24-Hour Electrical Load Predictions by the Complementary Use of PLS and Neural Network

M. Lu and H. Xue (PRC)


Power system, load forecasting, artificial neural network, partial least squares, modeling, simulation.


Electrical load prediction is the foundation of real time control, security analysis and economic operation in power system. In this paper, a novel Hybrid PLS-ANN (HPA) model is proposed and utilized to eliminate the multicollinearity and redundance of the input data space. In the proposed HPA approach, Partial Least Squares (PLS) is used to reduce data dimension of predictor variables and to formulate the architecture of Artificial Neural Network (ANN) whereas ANN is employed to extend the nonlinear mapping capability of PLS model. The proposed model has the advantage of simple structure, high efficiency and good general capability, and provides faster convergence and more precise prediction results. After having completed learning or training, the model can be used to forecast electric load of 24 hours ahead. Field data from northwestern certain district in China are utilized to test the predictive power of the proposed model in Short Term Load Forecasting (STLF) in power system. Simulation analysis results showed that this approach could reach good predictive performance.

Important Links:

Go Back