Although integrated energy systems (IES) are currently modest in size

their scheduling faces strong challenges

stemming from both wind generation disturbances and the system's complexity

including intrinsic heterogeneity and pronounced non-linearity. For this reason

a two-stage algorithm called the Multi-Objective Group Search Optimizer with Pre-Exploration (MOGSOPE) is proposed to efficiently achieve the optimal solution under wind generation disturbances. The optimizer has an embedded trainable surrogate model

Deep Neural Networks (DNNs)

to explore the common features of the multi-scenario search space in advance

guiding the population toward a more efficient search in each scenario. Furthermore

a multi-scenario Multi-Attribute Decision Making (MADM) approach is proposed to make the final decision from all alternatives in different wind scenarios. It reflects not only the decision-maker's (DM) interests in other indicators of IES but also their risk preference for wind generation disturbances. A case study conducted in Barry Island shows the superior convergence and diversity of MOGSOPE in comparison to other optimization algorithms. With respect to numerical performance metrics HV

IGD

and SI

the proposed optimizer exhibits improvements of 3.1036%

4.8740%

and 4.2443% over MOGSO

and 4.2435%

6.2479%

and 52.9230% over NSGAII

respectively. What's more

the effectiveness of the multi-scenario MADM in making final decisions under uncertainty is demonstrated

particularly in optimal scheduling of IES under wind generation disturbances.