To reduce the cost， development cycle， and safety risks associated with the traditional trial-and-error paradigm in energetic materials research， and to improve the design efficiency of high-energy oxygen- and nitrogen-rich azole energetic compounds， a precise design strategy based on interpretable machine learning was proposed. A dataset containing 150 samples was constructed from reported detonation velocity data of azole energetic compounds. Twenty-four structural and molecular descriptors were selected， and four detonation velocity prediction models were developed. The results show that the support vector machine （SVM） model exhibited the best predictive performance， with a test-set coefficient of determination of 0.9367， a root mean square error of 0.14 km·s^-1， and average and maximum relative errors of 1.6% and 3.4%， respectively. Model interpretability analysis indicates that oxygen balance， the mass fraction of nitrogen gas in detonation products， the number of nitro groups， and MolLogP are key descriptors affecting detonation velocity. Guided by these findings， a low-sensitivity azole-fused ring scaffold was selected， and 48 novel oxygen- and nitrogen-rich azole energetic compounds were designed. Predictions from the optimal model show that all designed molecules have detonation velocities higher than or comparable to that of RDX， among which 11 molecules， accounting for 22.9%， surpass HMX in detonation velocity， with predicted impact sensitivities comparable to or better than that of HMX. These results demonstrate that interpretable machine learning can be applied to the rapid screening and precise design of high-energy azole energetic compounds， providing an effective approach for the development of novel high-performance energetic materials.