<p>The spatiotemporal evolution of water quality is fundamental for the management of water resources in long-distance water transfer projects (LWTPs). Due to the multidimensional and nonlinear characteristics of water quality monitoring data, a novel method is proposed to depict the spatiotemporal evolution of water quality in LWTPs. First, principal component analysis (PCA) is employed to identify the key indices influencing water quality in LWTPs. Subsequently, an improved machine learning method, IPSO-SVR, is introduced to characterize the spatiotemporal evolution of water quality. This method integrates particle swarm optimization (PSO) and support vector regression (SVR) with the commonly used K-fold cross-validation in machine learning. Finally, the effectiveness of this improved method is validated using water quality monitoring data from three automatic monitoring stations along the middle route of the South-to-North Water Diversion Project. The results are compared with actual values and simulated values derived from SVR, PCA-SVR, PSO-SVR, random forest (RF), XGBoost, and linear regression (LR). The findings indicate that IPSO-SVR outperforms the six other models, improving the coefficient of determination (<i>R</i>²) by 0.9% to 6.5% and reducing the mean square error (MSE) by 17% to 52%. Furthermore, its stability, reliability, and applicability are markedly superior to those of the other models. This research presents a novel approach and methodology to enhance the water quality management capabilities in medium- and large-scale water transfer projects.</p>