An industrial integrated energy system gathers a variety of production capacity and energy-related units, which also involves the production, transformation, and consumption of cold, heat, power, gas, etc. The combined cooling, heating, and power (CCHP) system built upon the cascade utilization of energy plays a significant role in improving the utilization rate of multiple energy resources. In this article, a controlled hierarchical relevance vector machine is modeled for identifying the scheduling event in the by-product gas system of a steel industry, which can effectively overcome the disturbance triggered by the changing of operational modes (OMs). Considering the multiple time scales (MTSs) property and the changing of seasons and load conditions of the CCHP, a unified modeling approach is proposed to describe MTSs and multiple OMs of the CCHP by using fuzzy singular perturbation here. Furthermore, an extended $Q$-learning approach by incorporating a hierarchical interval type-2 fuzzy system is designed for dynamically choosing the optimal controller’s parameter for the proposed unified model along with the changing of OMs. The performance of the proposed method can be experimentally validated by a series of industrial application cases to highlight the necessity of modeling and self-adaptive parameter of the controller.