Energy level is the fundamental property of energetic compounds and is one of the core criterion for determining whether an energetic compound has application potential. Experimental determination of detonation parameters is the basic method to evaluate energy level. However， due to the large dosage required and the poor reproducibility of the result， it is impossible to directly provide the theoretical maximum energy， which fails to meet the needs of new compound development. Non-experimental methods， including detonation equations of state， semi-empirical formulas and machine learning models， can predict detonation parameters via basic physicochemical properties or even only molecular structures. These methods become effective approaches for designing and screening new energetic compounds. This study systematically reviews these non-experimental evaluation methods. It summarizes and analyzes their basic principles， applicable conditions， advantages and disadvantages. Equations of state are theoretically rigorous and reliable but need precise physicochemical properties as input. Semi-empirical formulas are the most convenient to apply but lack sufficient accuracy and still require physicochemical properties as input. Machine learning models can predict detonation parameters only by molecular structures but have poor generalization performance. Overall， all existing methods are established based on available experimental data. Their weak data foundation limits extrapolation capability for unknown compounds， failing to meet the requirements for the design and screening of novel energetic compounds. Herein， we propose physics-mechanism-constrained hybrid models and dimensional expansion of energy evaluation parameters， to overcome the experimental data dependence of existing methods and improve their prediction accuracy and generalizability. This study provides solutions to the shortcomings of existing prediction methods and offers reference for establishing more accurate and generalizable ones.