For the iterative image reconstruction algorithm of electrical capacitance tomography (ECT), linear forward problem solution is usually adopted to speed up image reconstruction. However, image reconstruction error is inevitably produced. In this paper, a nonlinear forward problem solution based on extreme learning machine (ELM) of ECT is proposed. The inputs and outputs of ELM network are permittivity distribution and predicted capacitance measurements, respectively. Image reconstruction is carried out based on the combination of the presented method and conventional Landweber iterative algorithm, which is named as ELM-Landweber iterative algorithm. In order to make the samples more representative, the distribution positions and sizes of objects in each phantom are randomly generated, and the corresponding normalized capacitance values are calculated as ELM network training and test samples. Simulation and static experiments are conducted for ELM-Landweber iterative algorithm and the reconstructed images are compared with those of conventional Landweber iterative algorithm. Experimental results show that the convergence speed of ELM-Landweber iterative algorithm is significantly enhanced, and the quality of the reconstructed image is obviously improved compared with conventional Landweber iterative algorithm. The average image relative error of training samples and test samples decreases from 0. 728 to 0. 504 and from 0. 596 to 0. 475, respectively.