Abstract
The pass rate of granules is an essential indicator during the high-pressure grinding process, as it accurately reflects the processing quality. Currently, the pass rate of granules is detected primarily based on manual experience judgments or offline inspections. Hence, this article presents a methodology for predicting the pass rate of granularity via an optimized support vector regression approach improved through genetic algorithms. Initially, a time-delay analysis method based on the particle swarm optimization algorithm is applied to mitigate the effects of time delays between the granularity pass rate and other data, thus aligning the dataset on a temporal scale. Subsequently, the feature data were selected using the maximum information coefficient analysis technique, which identified the most significant variables for inclusion in the training and testing sets of the predictive model. Predictions are then made using a support vector machine model that has been enhanced via genetic algorithm optimization. Furthermore, an online prediction model has been established, enabling real-time forecasting of the granularity pass rate and online model updates through root mean square propagation gradient descent optimization algorithm. This method leverages end-edge-cloud collaboration to provide a smart detection mechanism for the throughput rate of particles in high-pressure grinding mills. Experimental results demonstrate that, compared to traditional time-delay analysis, the improved time-delay analysis method proposed in this study is more effective and accurate. Simultaneously, the ɛ-GASVR granularity pass-rate prediction model proposed in this article achieved an R (2) of 0.89.