The power state evaluation plays a decisive influence on the safety implication of the lithium battery packs, and there is no effective online evaluation method so far due to the imbalance phenomenon among the internal connected battery cells, which cannot be abstained by the advancement of the materials and techniques. A novel power state mathematical evaluation method is proposed in this paper by investigating the improved external parameter coupling treatment, in which the mutual relationship description is conducted by the parameter information feature decomposition together with the Bayesian sequential decision algorithm. The complicated power state evaluation model with the coupling relationship decomposition is constructed by investigating the non-convex optimization treatment under complex working conditions for the lithium battery packs. The evidence combination is realized by introducing the information fusion strategy, according to which the multi criteria decision is realized by using the evidence theory. As can be seen from the experimental results, the voltage difference is within 10 mV in both of the first and the second phases, which increases rapidly in the third phase and reaches a maximum of 120 mV. Meanwhile, its power state evaluation accuracy is 95.00% and has a good output voltage tracking effect in the complex working conditions. The power state evaluation can be realized effectively by the proposed model constructing method, which is suitable for the complex battery cell combination structures and environmental influences, protecting the reliable and hierarchical working state monitoring and management of the lithium battery packs. It provides safety protection and energy management basis for the reliable power supply in the cleaner production of the power lithium battery packs.