基于PSO-ELM算法的红外目标模拟器校准数据拟合方法研究

Research on Calibration Data Fitting Algorithm of Infrared Target Simulator Based on PSO-ELM Algorithm

提高红外目标模拟器校准数据的拟合精度,对于红外目标的辐射照度等辐射特性的测量有着重要意义;针对校准数据具有很强的非线性,传统的拟合算法精度不高的问题,引入一种基于粒子群算法优化的极限学习机算法(PSO-ELM),以标准黑体辐射温度作为输入因子,以MCT探测器实际测量出的辐射照度作为输出因子,建立PSO-ELM模型,利用粒子群算法(PSO)对连接隐藏神经元和输入层的权值和隐藏神经元阈值进行优化,拟合出输入参数和输出参数之间的非线性关系;这两个参数的优化提高了极限学习机算法(ELM)的性能,该方法的主要优点是具有较强的容错性、较好的对复杂非线性数据处理性能和ELM算法参数设置上的优化机制;通过与GA-ELM模型、ELM模型进行对,验证了与传统数据拟合方法相比,基于PSO-ELM的方法拟合精度有了很大提高,为红外目标模拟器校准数据拟合提供了新的方法。

Improving the fitting accuracy of calibration data of the infrared target simulator is of great significance for the measurement of irradiance and other radiation characteristics of the infrared target.In view of the strong nonlinearity of calibration data of the infrared standard simulator and the poor accuracy of the traditional fitting algorithm,a particle swarm optimization extreme learning machine(PSO-ELM)is introduced in this paper.The standard black body radiation temperature is taken as the input factor and the irradiance,which is actually measured by the MCT detector as the output factor,and the PSO-ELM-based model is established.In the PSO-ELM-based method,the connection weight matrix from the input layer to the hidden layer and the bias vector of the hidden layer are optimized by the PSO algorithm,and a nonlinear relationship between input parameters and output parameters is fitted.The optimization of these two parameters has greatly improved the predictive ability of original ELM.This method has main advantages of strong fault tolerance,better processing performance for complex nonlinear data,and the optimization mechanism in a kernel parameter setting of ELM.Comparing with genetic algorithm extreme learning machine(GA-ELM),extreme learning machine(ELM),The superior fitting accuracy of the PSO-ELM-based method is verified in contrast to the conventional data fitting method,which provides a new method for infrared target simulator calibration data fitting.