Particle swarm optimization (PSO) is a global random search algorithm put forward by simulating the flock foraging in the process of social behavior based on swarm intelligence. Researchers have proved that PSO algorithm is an effective geophysical inversion method, and it does not rely on the initial model. Because the conventional PSO is easy to be stuck in relative extremum, slow convergence speed in the late and the inversion accuracy is not high, this paper presented an improved fully chaotic oscillations particle swarm optimization algorithm based on same conventional PSO theory. It improved the formula of updating speed, made the particles getting the difference between the current global best position quickly, enhanced the learning ability of particles. The paper did a two-dimensional numerical test on ERT data in matlab2012b programming environment,the results show that this algorithm inversion is not dependent on the initial model, increases the search space,and have higher inversion in accuracy than the standard PSO, and the image quality is better than that of Levenberg-Marquardt method.