关键词: OSELM定位模型, 高动态井下环境, 正则化技术, 遗忘因子机制, 增量式学习, 位置指纹定位

Abstract: It is important to obtain the location of underground operators in a timely and accurate manner in the complex and changing underground coal mine environment, where the highly dynamic changes in the underground communication environment lead to a decrease in model localization accuracy. In this paper, the online sequential extreme learning machine (OSELM) algorithm is used for underground localization. Compared with the batch-type localization algorithms GA-BP and extreme learning machine (ELM), the OSELM algorithm can maintain the original model localization accuracy more effectively. However, the OSELM algorithm still suffers from the deficiencies of sickness matrix inversion and equal treatment of all added data, which leads to the poor stability and adaptability of the algorithm to dynamic environments. Based on the OSELM algorithm, the regularization OSELM algorithm and the forgetting factor OSELM algorithm are proposed respectively, while the OSELM algorithm that integrates the regularization technique and the forgetting factor mechanism is proposed. The experiments show that the localization accuracy of the OSELM algorithm is 0.566 and 0.628 2 m higher than that of the GA-BP and ELM algorithms, respectively, after the change of the experimental environment; the localization accuracy of the proposed OSELM algorithm with regularization and forgetting factor is higher than that of the OSELM algorithm in the 3 m error distance range; the localization accuracy of the OSELM algorithm that fuses the two mechanisms is the highest. The localization accuracy of the OSELM algorithm with the fusion of the two mechanisms is about 5% higher than that of the OSELM algorithm. Both the proposed OSELM and its improved algorithm can effectively improve the model localization accuracy.

Key words: OSELM positioning model, highly dynamic downhole environment, regularization techniques, forgetting factor mechanism, incremental learning, location fingerprinting