Informer neural network-based prediction of degradation trajectory of energy meter clock battery life

Accurate prediction of the life degradation trajectory of the clock battery, a critical component of the power meter, is essential for assessing the reliability of the device. Addressing the computational inefficiency and limited generalization capabilitie s of traditional empirical recession models, this paper introduces a prediction model based on the Informer neural network. The battery degradation data are initially segmented and reassembled using a sliding window strategy. The encoder-decoder structure is then employed to achieve multilevel feature extraction from the input sequences. By setting various prediction starting points, the model demonstrates high sensitivity in predicting battery degradation under temperature conditions identical to those in the training set. Additionally, the model is capable of predicting the degradation trajectories of batteries under different temperature conditions, such as 60℃.