Given the challenges of integrating research on soil and water ecosystem services into practical land use planning, this study targets the Loess Plateau as its area of study. The K-means clustering algorithm classifies land use structures at the small watershed scale. The InVEST model is then used to explore the impact of these land use structures on soil and water ecosystem services. Taking Yangquan City as a typical case study, multiple scenarios are set, and future land use simulations are conducted using a combination of multi-objective optimization algorithms and the PLUS model. The results indicate that, at the small watershed scale, the land use structures of the Loess Plateau can be categorized into six types: agricultural land, forest land, agro-pastoral, grassland, agro-silvo-pastoral, and bare land structures. The optimal land use proportions are 0.6—0.7, 0.5, 0.45/0.4 (agricultural land/grassland), 0.75—0.85, and 0.15/0.4/0.25—0.35 (agricultural land/forest land/grassland), respectively. The constraint curves between ecosystem services in each land use structure generally exhibit hump-shaped, concave wave, semi-concave wave, and convex wave patterns. In terms of future land use quantities, the areas of water bodies and shrubs are at relatively low levels under all three scenarios. Among them, the area of urban construction land is the largest in the control scenario, followed by the baseline scenario, and the smallest in the synergy scenario. In all three scenarios, Yangquan City is characterized by a concentric distribution pattern, with urban construction at the center, followed by agricultural land, grassland, and forest land. During the development process of land use, there is a substitution pattern among various land types, whereby urban construction tends to replace agricultural land and grassland, grassland may replace forest land and agricultural land, and forest land may be substituted by grassland. The research findings offer significant guidance for land use planning in the Loess Plateau.