Constrained neural network based identification of harmonic sources

Constrained neural nets are used to identify the location and magnitude of harmonic sources in power systems with nonlinear loads, in situations where sufficient direct measurement data are not available. This approach permits measurement of harmonics with relatively few permanent harmonic measuring instruments. A simulated power distribution system is used to show that neural nets can be trained to use available measurements to estimate harmonic sources. These estimates are constrained to conform to the available direct harmonic measurements, which improve their accuracy. It is shown that suspected harmonic sources can be identified and measured by a process of hypothesis testing.<>