To address the safety risks associated with the instantaneous transition of driver workload triggered by work zones on typical long straight sections of prairie highways, a multimodal data-driven workload recognition model was developed to enhance traffic safety monitoring capabilities in unconventional scenarios. Representative long straight sections of prairie highways were selected to conduct driving simulation experiments, featuring two specific work zone scenarios: inner lane closure and outer shoulder closure. Multimodal data, including eye movement (EM), electrocardiogram (ECG), and vehicle operation and control (VOC), were synchronously collected. Meanwhile, subjective workload scores were obtained as supervised labels using the NASA task load index (NASA-TLX). Core features were identified through correlation and factor analysis. Frequency-domain indicators unsuitable for short time windows were excluded via physiological validity verification, resulting in a refined feature system comprising 12 core indicators. Five machine learning models were parameterized and optimized using a Bayesian optimization algorithm, followed by a comparative analysis of recognition performance across different models and feature combinations. The results indicate that the extreme gradient boosting (XGBoost) model exhibited superior performance in prairie highway work zone scenarios, achieving a classification accuracy of 91.60% and an Area Under the receiver operating characteristic curve (ROC-AUC) of 0.948, which significantly outperformed random forest (RF) and support vector machine (SVM) models. Regarding feature combinations, the fusion of all dimensions (EM-ECG-VOC) yielded the highest recognition efficiency. Among dual-modal combinations, the ECG-VOC combination demonstrated better performance than those incorporating eye movement indicators. It is concluded that the fusion of multimodal features combined with optimized ensemble learning algorithms can effectively identify instantaneous fluctuations in driver workload within prairie highway work zones. This study provides theoretical evidence and algorithmic support for the development of active safety warning systems and the optimization of autonomous driving takeover strategies for high-risk road sections.