Research Of Models On Forecast Of System Marginal Price In Spot Market

In power market, system marginal price(SMP) is the product price of an individual energy enterprise whose profit relies on successful bidding policy, and bidding policy is generally based on accurate mastery of short-term market tendency whose key is to carry out forecast of SMP. Therefore, research on forecast of SMP is an urgent task for all individual energy enterprises in competitive circumstances. Meanwhile, from the point of customers, SMP forms their unit cost of using energy and forecast of SMP makes dynamic cost control possible; from the angle of market supervisors, forecast of SMP promotes the healthy, stable, and orderly development of energy market and supplies foundations for establishment of all energy price policies. Accordingly, forecast of SMP is significant for three kinds of participators in power market.Combining with the requirement of power enterprises' bidding in power market, in this thesis, several sorts of models for spot price forecast are proposed:Based on historical price materials, from the point of time-serials analysis, models of fuzzy exponential flatness, linear remove and auto-regression, and mean generating function are constructed to probe the tendency and regulation hidden in price materials and to forecast spot price next day.With the thought of econometrics, models of three-points, removing probability of Markov, and Holt-Winter are applied to forecast of tomorrow spot price.Considering influencing factors and changing discussion angles, this thesis also forms models of ANN based on weather sectors, holiday price forecast based on selection of similar year, and combination of fuzzy categorization and statistic correlation.According to historical price materials, through introducing combining thought, model of optimal combination and model of linear remove combining with exponential flatness are used to forecast tomorrow spot price.Through simulating forecast, considering the predicting accuracy and stability, author thinks that models of linear remove and auto-regression and linear remove combining with exponential flatness do a better job in prediction of spot price.