With the complexity of modern power system and the susceptibility to external weather influences, it brings challenges to build an accurate load model. This paper proposes a variational autoencoder (VAE) long short‐term memory (LSTM) load model based on the attention mechanism (Attention). First, the Prophet data decomposition method is used to decompose long sequences of load data at multiple time scales. Second, the correlation‐based feature selection with maximum information coefficient (CFS‐MIC) method is employed to select weather features based on their relevance, a subset of features with high correlation and low redundancy is chosen as model inputs. Finally, the Attention‐LSTM‐VAE model is constructed to capture the temporal variations laws of load. The dataset includes 2 years of load values and weather data collected in Caojiaping, Hunan Province, China. The experimental results show that the Attention‐LSTM‐VAE model has the lowest mean absolute error of 0.0374 and the highest R‐squared value of 0.9714, verifying the accuracy of the model. Therefore, the performance of the Attention‐LSTM‐VAE model is better than the general deep learning load models, which has important reference for the research of power load models. Comparisons with other deep learning methods, the experimental results show that the Attention‐LSTM‐VAE model has the lowest mean absolute error of 0.0374 and the highest R‐squared value of 0.9714. The Attention‐LSTM‐VAE has better robustness, stability, and accuracy in load modeling, which has an important reference for the research of power load models.

Wireless sensor network (WSN) is usually deployed in unattended monitoring environments, with limited energy and large-scale characteristics. Therefore, the performance of WSN is susceptible to the influence of external environment and its own energy. The existing routing protocols try to optimize energy efficiency and routing reliability from the perspective of the network itself. However, they do not take into account the environmental impact from the outside, which makes them unable to properly respond to harsh environment. Therefore, a new Environment Fusion Routing Protocol (EFRP) is proposed. Cluster heads are selected in a probabilistic manner according to the ratio of the energy of each node and environmental factors. Nodes with excellent environment have a higher probability of being selected as cluster heads. Besides, the EFRP optimal path is determined according to the comprehensive potential field superimposed by the distance field, residual energy field and environmental field. The best path can be weighed among delay, energy saving and routing invulnerability. Therefore, the balance of network energy is realized and the invulnerability of the network in harsh environment is improved. The experimental results show that EFRP can significantly improve the data transmission and invulnerability of the network in harsh environments.