Forecasting the speed of an autonomous dump truck on quarry routes using spatial - temporal graph convolutional networks

Introduction. Accurate speed prediction for autonomous mining trucks (AMTs) is paramount to addressing critical aspects of safety, efficiency and productivity in mining operations. The application of spatiotemporal graph convolution networks (ST-GCNs) represents a promising and innovative approach to modeling the behavior of autonomous mining trucks (AMTs) in quarrying operations. This advanced computing technology is well suited for solving dynamic and complex problems associated with the movement of ACS in spatial and temporal dimensions. Methodology. The teaching and assessment methodology consists of several key stages. First, the dataset is loaded and preprocessed: temporal features are extracted from timestamps and spatial features are normalized. The data set is then split into training and test sets to train and evaluate the model. An ST-GCN model architecture is defined, consisting of convolution layers followed by a fully connected layer. The model is trained using a mini - batch stochastic gradient descent with a mean squared error (MSE) loss function and an Adam optimizer. Computational experiments and their discussion. The model was trained. Graphs of actual and expected speeds were constructed. A comparison of actual and expected speeds depending on the number of epochs is shown. The model demonstrates reasonable accuracy in predicting speed, but further evaluation and validation on additional data sets will provide better insight into its generalization ability. Conclusion. The analysis made it possible to find out the factors influencing the speed prediction. The ST-GCN model captures the complex spatiotemporal relationships in ACN trajectory data very effectively. The combination of ST-GCNs and accurate speed prediction allows for safer and more efficient quarrying operations.