Obtaining accurate information of temperature distribution plays an important role in chemical industry. As a result of advantages such as low cost and non-intrusive sensing

acoustic tomography (AT) is considered to be a promising visualization measurement method for temperature distribution. An SA-ELM algorithm was proposed to improve reconstruction quality of temperature distribution after AT measurement. First

robust estimation was used to establish the L1 norm objective functions. Then

the objective functions were solved to obtain temperature distribution on coarse discrete grids by simulated annealing algorithm (SA). Finally

temperature distribution on fine grids was predicted by extreme learning machine (ELM) method. Numerical simulations and experimental study showed that the SA-ELM method could improve quality and robustness of temperature distribution reconstruction. Hence

an effective new method is developed for solving reverse challenge in AT measurement.