A dual-factor auto-regressive integrated moving average model (ARIMA model) has been established to tentatively improve the low accuracy of the ARIMA model in forecasting modern urban cooling loads, followed by a simultaneous predication testing by these two models. A final comparison between the two prediction results shows that the new model, by introducing an average temperature operator to the forecasting process, has effectively improved the urban cooling load forecasting accuracy with a universal applicability on its former basis of an ARIMA model.