Research On The Model And Method Of Medium And Long Term Power Load Combination Forecast

The medium and long term power load forecast is the main evidence for the power department in making such major decisions as the expansion outline of power grids, the stationing of power supplies, the dispatch of electricity generation, etc. Therefore, raising the precision of the medium arid long term power load forecast is of great significance to ensure security, economy and the excellent operation of power grids.Combination forecasting method is a more effective prediction strategy, which can take full advantage of the merits of each site prediction methods and contains useful information, and can also reduce the individual risk prediction model forecast.In the course of the traditional combination prediction, when making the electric charge prediction in different regions and different periods of time, the predictor often combines the prediction model based on his own subjective experience and the single prediction model construction on knowledge choice, which doesn't predict the strategy in a scientific way in order to direct the choice of prediction models. This paper firstly presents the necessity of introducing the idea of prediction decision-making in the course of combination prediction. Based on the merit and demerit of the current prediction model screening method which have been analyzed thoroughly in this paper, this paper, then, establishes a evaluation indicator system for prediction model on the basis of two indicators:Grey Relational Grade and prediction validity, innovates in putting forward some concepts like concordant gene, comprehensive validity indicator and so on, and sets up the prediction model screening method based on the comprehensive validity indicator and model redundancy verification.What is crucial to combination prediction is how to identify the weights of single prediction models of various kinds. Overcoming the demerit of fixed weight combination, the variable weighting combination prediction strategy that is presented in this paper is more practical, which produces better combination weight. The medium and long term variable weighting combination prediction model based on the comprehensive validity indicator system is presented in this paper, combined with the prediction model screening method based on the comprehensive validity indicator.An improved particle swarm optimization with immunity algorithms (IA-PSO) based on equity theory and adaptive adjustment is proposed to solve the shortcomings of IA-PSO for slow convergence rate and relatively low accuracy. On the one hand, through leading perturbation variables into the generation process of particle population, a balance is reached between the order and random behaviors. On the other hand, an adjustable mechanism of the adaptive particle velocity is proposed through the division of particle levels, which obtained by computing adaptive value.In view of the negative weights and not making different treatment on the influence of errors in different historical time periods on weights in traditional load combination forecasting models as well as the complexity of calculating variable weights, the authors propose a combination forecasting model using fuzzy adaptive variable weight based on fresh degree function and forecasting availability and apply it in the medium and long term load forecasting. The fresh degree function is adopted to embody the impact extent of historical data in different time periods.The analysis of instances proves that the prediction model screening method and two kinds of variable weighting combination prediction methods, which are presented in this paper on the basis of comprehensive validity indicator, have a better prediction effect in the course of the medium and long term electric charge prediction.