The invention provides a non-stationary differential weighted superposition seismic data processing method. The non-stationary differential weighted superposition seismic data processing method comprises the following steps: 1, performing Gaussian filter processing on original data to reduce an influence of noises on a signal of a reservoir; 2, performing sliding three-point smoothing processing on a seismic channel; 3, calculating an even-order derivative channel based on the smoothed data to highlight high-frequency information of a thin layer and a thin interbedding layer; 4, performing standard processing on a high-order derivative channel and amplitude-preserved processing based on a Gaussian window so that the processed data longitudinally has relatively high amplitude preserving performance; and 5, performing certain weighted superposition on the original data, the smooth data and the high-order derivative channel to acquire a final processed result. The non-stationary differential weighted superposition seismic data processing method provides high-quality data for carrying out subsequent attribute extraction, oil and gas reservoir identification and prediction of the reservoir, so that the exploration and development capabilities of the oil and gas reservoir of the thin layer and the thin interbedding layer are improved.