Technology Of Short-Term Power Load Forecasting

Short-term load forecasting is one of the most important procedures in the real-time control of power generation and efficient energy management systems. It is used for establishing the power station operation plan and the unit operation plan, together with spinning reserve planning of energy exchange. This paper presents a new forecasting technique by combining several traditional methods for the accuracy load forecasting.To avoid the disadvantages of single forecasting method, a short-term load forecasting technique is used which is based on combination of Wavelet Transform, Time series Analysis and Neural Network.Firstly, Mallat Algorithm is used for wavelets decomposition and reconstruction. According to the characteristics of different kinds of Wavelets, Daubechies 4 (Db4) Wavelet is chosen as the wavelet orthogonal bases to decompose load series at level 4. As a result, one approximation series in low frequency domain and four details series in high frequency domain are obtained successively.Secondly, due to the periodicity of the approximation series in low frequency domain, AR linear model is chosen for low frequency series forecasting and its rank is determined by correlation analysis. Due to the randomness of the details series in high frequency domain, RBF Neural Network is chosen based on Gaussian core function to establish the forecasting model. The weights of this network are adapted by using gradient descent method. The neural network is trained to obtain high frequency forecasting data.Finally, the whole forecasting load curve is obtained by combining the forecasting data of approximation and details series.The effectiveness of the proposed technique is validated by experiments with historical load data in DUT substation, and it can also be proved better than the single forecasting method.