Video action segmentation is a key technology for understanding videos. Due to the limitations of existing video action segmentation methods, this paper proposed a video action segmentation network based on an improved ASFormer with multi-scale temporal features. Aiming at the problem of local temporal feature disappearance caused by the multilayer structure of ASFormer network, we first introduce the multi-scale temporal feature extraction module and the multi-scale attention module to improve the network’s temporal feature extraction capability of actions. Second, to improve the network’s ability to localize action boundaries, boundary-aware loss is introduced to enhance the network’s attention at action boundaries. Then, a smoothing post-processing operation is introduced to suppress over-segmentation caused by mutated frames. To verify the effectiveness of the method proposed in this paper, an experimental study is finally performed on a publicly available dataset.

Cities need emergency-shelter systems that are computationally efficient, socially fair, and consistent with long-term goals for sustainable urban development. This paper proposes a GIS-integrated, two-stage location–allocation framework for urban earthquakes that jointly optimizes shelter siting and evacuee assignment under time-varying demand. The model incorporates equity constraints that cap extreme travel burdens for vulnerable groups and robust capacity safeguards against demand uncertainty, helping prevent over- or under-investment in shelter infrastructure and promoting efficient use of land and public resources. A customized Phased Nested Local Search (PNLS) heuristic enables city-scale application and is benchmarked against a mixed-integer programming baseline solved by CPLEX. In a district-level case study of Chengdu, China, the framework reduces total assignment distance by 12.3% and the 95th-percentile travel burden by 15.8% while maintaining feasibility during the peak demand window. The results show that integrating equity, robustness, and spatial efficiency in shelter planning can strengthen urban resilience and directly support SDG 11 on sustainable cities and communities.

The prediction, prevention and control of disasters have occurred throughout the history of human development, so the development of China's new era has put forward new requirements for China's disaster prevention policies and plans, especially for the formulation and implementation of policies by local governments. Local governments are the main administrators of a region and the main responders to disasters, so they have an unshirkable responsibility for the prediction and prevention of local disasters. In recent times, with the rapid development of digitalisation and informationisation, productivity and technology have been further improved, and disaster models can be predicted with the help of computers, mobile platforms and other devices. Compared with the traditional prediction and management models, digital management and prediction make simulation and calculation faster and more accurate. Therefore, this paper is based on the digital system model of the local government in the past 10 years for disaster policy as the object of study, the use of quantitative and qualitative combination of risk models and disaster prediction models, through data analysis and integration of the overall analysis of the performance characteristics of the disaster management policy and the relevant preventive measures, analysis and prediction of the tendency of the local government to needle disaster prevention policy. Through the digital analysis of statistics related to the frequency of disasters and the frequency of local government policymaking, analysis of the inevitable connection between the two, and the policy text to disaster prevention level of the hierarchy, from which to analyse the logic of disaster prevention and relief ideas and government experience, in order to better improve the prevention and relief of disasters related to the laws and regulations, and to provide relevant recommendations.

Accurate prediction of PM_2.5 concentrations is essential for public health protection and environmental management. However, existing forecasting models face critical challenges in capturing complex temporal dependencies and providing interpretability for decision-making. This study proposes a novel hybrid framework integrating one-dimensional convolutional neural networks (1D CNN), extended Long Short-Term Memory (xLSTM), and attention mechanism (CNN-xLSTM-AM), combined with SHapley Additive exPlanations (SHAP) for interpretable PM_2.5 forecasting. The 1D CNN unit captures local temporal relationships of multivariate sequences. The xLSTM design, based on sLSTM-mLSTM-sLSTM blocks, addresses limitations of traditional LSTM architectures through exponential gating and dual memory structures: sLSTM captures the state-tracking dynamics of scalar memory mixing, and mLSTM captures multivariate dependencies using matrix memory with larger capacity. The attention mechanism assigns dynamic weights to temporal features according to the importance of each time step. Extensive experiments on Beijing hourly data (2020–2021) show that CNN-xLSTM-AM achieves an RMSE of 9.845 μg/m³ and an R² of 0.977 for single-step prediction. Compared with CNN-xLSTM, it reduces RMSE by approximately 8.5% in single-step prediction and by 8.9% and 12.3% at the 2 h and 10 h forecasting horizons, respectively. The SHAP interpretability analysis identifies atmospheric pollutants as the highest contributors to predictive power (57.09%), with carbon monoxide as the top feature (32.791%). SHAP dependence analyses further suggest nonlinear associations between the input factors and model predictions, with potential turning points identified for selected variables, including carbon monoxide, relative humidity, wind speed, and wind direction. These model-derived patterns indicate potential changes in predictive relationships but should not be interpreted as definitive physical thresholds or universal regime shifts in PM2.5 behavior. These results demonstrate the potential of combining predictive modeling with post hoc interpretation to support environmental decision-making and may inform the development of targeted emission-control strategies and threshold-informed early-warning approaches.