Piezoelectric high-entropy energy devices have been in service for a long time in the complex environment of intelligent connected road, and have experienced problems with device anomalies and randomness in output during energy harvesting. To determine the availability of piezoelectric devices, this study characterizes their output characteristics and proposes a definition for availability of piezoelectric devices. According to this definition, a reliability model considering competitive failure mechanisms is established. Furthermore, a deep learning model based on group characteristics is built up, achieving real-time assessment of the anomaly degree of piezoelectric devices. The effectiveness of the proposed availability evaluation method is demonstrated through case analysis. The results indicate that this method effectively manages uncertainties in evaluating the operational state of piezoelectric devices. For the tested devices experiencing output power attenuation exceeding 30%, the anomaly detection achieves a precision of 0.928 and a recall of 0.775, thereby providing a novel approach for monitoring anomalies in such devices.