基于集合卡尔曼滤波动态优化CA模型参数的方法

Parameter Optimization for CA Model Using Ensemble Kalman Filter

传统元胞自动机(CA)模型的转换规则不随模拟过程的时间和空间而变化,难以模拟和表达非线性地理过程。提出基于集合卡尔曼滤波(EnKF)动态优化CA模型参数的方法,以提高模型对复杂地理过程模拟的适应能力。通过引入集合卡尔曼滤波到CA模型中,将模型参数与模型状态整合成一个联合状态矩阵(joint state matrix)。再把该矩阵与观测数据输入到EnKF更新方程中,计算出新的参数值,并自动更新到模型中,从而实现动态调整模型运行轨迹,以更好地适应城市发展的过程模拟。将此方法应用于东莞市的城市模拟试验中,优化后的CA模型能在单参数和多参数优化中正确地调整模型参数,使其迅速地收敛于真值并趋于平缓,也能降低模型误差并获得更好的模拟结果。

A new method to optimize the parameters in cellular automata（CA） using ensemble Kalman filter（EnKF） is proposed.Most of the existing methods assume that transition rules are invariant in the spatio-temporal dimension.The same set of transition rules and their parameters will be applied to any location and time,regardless of the possible changes to the simulation environment.The use of ＇fixed＇ rules will produce poor simulation results,especially when the simulation period is long.A new way was developed to calibrate CA model parameters using EnKF.The method merges the status and parameters into a joint status matrix,and then updates the parameters in the transition rules by assimilating observations.The method was also applied to simulate the development of a fast growing city,Dongguan.Studies based on the experiments indicate that the results are better in single parameter and multi-parameters optimization.It is capable of reducing the error obviously.Besides,it also can get better results and make a more accurate prediction.