Cluster Consistency Constrained Informer Model for Long-Term Battery Health Prediction in Energy Storage Stations

Abstract

 Numerous battery packs and clusters operating under different current, temperature, and ageing conditions are found in the energy storage systems. In these situations, the battery state-of-health's long-term deterioration and pack-to-pack variability must be taken into account for estimate. This research proposes a cluster-aware long-sequence Informer model for battery SOH estimation in energy storage facilities. The model includes degradation-stress sparse attention and consistency-constrained estimate after reconstructing degradation sequences from voltage, current, temperature, capacity, internal resistance, charge-discharge depth, and operational stress variables. The suggested model obtained 0.82% MAE, 1.17% RMSE, and 0.974 R² in experiments based on station-level battery operating data; concurrently, the cross-cluster SOH dispersion error was lowered by 23.6% in comparison to the conventional Informer. The findings demonstrate how long-sequence attention may increase SOH estimation accuracy under challenging energy storage operating settings and solve the issue of slow-degradation trends.