Long-term Load Forecasting And Effects Of Uncertainty In Related Elements

In order to reflect the influence of each element on the load forecasting result, an Artificial Neural Network (ANN) based approach for long-term load forecasting is investigated. A three-layer back propagation (BP) network is proposed using the history data of loads and correlative factors as exemplar. An elimination based method and variance ratio test based method are used for selecting inputs for the proposed ANN. The products of random initial value matrixes and input elements are used as the modified inputs of the ANN to improve its performance. The influence of the uncertainty of each element on the load forecasting result is researched. When the probability density functions of all elements are normal,the influence of the uncertainty of each element on the load forecasting result can be determined by the derivative of the forecasted load with respect to the corresponding element. On the basis of which, the range of the forecasted load is obtained. To general situations,a new concept of sampled blind number is defined with its values expressed by the averaged values of the corresponding intervals. Sampled blind numbers are adopted to describe the uncertainties. According to the operation properties of blind numbers and the load forecasting results under various probable cases, the probability density function and the confidence interval of the forecasted load are worked out. The load of Shanxi province is forecasted using the proposed method. Analysis results show that the best effect is obtained when six factors are selected as inputs for ANN, such as GDP, heavy industry production, light industry production, agriculture production, production of primary industry and tertiary industry. The proposed method is proved to be feasible and accurate by comparison the forecasting results to the actual data.