Aiming at the problems of the existing energy-saving algorithms for wireless sensor networks that have large calculations, high energy consumption of terminal nodes, and low real-time data update of the upper computer, an adaptive sampling algorithm for sensor networks based on multi-step prediction was proposed. An autoregressive prediction model between the upper computer and the terminal node for simultaneous prediction was established. At the same time, the effect of adaptively changing the sampling step was achieved by comparing the forward multi-step prediction value of the prediction model with the data change trend fitting value. In order to verify the energy saving of the algorithm, the experiments were carried out on the ZigBee-based ship launching airbag pressure monitoring system platform. The results show that with the root mean square error of 0.089 2, the proposed algorithm saves 36.252% energy compared with fixed-period sampling, and saves 26.912% energy compared with traditional adaptive communication algorithms, and has better energy performance.