Hongguang Pan

Personnel positioning is critical for coal mine safety. To address UWB ranging fluctuations from multipath effects and PDR drift under varying gaits, an adaptive fusion method is proposed. A GWO-GRU model is developed for gait recognition and is combined with an adaptively improved Weinberg model to optimize step length estimation. Furthermore, a CNN-BiLSTM network identifies line-of-sight/non-line-of-sight (LOS/NLOS) states from channel impulse response (CIR) signals with 89.60% accuracy, enabling dynamic weight adjustment. Experimental results show the method effectively suppresses cumulative drift, reducing the average error to 0.39 m, achieving a 34% improvement over uncorrected PDR. This approach ensures stable and continuous positioning in complex mines.

In recent years, the steady increase in university graduates has intensified competition in the job market. Traditional career platforms often display static job listings without personalized guidance. To address this gap, we propose an intelligent employment recommendation system that integrates artificial intelligence (AI) and Internet of Things (IoT) data. We first extract graduate profiles using a hybrid conditional random field (CRF) and long short-term memory (LSTM) neural network. Next, we apply an optimized K-means clustering algorithm with weighted similarity to group similar profiles. We then enhance a user-based collaborative filtering model with enterprise popularity scores and influencer impact factors to compute job-seeking similarity. Experiments on data from a Xi’an university demonstrate that our method outperforms standard user-based and cluster-based approaches, improving recommendation accuracy by 17.8% and 15.9%, respectively. This system delivers context-aware job suggestions that increase the success rate of graduate employment.

Recent studies have demonstrated the potential application of speech imagery neural signals in brain–computer interface (BCI) technology. Text generation based on speech imagery offers a natural communication method for individuals with speech disabilities. However, the limitations in imagined content and the immaturity of text generation technology currently constitute an obstacle to its applications. Therefore, this study proposes an enhanced CTA-BiLSTM model for efficient text generation utilizing speech imagery electroencephalography (EEG) signals, significantly enhancing the accuracy and fluency of text generation. Firstly, distinct from the prevailing imagination of characters and words, this study has assembled a sentence-level EEG dataset from ten subjects to facilitate communication. Subsequently, addressing the temporal dynamics characteristics and sequence dependencies of sentence signals, we employ dynamic time warping (DTW) and hidden Markov models (HMM) for accurate temporal alignment and signal annotation to generate fine-grained sentence labels. Finally, the proposed CTA-BiLSTM model leverages channel-time attention mechanism to dynamically adjust weights across channels and time, emphasizing critical features. Concurrently, the bidirectional long short-term memory (BiLSTM) network captures and utilizes long-term dependencies in the EEG signals, thereby enhancing the accuracy of the model in decoding complex temporal patterns. The experimental results demonstrate that the average sentence decoding accuracy can reach 67.50% on the self-built dataset, realizing a better evaluation accuracy and validating its potential for application.

It has been proven that electroencephalography (EEG) is an effective method for evaluating an individual's mental state. However, when it comes to the evaluation of miners' mental state, there are still some issues with missing EEG dataset and unsatisfactory evaluation accuracy. Therefore, this article proposes a miner mental state evaluation scheme with decision-level fusion based on multidomain EEG information. First, in the comprehensive lab for coal-related programs of Xi'an University of Science and Technology, the coal mine environment is simulated, and a realistic EEG dataset is constructed. Second, the multidomain features are extracted to represent abundant information in time, frequency, time-frequency, and space domain. These features with low dimension are classified adopting support vector machine (SVM), k-nearest neighbor (kNN), and back propagation (BP) network to obtain the optimal evaluation submodel (four domains corresponding to four submodels). Finally, based on the state probabilities provided by the optimal evaluation submodel, we adopt stack fusion and an improved Yager rule to fuse four submodels in order to find the most suitable fusion algorithm. The experimental results demonstrate that the average accuracy can reach 93.19% on the self-built dataset when utilizing the improved Yager rule with weight, and it realizes a better evaluation accuracy.