论文部分内容阅读
带输入项的线性自回归模型是一种综合性预测模型，较之常用的树木物候预测模型更为优越。模型结构属动态随机差分模型范畴；集中了线性自回归和多元线性回归模型两者的优点；模型有时滞，使预测值不但和现时刻输入（长期天气预报结果）有关，还受历史输入及自身滞后量的影响，即削弱了长期天气预报结果对物候预测的影响，提高了精度；模型参数的修正采用递推最小二乘估计法，参数随预测期数的增加而不断修正，使预测值更靠近真值（观测值）。从日本樱花、绯红晚樱、刺槐３树种预测误差对比可明显看出，新法预测误差总是稳定在１～２ｄ内，而不致于如其他方法预测误差那样在１～１１ｄ不定。说明新预测方法更加符合树木物候动态随机变化之实际。
The linear autoregressive model with inputs is a comprehensive prediction model, which is superior to the commonly used prediction models of tree phenology. The model structure belongs to the category of dynamic stochastic difference models; the advantages of both linear autoregressive and multivariate linear regression models are concentrated; the models have lags so that the predicted values are not only related to the current inputs (long-term weather forecasts) but also to historical inputs and to Hysteresis, which weakens the effect of the long-term weather forecast results on the phenological forecast and improves the accuracy. The model parameters are corrected by the recursive least squares estimation method, and the parameters are continuously revised with the increase of the forecast period so that the forecasting value is more Near the truth (observation). From the Japanese cherry, crimson evening cherry, black locust tree species error prediction comparison of three species can be clearly seen that the new method of prediction error is always stable within 1 ~ 2d, rather than in other methods as expected error in 1 ~ 11d indefinite. Indicating that the new prediction method is more in line with the actual dynamics of tree phenology changes randomly.