Application of synthetically smooth algorithm in power load sequence pretreatment

Since neural networks are sensitive to inaccurate data, historical load data must be executed pretreatment before load data is input into forecasting model. By analyzing the exceptional wave data of historical load data, continuous curve of primitive load data is obtained to discrete load curve. The basic conception about load data pretreatment is adopted. First, primitive load data curve and typical load curve are compared. The elimination of different point data and the smoothing of load curve are proceeded to realize the vertical processing. Then, using the method of median data generates a smooth estimated sequence for primitive load sequence. Before load horizontal processing, deviation rate about practical load sequence comparing to smooth estimated sequence should be got. By this way, a smaller and more precise input variables set can be gained by this method. Simulation experiment verifies validity of the algorithm.