116 / 2021-08-04 20:35:15

SOC prediction method of wireless sensor nodes batteries based on Attention-LSTM

Lithium batteries are usually used to power wireless sensor nodes (WSNs). However, due to the nonlinearity of the battery, the prediction of its state of charge (SOC) has become a challenging task. In order to further improve the prediction accuracy, a node battery SOC prediction method based on attention mechanism and long short-term memory (LSTM) recurrent neural network is proposed, which maps the easily observed voltage and current to the target SOC. Firstly, one layer LSTM is used to make full use of the timing characteristics of lithium battery data, and then the attention mechanism is used to highlight the input characteristics that play a key role in SOC prediction. Finally, the full connection layer is used to output the prediction results. In addition, by comparing the performance of the model under different parameters, the optimal model setting is determined, and the proposed model is compared with other models. The results show that the proposed Attention-LSTM model has the best prediction performance, and the mean absolute error (MAE), mean square error (MSE) and determination coefficient (R²) reach 1.0250%, 0.0226% and 0.9027 respectively, which meets the requirements of node battery SOC prediction. In addition, a model-based node dynamic sleep scheduling algorithm is proposed to prolong the lifetime of nodes.