Efficiency prediction of excavator slope brushing operation based on pilot pressure and the main pump pressure

Aiming at the difficulty of obtaining efficiency of excavator brush slope condition, an excavator efficiency prediction method based on pilot pressure and main pump pressure was proposed. Firstly, the method took the actual brushing slope operation of excavator as the research object, and completed the collection of original data such as flow and pressure and hydraulic cylinder displacement based on LabVIEW platform, and the original data was processed by eigenvalue extraction based on time window and efficiency acquisition. Then, a principal component analysis–random forest model was established, including using principal component analysis theory to reduce the dimension of input data, improving the operation speed and anti-overfitting ability of the model, and using random forest algorithm to establish an excavator efficiency prediction model to achieve accurate prediction of excavator efficiency. Finally, the random forest prediction model was compared with support vector machine prediction model and BP neural network prediction model, and the influence of sample size and time window on model prediction accuracy, as well as the importance of each principal component in predicting excavator efficiency were explored. The results indicate that principal component analysis can effectively reduce the dimensionality of input data from 18 dimensions to 6 dimensions. Among the selected 6 principal components, the descending motion of the excavator bucket and the degree of load fluctuation of the excavator have a significant impact on the efficiency prediction results of the excavator. Compared with support vector machine prediction model and BP neural network prediction model, the random forest prediction model has higher accuracy. When the sample size is 0–50 000 and the time window is 0.05–0.25 s, the accuracy of the prediction model based on principal component analysis random forest algorithm increases with the increase of sample size and time window width. The optimal performance is achieved when the sample size is 40 000 and the time window width is 0.1 s, with a root mean square error of 0.029 2. © 2024 Central South University of Technology. All rights reserved.