Research on Lop Nur Potassium Salt Grade Prediction Model Based on Informer

The Lop Nur Salt Lake in Xinjiang harbors abundant brine resources and is the world’s largest sulfate-type potassium-bearing brine deposit. It serves as an essential potassium fertilizer supply base in China, boasting a high level of automation. In the process of producing potassium sulfate from sulfate-type brine, the grade of potassium salt significantly impacts the quality of the final product. Accurate and intuitive analysis and understanding of the potassium salt grade in the salt pools are crucial for resource utilization, automation processes, and product quality enhancement. Data were collected using unmanned water sampling machines equipped with GPS and sensors. By integrating chemical engineering principles and automated manufacturing process analysis, multiple variables influencing the potassium salt grade were selected as model inputs. Data preprocessing included outlier detection, missing value imputation, and noise reduction. An Informer-based prediction model for the potassium salt grade in Lop Nur was developed. The model was trained and tested, and the results were compared with those from LSTM and CNN models. The findings indicate that the Informer-based model exhibits higher prediction accuracy, stable error fluctuation, a narrower range of variation, and stronger generalization capability, outperforming the LSTM and CNN models.