measurements along with the power flow information are fed into neural networks. The outputs from the neural networks consist of estimates of the harmonic currents at other busses and loads. These estimates, called pseudo-measurements, are then verified and improved using the actual measurements through state estimation. The pseudo-measurements, in effect, replace actual monitoring stations. A similar method is presented in this paper, except both voltage and current measurements are taken at metering points in the network (as most power quality meters record both) whereas the method in [4] utilised current measurements only. A different neural network structure than [4] investigated in this paper. The test system also contains stochastic loads, representing real power systems more closely.
