Abstract: A wheat harvesting robot is often required for the high stability and accuracy of the load monitoring during operations, particularly under the complex and variable working conditions. In this study, the load status was quantified using the throughput indicator. An online load monitoring was also proposed to fuse the multi-source power flow sensing and intelligent algorithms. Firstly, the power sensors were developed for the feeding auger, conveyor, and threshing drum, with emphasis on the energy transfer paths of the key harvester. A highly synchronous multi-channel load monitoring terminal was developed to combine with the engine CAN bus data using a virtual instrument. The parameters of the power flow were synchronously acquired in real time, including the feeding auger power, conveyor power, threshing drum power, instantaneous engine fuel consumption, and engine torque percentage. Secondly, the field trials were conducted to cover the various wheat varieties, yields, and moisture contents. A high-fidelity correlation dataset between power flow parameters and throughput was constructed after acquisition. A moving average algorithm was used to smooth the raw data, effectively suppressing the random noise and impulse interference during field operations. In the core modeling, a hybrid prediction model was proposed to integrate multiple sensor information using machine learning. The more sufficient stability was achieved than in the single-signal prediction. Support vector regression (SVR) was used as the predictor. The multi-source heterogeneous sensor signals were integrated to capture the variation in the throughput. An advanced nutcracker optimization algorithm (NOA) was introduced to globally optimize the hyperparameters, such as the penalty coefficient and the kernel function parameters in the SVR. Finally, the prediction accuracy and generalization were significantly enhanced due to the high computational efficiency. Experimental results demonstrate that the best performance was achieved after optimization. Sensor calibration tests show that the reference error of all power sensors was below 0.5%, thereby fully meeting the requirements of the engineering accuracy. Univariate linear regression showed that there was the highest correlation between the threshing drum power and the throughput, with a coefficient of determination (R²) of 0.86, followed by the engine torque percentage, instantaneous fuel consumption, conveyor power, and feeding auger power. The SVR-NOA hybrid model performed best on the independent test set, with an R² as high as 0.96 and mean squared error (MSE) and mean absolute error (MAE) as low as 0.23 and 0.39, respectively. Horizontal comparative analysis showed that the absolute value of R² was improved by 0.06 (from 0.90 to 0.96), compared with the unoptimized baseline SVR model; Compared with the optimal univariate linear model using threshing drum power (R²=0.86), the absolute value of R² was significantly improved by 0.10, while the MSE and MAE were reduced by 66.7% and 40.9%, respectively, fully demonstrating the significant advantages of the multi-source information fusion model. The SVR-NOA model also performed best after comparison, which significantly outperformed the SVR-GWO model (^R2 leading by 0.05), and the second-best SVR-PSO model (R²=0.02). The final online verification showed that the monitoring system shared the maximum absolute error of 0.37 kg/s and an average absolute error of 0.15 kg/s for the real-time estimation of the throughput. The relative average error remained stable within 2.5%, indicating the excellent real-time monitoring accuracy and robustness. The high-precision online perception was achieved in the operating load of the combine harvesters using a precise sensing-algorithm fusion-intelligent optimization. The finding can provide direct technical support for the intelligent speed regulation, efficiency optimization, and fault early warning of the harvesters. It is also important theoretical value and practical significance to promote the grain harvesting machinery in precision agriculture.