Prediction for down-hole working conditions of beam pumping units is an effective strategy to timely control oil well's working state

and is important to improve production efficiency and to reduce maintenance cost.Chaos theory was used in transient prediction for oil well's down-hole working conditions.First

moment eigenvalues of invariant curves were extracted from dynamometer chart as predictive variables.Then

after data sequence of these predictive variables were proved to have chaotic characteristics

chaotic time series prediction model was established by ELM(kernel extreme learning machine) method and several uncertain variables of the model were optimally solved by IFOA(improved fruit fly optimization algorithm) with two strategies of global population diversity-evolution and local individual random-variation.Finally

model predictive results were analyzed to determine fault types according to MEA(matter-element analysis) method.Case study of two oil wells in one oilfield showed that the IFOA-KELM-MEA prediction model was reasonable and effective.