关键词: 开普勒优化算法(KOA)-双向长短期记忆网络(BiLSTM)模型, 淋水井筒, 风温预测模型, 可解释性分析, 皮尔逊相关性

Abstract:

This study aims to improve the accuracy, stability and interpretability of the model for the prediction of the air temperature in the mine water-drenched shaft. Firstly, characteristic variables were analyzed by Pearson correlation coefficient. Secondly, BiLSTM model was optimized by KOA, and the prediction model of mine shaft air temperature based on KOA-BiLSTM was established. Then, under the same sample conditions, the algorithm was compared with back propagation (BP), random forest (RF), least squares boosting (LSBoost) and support vector machine (SVM). Finally, interpretability analysis was conducted using the shapley additive explanations (SHAP) algorithm, which was verified by an example. The results show that the absolute error range of KOA-BiLSTM model is -1.24-0.5 ℃, which is 3.98% higher than the prediction accuracy of the unoptimized model. Compared with the other four models, the average absolute error (MAE), average absolute percentage error (MAPE) and mean square error (MSE) of the proposed model are the smallest, indicating that the model has the best prediction effect and generalization ability. The SHAP analysis shows that the wellhead air flow temperature has the greatest impact on the prediction results, while the surface pressure has the least impact. The absolute error range of KOA-BiLSTM model example verification is -0.49~0.38 ℃, and the prediction accuracy can meet the work needs.

Key words: Kepler optimization algorithm (KOA)-bidirectional long short-term memory (BiLSTM) model, water-drenched shaft, wind temperature prediction model, interpretability analysis, Pearson correlation