As a traditional performance evaluation method, DEA does not embody decision preference as well as construct a total ordering of effective evaluation units. Under the premise of considering the preference of decision makers, built is a DEA preference matrix through using subjective evaluation methods. Based on the principle of TOPSIS sequencing, built are positive ideal points and negative ideal points by creating the virtual best and the virtual worst frontier. According to these results, the DEA sequencing value of each evaluation unit is determined via the relative closeness degree. An empirical study is done with the data of 12 oil production plants and conducted is their performance evaluation at various priority conditions such as oil and gas production, safety production, and environmental protection production. Results show that the improved method has practical application value in oil production plant performance evaluation optimization.