The strength of high-performance concrete has an intricate relationship with its components. It is evident that traditional black-box models are characterised by an absence of interpretability. As a result, they fail to reveal the actual dependency between concrete strength and its components. Using a large sample dataset of high-performance concrete, machine learning models are constructed and optimized with the Hyperopt algorithm. The models are integrated with SHAP explainability techniques and feature dependency algorithms. A visual representation of the nonlinear relationship between the strength of high-performance concrete and its components is calculated. Furthermore, the interaction and coupling effects between multiple components are examined. The results indicate the following: Hyperopt-extreme gradient boosting model is a trusted machine learning model with high accuracy and robustness. Among the eight components(factors), age, cement, water and slag play a key controlling role in compressive strength, and their relationships with compressive strength satisfy the exponential function, power function, Gaussian function, and exponential function, respectively. At the same time, there is a significant interactive coupling effect between the four key factors; the interaction curves between age and cement, water, and slag adhere to the LogisticCum, LogNormal2D, and Power2D functions. The interaction between cement and water, as well as between cement and slag, conforms to the Poly2D function, while the interaction between slag and water conforms to the ExtremeCum function.