Abstract: Cognition of complex systems faces the following three intertwined challenges: (1) mechanism analysis of deterministic model behavior, (2) handling of high-dimensional uncertainty scenarios, (3) improvement in model authenticity. The methodology (AI+WRT) that integrates artificial intelligence (AI) with whole reductionism thinking (WRT) provides a systematic cognitive framework to solve this dilemma. WRT constructs entropy-preserving bidirectional channels between the long-term mutually exclusive holism and reductionism, achieves whole reductionism of deterministic complex models. It provides the safety margin of the overall model, bifurcation conditions, and microscopic features on any time section, such as trajectory eigenvalues and energy conversion rate, and supports sensitivity analyses and optimal decision. On the other hand, after the ultra-high-dimensional uncertainty space grid-based, various AI technologies (including LLMs and expert systems (ES) sort the discrete deterministic cases by attention level according to forecasting and early warning information, and assign them to the lower-layer, WRT, for case-by-case quantitative analysis. The two-layer framework, AI+WRT, decouples the study of uncertainty and complexity into two relatively independent yet collaborative frameworks, achieving a closed cognitive loop. This methodology originates from the quantization algorithm for transient stability analyses of power systems (Extended equal area criterion EEAC) established in 1987. Subsequently, by expressed by functionals and operators, the above thinking is separated from embodied knowledge of application domain problems, generalized into the WRT methodology, and further expanded into the methodology for researching uncertain complex models, AI+WRT. The application areas have expanded from transient stability of power systems to smart grid (CPS in power), CPS in energy, cyber-physical-social-system (CPSS) in energy, and CPSS in energy-environment-economy (CPSS-EEE). After nearly forty years of engineering practice and refinement, a complete system has been formed, from epistemological foundations and operational frameworks to engineering validation. This paper defines it as a logical chain composed of several core points, including the design of a two-layer cognitive architecture of deterministic WRT and uncertainty AI+WRT. Through engineering practice in different fields, it verifies its applicability and transferability, summarizing it as a cognitive methodology for uncertain complex model behavior. Finally, it looks forward to enhancing human understanding of the evolution of objectively existing ontological systems starting from behavioral cognition based on complex subjective models. The article emphasizes that this will be a tough journey forever on the road.