Power output is one of the main working parameters of a diesel engine, which not only reflects its dynamic performance but also acts as an important indicator of its operating state. At present, diesel engine power monitoring mainly depends on dynamometers, but most engines cannot be so equipped during operation. In addition, other techniques, such as in-cylinder pressure measurements used to gauge power output are invasive and have higher requirements for sealing and reliability. Therefore, it is important to study other means for achieving real-time monitoring of engine power that do not damage the engine. In this paper, we use the vibration acceleration of the combustion phase in the angle domain as the original signal to identify the output power. However, the signal contains many interference signals and must be pre-processed by purification techniques such as filtering, de-noising and empirical mode decomposition. Finally, the automatic identification of the power of a diesel engine is realized by means of inductive monitoring system algorithms. The test results show that the recognition accuracy under the five different working loads ranges between 83% and 95%, which demonstrates the effectiveness of the proposed method.