Abstract: This paper addresses challenge issues of current sensor fault diagnosis methods, including dependence on expert knowledge, ignorance of spatiotemporal correlations with bypass terminals and redundant feature impacts. An approach for sensor health assessment is consequently proposed, utilizing information fusion and one-dimensional convolutional neural networks. Firstly four types of sensor characteristics are selected according to the strong correlation with photovoltaic power prediction, which are statistical features of sensor data streams, temporal characteristics of sensor data streams, data characteristics of bypass terminal, and weather forecast data. Subsequently, a random forest algorithm is employed to select the core features of the sensors. Finally, health status assessment models are separately trained for the two types of sensors. Experimental results demonstrates that the proposed method has achieved accuracy of 99.29% and 99.07% respectively in the health status assessment of temperature and light intensity sensor.