Study On The Methods Of Electric Power Demand Forecasting In Distribution Network Planning

Electric power demand forecasting is the precondition and foundation of power distribution network planning, the level of forecasting precision directly affects the quality of distribution network planning and the safe reliable and economical operation of the distribution network; electric power demand forecasting is also an important part of power system modernizing operation and management. Distribution network planning demands not only the gross of future load, but also its position in the planning area, namely gross load forecasting and spatial load forecasting, and which have great guiding significance for a reasonable distribution network planning.Based on analysis of various gross load forecasting methods, considering that the model should solve less samples problem and multiple correlation problems in the prediction of annual electricity consumption, and that historical load samples each have asymmetrical status which should be assigned to different weightings, a weighted partial least squares regression (WPLSR) algorithm is presented. The specific modeling procedures are:1) analog deviation between the historical samples and predicting ones is computed,2) identify abnormal samples,3) adjust sample weights,4) partial least squares regression analysis. Analog deviation is introduced in the process of forecasting, thus achieves abnormal data identification and samples weighted forecasting, and avoids the case that bad samples and good ones are treated equally in traditional PLSR model. At last, the reliability and availability of WPLSR model are validated by an actual example.Based on analysis of various spatial load forecasting methods, considering that tradition methods based on experience or simple comparison can hardly meet the accuracy requirements in load density forecasting, a novel load density index methodology based on grey relational analysis and least squares support vector machine for spatial load forecasting is presented. The specific procedures are:1) establish elaborate load density index system,2) select the training samples which are more similar to the predicting one based on grey relational analysis,3) optimize the parameters of LSSVM prediction model automatically based on chaos particle swarm optimization algorithm. In the process of calculating the load density index, grey relational analysis and chaos particle swarm optimization algorithm are introduced to improve the generalization capability and adaptability of the LSSVM prediction model, further accuracy of the model. At last, the feasibility and availability of this model are validated by an actual example, and a novel and practical load density calculating method for distribution networks spatial load forecasting is presented.