| Accurate forecast of short-term electrical load is very important to the security ofpower system and economy. Especially under power market condition, loadforecasting is becoming even more important, not only for system operators, but alsofor market operators, transmission owners, and any other market participants. Thusthe high forecasting accuracy and speed are required not only for reliable systemoperation, but also for adequate market operation, as both under-forecasts andover-forecasts would result in increased operational costs and loss of revenue. So it isnecessary to research advanced theory of intelligent technology.Based on the load data of meritorious power of Xian yang area power system,this paper establishes two models, ANN and WNN, to carry out load forecasting work,and compare the results. Since the traditional BP algorithm has some unavoidabledisadvantages, such as slow training speed and possibility of local minimizing theoptimized function, an optimized L-M algorithm, which can accelerate the training ofneural network and improve the stability of the convergence, should be applied toforecast the reduction of the mean relative error. Meanwhile, because the badforecasting of the error of peak load is based on the model of LMBP, this thesisintroduces a MRA & LMBP model. Through applying this model, the load serials aredecomposed to different sub-serials, which show the different frequencycharacteristics of the load and show significant regularity and periodicity than theoriginal load serials. Meanwhile, an artificial neural network is constructed to predicteach sub-serial corresponding to its characteristics. Finally, we achieved theforecasting result by reconstructing all predicted results of sub-serials together.The simulation results, which have a brightly applicable future, proved that thecombined model indeed improved the predicting precision, especially developed theprecision of daily peak load in certain degree.
|
PDF Full Text Request