Human activity recognition using sensor data has become a research hotspot in the field of ubiquitous computing and has a wide range of application scenarios in real life. Effectively extracting and fusing the rich spatiotemporal information contained in the sensor data from various sensors and multiple wearable devices is a challenging task. Moreover, for practical deployment applications, the limited computational and storage resources of edge devices also impose requirements on the model size and inference speed. Hence, we propose STFNet, an enhanced and lightweight spatiotemporal fusion network for wearable human activity recognition. In which, we innovatively divide the extraction and fusion process of spatiotemporal features into three stages, including sensing feature extraction, temporal feature extraction, and channel fusion, to learn incrementally and reduce information loss. We first extract shallow intra-channel features and learn local inter-channel information by using a lightweight local attention mechanism. Then, the learned features are fed into a hierarchical convolution structure with residual connections to extract multi-scale and comprehensive temporal features. Finally, we use graph convolutions to further fuse the deep inter-channel information. By replacing the conventional global attention mechanism with a local one and using hierarchical convolution to replace dilated or multi-kernel convolution, STFNet achieves the purpose of lightweight design. In the experiments, STFNet performs better on public datasets than the selected WHAR models published in recent years and demonstrates its lightweight characteristics on edge device deployment.

Using wearable sensors to identify human activities has elicited significant interest within the discipline of ubiquitous computing for everyday facilitation. Recent research has employed hybrid models to better leverage the modal information of sensors and temporal information, enabling improved performance for wearable human activity recognition. Nevertheless, the lack of effective exploitation of human structural information and limited capacity for cross-channel fusion remains a major challenge. This study proposes a generic design, called GT-WHAR, to accommodate the varying application scenarios and datasets while performing effective feature extraction and fusion. Firstly, a novel and unified representation paradigm, namely Body-Sensing Graph Representation , has been proposed to represent body movement by a graph set, which incorporates structural information by considering the intrinsic connectivity of the skeletal structure. Secondly, the newly designed Body-Node Attention Graph Network employs graph neural networks to extract and fuse the cross-channel information within the graph set. Eventually, the graph network has been embedded in the proposed Bidirectional Temporal Learning Network , facilitating the extraction of temporal information in conjunction with the learned structural features. GT-WHAR outperformed the state-of-the-art methods in extensive experiments conducted on benchmark datasets, proving its validity and efficacy. Besides, we have demonstrated the generality of the framework through multiple research questions and provided an in-depth investigation of various influential factors.

In the field of intelligent transportation, the planning of traffic flows that meet energy-efficient driving requirements necessitates the acquisition of energy consumption data for each vehicle within the traffic flow. The current methods for calculating vehicle energy consumption generally rely on longitudinal dynamics models, which require comprehensive knowledge of all vehicle power system parameters. While this approach is feasible for individual vehicle models, it becomes impractical for a large number of vehicle types. This paper proposes a digital model for vehicle driving energy consumption using vehicle speed, acceleration, and battery state of charge (SOC) as inputs and energy consumption as output. The model is trained using an optimized CNN-BiLSTM-Attention (OCBA) network architecture. In comparison to other methods, the OCBA-trained model for predicting PHEV driving energy consumption is more accurate in simulating the time-dependency between SOC and instantaneous fuel and power consumption, as well as the power distribution relationship within PHEVs. This provides an excellent framework for the digital modeling of complex power systems with multiple power sources. The model requires only 54 vehicle tests for training, which is significantly fewer than over 2000 tests typically needed to obtain parameters for power system components. The model’s prediction error for fuel consumption under unknown conditions is reduced to 5%, outperforming the standard error benchmark of 10%. Furthermore, the model demonstrates high generalization capability with an R² value of 0.97 for unknown conditions.