Video action recognition based on skeleton nodes is a highlighted issue in the computer vision field. In real application scenarios, the large number of skeleton nodes and behavior occlusion problems between individuals seriously affect recognition speed and accuracy. Therefore, we proposed a lightweight multi-stream feature cross-fusion (L-MSFCF) model to recognize abnormal behaviors such as fighting, vicious kicking, climbing over the wall, et al., which could obviously improve recognition speed based on lightweight skeleton node calculation, and improve recognition accuracy based on occluded skeleton node prediction analysis in order to effectively solve the behavior occlusion problem. The experiments show that our proposed All-MSFCF model has a video action recognition average accuracy rate of 92.7% for eight kinds of abnormal behavior recognition. Although our proposed lightweight L-MSFCF model has an 87.3% average accuracy rate, its average recognition speed is 62.7% higher than the full-skeleton recognition model, which is more suitable for solving real-time tracing problems. Moreover, our proposed Trajectory Prediction Tracking (TPT) model could real-time predict the moving positions based on the dynamically selected core skeleton node calculation, especially for the short-term prediction within 15 frames and 30 frames that have lower average loss errors.

The integration of a high proportion of wind power has brought disorderly impacts on the stability of the power system. Accurate wind power forecasting technology is the foundation for achieving wind power dispatchability. To improve the stability of the power system after the high proportion of wind power integration, this paper proposes a steady-state deduction method for the power system based on large-scale wind power cluster power forecasting. First, a wind power cluster reorganization method based on an improved DBSCAN algorithm is designed to fully use the spatial correlation of wind resources in small-scale wind power groups. Second, to extract the temporal evolution characteristics of wind power data, the traditional GRU network is improved based on the Huber loss function, and a wind power cluster power prediction model based on the improved GRU network is constructed to output ultra-short-term power prediction results for each wind sub-cluster. Finally, the wind power integration stability index is defined to evaluate the reliability of the prediction results and further realize the steady-state deduction of the power system after wind power integration. Experimental analysis is conducted on 18 wind power farms in a province of China, and the simulation results show that the RMSE of the proposed method is only 0.0869 and the probability of extreme error events is low, which has an important reference value for the stability evaluation of large-scale wind power cluster integration.

The ultrashort-term wind power prediction (USTWPP) technology assists the grid to arrange spare capacity, which is important to optimize power investment reasonably. To improve the accuracy of USTWPP and optimize power investment requirements, a USTWPP method with dynamic switching of multiple models is proposed. For high wind speed fluctuation samples, the wind speed-power curve (WSPC) is fitted in a large sample of historical data, and the corrected wind speed is the input of WSPC. The spatiotemporal attentive network model (STAN) is built for the prediction of low wind speed fluctuation samples. According to the real-time fluctuation characteristics of the correction wind speed, a switching mechanism between multiple models is established to reconstruct the prediction results along the time axis direction, and the predicted power is set to zero for the samples whose correction wind speed is lower than the cut-in wind speed. We conducted simulation experiments with data provided by a wind farm with an installed capacity of 130.5MW in China. The normalized root mean square error (NRMSE) for the 4h ahead predicted power reaches 0.0907, which verified the validity and applicability of the proposed model.

Amid the intensification of demographic aging, home robots based on intelligent technology have shown great application potential in assisting the daily life of the elderly. This paper proposes a multimodal human–robot interaction system that integrates EEG signal analysis and visual perception, aiming to realize the perception ability of home robots on the intentions and environment of the elderly. Firstly, a channel selection strategy is employed to identify the most discriminative electrode channels based on Motor Imagery (MI) EEG signals; then, the signal representation ability is improved by combining Filter Bank co-Spatial Patterns (FBCSP), wavelet packet decomposition and nonlinear features, and one-to-many Support Vector Regression (SVR) is used to achieve four-class classification. Secondly, the YOLO v8 model is applied for identifying objects within indoor scenes. Subsequently, object confidence and spatial distribution are extracted, and scene recognition is performed using a Machine Learning technique. Finally, the EEG classification results are combined with the scene recognition results to establish the scene-intention correspondence, so as to realize the recognition of the intention-driven task types of the elderly in different home scenes. Performance evaluation reveals that the proposed method attains a recognition accuracy of 83.4%, which indicates that this method has good classification accuracy and practical application value in multimodal perception and human–robot collaborative interaction, and provides technical support for the development of smarter and more personalized home assistance robots.

Data with time attributions such as price, load, and stock, which directly reflect the variation tendency, are the most common type of data character available. However, it is difficult to predict complex and volatile time-series character data. Further, most density cluster methods employ existing data to train the initial radius; however, a certain density radius is hard to be made suitable for continuously generated on-going datasets. Therefore, how to select a suitable timespan according to the time-series character in a way that makes it possible to support an adaptive updated density radius for real-time calculation is a core process. In this paper, a self-adaptive multi-density (SAMD) prediction model is proposed for solving the dynamic density radius selection problem in time-series data so as to improve the accuracy of real-time prediction. This multi-density clustering method can effectively shorten the iteration times and achieve dynamic clustering by the proposed jump sequence, which can optimize the jump points in the electricity price sequence. Moreover, we especially focus on the time interval features and other multi-source influencing factors together to construct the multi-core function with double-layer optimization to calculate the weighted coefficients, which have good adaptability and improve the classification and recognition performance. The experimental results show that the model had higher prediction accuracy and reduced processing time consumption in order to achieve real-time prediction.