PSO-ELM辅助的GNSS-IR土壤湿度反演方法

PSO-ELM Assisted GNSS-IR Soil Moisture Retrieval Method

针对如何有效提高全球导航卫星系统多径干涉遥感(Global Navigation Satellite System Interferometric Reflection,GNSS-IR)技术观测量数据反演土壤湿度精度的问题,提出一种基于极限学习机(Extreme Learning Machine,ELM)模型的土壤湿度反演方法,并利用粒子群优化(Particle Swarm Optimization,PSO)ELM来获取该模型的最优参数。以GNSS-IR提取反射信号的相位作为输入向量,以PBO H 2 O的土壤湿度值作为输出向量,构建PSO-ELM神经网络模型,并进一步与ELM模型、BP神经网络模型和线性回归模型进行对比分析。实验结果表明,PRN10卫星在PSO-ELM模型中的土壤湿度反演结果与土壤湿度值之间的决定系数为0.8771,均方根误差为0.0252,平均绝对误差为0.0207,相比其他3种模型的土壤湿度反演精度更高、稳定性更强,具有较强的拟合能力,证明了该模型的可靠性和优越性。

In order to improve the accuracy of soil moisture retrieval from Global Navigation Satellite System Interferometric Reflection(GNSS-IR)observation data,a soil moisture inversion method based on Extreme Learning Machine(ELM)model is proposed,and the optimal parameters of ELM model are obtained by using Particle Swarm Optimization(PSO)optimization.The phase of reflected signal extracted by GNSS-IR is taken as the input vector,and the soil moisture value of PBO H 2 O is taken as the output vector to construct the PSO-ELM neural network model,which is further compared with the ELM model,BP neural network model and linear regression mode.The experimental results show that:The determination coefficient between the retrieval result of soil moisture from PRN10 satellite in PSO-ELM model and the true value of soil moisture is 0.8771,the root mean square error is 0.0252,and the mean absolute error is 0.0207.Compared with the other three models,the soil moisture inversion accuracy is higher,the stability is stronger,and the fitting ability is stronger,which proves the reliability and superiority of this model.