Abstract: The problem of wide-band oscillations caused by the high proportion of renewable energy and power electronic equipment is becoming increasingly prominent. However, the existing oscillation monitoring methods based on synchrophasor data are limited by the current communication bandwidth, and it is difficult to monitor the wide-band oscillations from several hertz to hundreds of hertz globally. Therefore, a wide-area location method of wide-band oscillations based on the signal compression of the autoencoder and long short-term memory（LSTM） network is proposed, which uses the data compression and decoding capability of the autoencoder to realize the wide-area monitoring analysis of wide-band oscillation signals. Firstly, the power system measurement signals are encoded and compressed at sub-stations to realize the transmission of wide-band oscillation signals under the existing bandwidth, and effectively reduce the redundancy of oscillation data. Secondly, a feature matrix can be directly generated based on the compressed data, and the LSTM network can be adopted to locate the source of oscillations at the master station. In addition, the master station can decode the compressed data uploaded by the sub-stations, so the compressed data or decoded data can be used for the analysis and control of wide-band oscillations according to the requirements. Finally, the subsynchronous, supersynchronous, as well as medium and high-frequency oscillations are fully considered, and the load changes and random noise are taken into account for simulation. The results show that this method has a high reproduction, location accuracy, and good anti-noise performance.