Electric Power Load Profile Estimation Based On Blind Source Separation

Power system state estimation is the core of electric energy management system and the bases of dispatch, control, security evaluation and so on. The theory of state estimation in power system had established in the early of the 1970s, improving constantly in the nearing more than ten years.It is difficult to estimate load profiles in the deregulated environment where competing entities need to access the load demands based on the partial knowledge of the system. Traditional state estimation techniques can not estimate load profiles because they work on the premise that knowledge of the power system parameters or topological structure can not be required. In this dissertation, the load profiles are estimated by using a blind source separation technique called Independent Component Analysis (ICA) and Kernel Independent Component Analysis(KICA).Firstly,the current state of the load profile estimation and development of the state estimation are surveyed. Secondly, many problems about ICA and KICA are discussed, such as theory, research information and some basic algorithms. The individual bus loads are statistical correlation and Gaussian distributed. A filter and a Laplacian distributed signal are used to match the ICA model's conditions. The branch power is regarded as linear combinations of the loads. ICA and KICA use the limited number of branch power measurements to estimate load profiles.The proposed approach is demonstrated for the IEEE-14 system to estimate eight active load profiles, using FastICA algorithm, Infomax algorithm and KICA algorithm respectively. The results of experiments show that the correlation coefficients between estimated load curves and original load curves are larger than 0.9 when ICA model is used, but more accurate results are obtained by using KICA algorithm which makes the correlation coefficients are almost 1.0.At last, the discussion about the sensitivity of the estimation algorithm with respect to the number of data points and measurements, measurement noise and so on is presented. The experimental results show that the modification of FastICA algorithm which is presented in this thesis performs well in removing noises and makes the correlation coefficients are nearly 0.95.