As the oil transport system’s primary power equipment, it is crucial to ensure the safe and reliable operation of oil pumps. Its monitored parameters and variables are often from various types of sensors, which can be fused to predict designated variable trends as a prerequisite for lifetime and performance prediction as well as fault diagnosis and prediction. In this paper, an operation variable prediction method is proposed by fusing multi-type data and using canonical correlation analysis (CCA) and multi-layer perceptron (MLP) for oil pumps. First, the multi-type original data of oil pumps needs to be collected, cleaned and labeled as available and analyzable data in advance. Then, the CCA is used to extract key variable features by analyzing correlations between multi-type data. Subsequently, the most relevant variables are input into the MLP to predict the given variable signal trends. Additionally, the predictive performance of the proposed prediction method is evaluated by two indicators with the root mean square error and root mean square percentage error. Finally, a case study of the oil pump with 13 on-site variables of four types is conducted to demonstrate the effectiveness of the proposed method.