Forecast Method of Multi-model Air Temperature Consensus in Tianjin

Based on genetic algorithm and particle swarm optimization, multi-model air temperature consensus forecast technology (MMATCFT) of hybrid evolutionary algorithm (HEG) is studied. The main technical thought of this method is that two integrated forecast models are set up respectively by using the genetic algorithm and particle swarm optimization, and then the final mixed forecasting model is established by the weight distribution scheme, which is founded through comparing forecast mean errors between the two models.In order to eliminate the impact of seasonal temperature characteristics of Tianjin, the daily rolling integrated forecast model based on 30-day data is adopted in practical operation applications with hybrid evolutionary algorithm. Using 2 m air temperature output data of four models of T639, GRAPES, TJWRF, BJ-RUC and observations of 35 automatic weather stations (AWS) in villages and towns of Tianjin from May to October in 2012, the forecast test of MMTCFT is carried out. Then the experimentation result is evaluated using the way of classification and station-separation, according to the meteorological standard that absolute error of temperature forecast is within 2℃. T639, GRAPES, TJWRF and BJ-RUC are separately run by China National Meteorological Center, Tianjin Meteorological Bureau and Beijing Meteorological Bureau. The analysis shows that the temperature consensus forecast model is effective and reliable. The technical scheme of the consensus forecast based on rolling model is more rational. The forecast errors are obviously smaller than any model mentioned above, and the forecast accuracy is higher. The average forecast accuracy of 6 h temperature, the daily maximum and minimum temperature in 35 AWS is 76.34%, 77.88% and 78.00% from May to October, respectively.