ARIMA-BP neural network model is used to predict the carbon intensity of National Ecological Civilization Pilot Zone under the target of carbon peak, The non-radial directional distance function, mate-frontier theory and MNMCPI are used to discuss the potential of emission reduction. Research findings: (1) The MSE of the carbon intensity prediction series from the ARIMA-BP neural network model is 0.005 and the MAPE is 1.95%, showing a high accuracy. Compared with the carbon emission intensity in 2005, the carbon emission intensity of the region in 2028 decreased by 67.5%, Achieve ahead of the goal of reducing carbon intensity by more than 65% in 2030 compared with 2005, Among them, Guizhou has the largest decrease, with an 77.5% decrease, while Fujian, Jiangxi and Hainan have all seen decreases of 65% or more. (2) The trend chart of the emission reduction potential of the zone has generally declined in fluctuation, and the average annual emission reduction space from 2016 to 2030 can reach 26.59%, which is equivalent to the average annual emission reduction of 216.4933 million tons of CO2. By 2030, the emission reduction potential will reach 21.9%, the lowest value between 2016 and 2030. Since the emission reduction potential is equal to 1-carbon emission efficiency, which means that the carbon emissions in the zone have achieved a fundamental transformation. (3) By exploring the change of CO2 emission efficiency, Under the MNMCPI analysis, the average value of MNMCPI in the region is 1.008, with an average annual growth rate of 0.8%. The average value of changes in technical efficiency, best practice gap and technology gap ratio is 1.0031, 1.0103 and 1.0007, with an average annual growth rate of 0.31%, 1.03% and 0.07% respectively. The potential optimal emission reduction technologies in the region are constantly approaching the national optimal emission reduction technologies, which promotes the rapid decline of carbon intensity.