Radio-frequency capacitance sensor is adopted in system of water-content detection in oil product

which is based on the principle that the dielectric constant of water is much higher than that of oil products

and thus the reflected specific properties are different in the respect of radio-frequency impedance.In order to solve the problems of the non-linearity characteristic of radio frequency capacitance sensor and the sensor’s cross sensitivity to temperature in the process of water content detection in oil product

the two-dimension non-linearity compensation of sensor adverse model based on multi-layer forward neural network using L-M arithmetic is brought forward.Anti-pulse mean filtering and the scope limited filtering arithmetic are used on sampled data to insure optimum effect of neural network training.The experiment conclusions validate the method’s feasibility and practicability.