基于遗传算法和高斯牛顿法的超声回波信号参数估计

在超声回波信号参数估计中，如果高斯牛顿法选取的迭代初值接近参数向量的真实解，则容易找到最优解；如果初始值远离最优解，则高斯牛顿法不收敛或者只收敛到局部最优解。针对高斯牛顿法对迭代初值敏感的问题，提出了遗传算法和高斯牛顿法结合的参数估计方法。该方法充分利用遗传算法善于进行全局搜索和高斯牛顿法善于进行局部快速搜索的优点，首先使用遗传算法求出超声回波信号的参数初值，然后利用这组初值进行高斯牛顿法迭代搜索。仿真结果表明，基于遗传算法和高斯牛顿法相结合的方法，具有收敛速度快、精确度高的特点。

Parameters estimation of ultrasonic echo signal based on genetic algorithm and gauss newton method

In ultrasonic echo signal parameters estimation tasks, it is easy to obtain the optimal solution when the selected initial values for Gauss Newton iteration is close to the real parameter vector. Otherwise, the algorithm does not converge or only converges to local optima. Gauss Newton Method is sensitive to initialization in parameters estimation of ultrasonic echo signal. To address this issue, a combined method of Genetic Algorithm (GA) and Gauss Newton Method (GNM) was proposed. GA is effective in global solution space searching, while GNM is effective in local searching. Full use of the merits of both GA and GNM to parameters estimation of ultrasonic echo signal, GA was used to solve the initial values of the ultrasonic echo parameters at first. Then, GNM iteration search based on these initial values was adopted. Simulation results show that the combination of GA and GNM obtains a fast convergence and a high precision.