Data-driven hydraulic pressure prediction for typical excavators using a new deep learning SCSSA-LSTM method

The hydraulic system pressure sensor is a crucial component for the excavator's automatic control. However, the working environment of an excavator is harsh and complex. If the pressure sensor malfunctions, the control system will lose its pressure signal, resulting in a complete loss of control over the excavator. To address this issue, a high-precision pressure data prediction model is proposed that combines an intelligent optimization algorithm with the long short term memory (LSTM) network. First, the basic principles of the excavator's hydraulic system and the common working conditions is introduced. Next, auxiliary variables are selected with the maximum information coefficient method, obtaining an optimal subset to serve as input for the prediction model. The sparrow search algorithm (SSA) is enhanced at three levels to determine the optimal parameters for the LSTM model, effectively addressing the challenges associated with hyperparameter selection. Finally, the effectiveness of the new deep learning method is validated. Experimental results demonstrate that the proposed data prediction method exhibits higher accuracy and stability compared to five other models. Specifically, the prediction accuracy improves by 49.29% compared to the traditional LSTM model under single pressure signal, and by 33.27% under multiple pressure signals.