基于遗传算法的门限自回归模型在海温预测中的应用

Application of threshold autoregressive model based geneticalgorithm for forecasting marine temperature

提出了建立门限自回归模型（ＴＡＲ）的一套简便通用的方法。用作者提出的改进遗传算法，可同时优化门限值和自回归系数，成功地解决了ＴＡＲ建模过程所涉及的大量复杂寻优工作这一难题，为ＴＡＲ模型的广泛应用提供了强有力的工具。实例计算的结果说明了这套方法的可行性和有效性，同时也说明了，通过门限值的控制作用，ＴＡＲ模型可以有效地利用如海洋资源所隐含的时序分段相依性这一重要信息，限制了模型误差，从而保证了ＴＡＲ模型预测性能的稳健性，提高了预测精度。该方法具有通用性。

A simple and general scheme is presented for establishing threshold autoregressive (TAR) model. With the improved genetic algorithm, both of threshold values and autoregressive coefficients may be optimized, and the difficulty point of modeling of TAR was resolved, which gives a strong tool for widely using TAR model. The result of the calculation example shows that the scheme was practical and efficient, and that TAR model can effectively utilize the important information of the section interdependence with time series such as marine observation dates by controlling the threshold values, reduce model errors, and ensure good stability and accuracy of the forecasting model. As a general method, the scheme has major theoretic value and wide application for predicting of the nonlinear time series.