Abstract: A chaotic parameter-optimized radial basis function (RBF) forecasting model was proposed. The chaos theory was used to obtain the embedded dimension and delay time of the ENPEMF， and the obtained parameters were used to optimize the RBF neural network. Finally， the trained optimized-RBF was utilized to forecast the strength trend of 14 d ENPEMF data. Numerical simulation results show that the improved RBF model could forecast the Rossler time series well with small error. Applying the improved RBF algorithm to the ENPEMF data before Ms7.0 earthquake in Lushan， it can effectively forecast the ENPEMF intensity trend 14 d before earthquake. The forecasting effect and accuracy are significantly better than that of the traditional RBF algorithm， which is expected to provide support for electromagnetic monitoring and analysis before earthquakes and geological disasters.

Key words: the Earth’s natural pulse electromagnetic field, intensity trend forecasting, chaos theory, parameter optimization, RBF neural network