Liquid time constant based neuromorphic active inference for resilient control and health aware battery management in hybrid AC/DC microgrids

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

The transition toward inverter-dominated renewable microgrids is frequently hindered by grid instability and prohibitive battery energy storage system replacement costs. Traditional control strategies often prioritize immediate grid regulation while neglecting electrochemical constraints, which accelerates battery health degradation. This paper proposes a neuromorphic active inference with liquid dynamics (NAI-LD) methodology, a bio-inspired architecture that interprets the microgrid as a dynamical system seeking operational homeostasis. Utilizing continuous-time liquid time-constant neural network, the controller adaptively modulates processing speed, contracting its temporal horizon during disturbances and expanding it during steady state to suppress noise. Evaluated in a simulation environment under severe stochastic transients, the NAI-LD controller restricted frequency excursions to ± 0.025 Hz and total harmonic distortion to a range of 2.3 to 2.7%. By incorporating expected free energy minimization, the framework establishes an intrinsic self-preservation mechanism to reduce daily micro-cycling. Based on the simulated trajectories, techno-economic analysis projects a potential 31.99% reduction in annualized operational expenditure and a theoretical 2.9-year extension of battery hardware lifespan. However, these findings are inherently limited by the simulation environment, the degradation projections are strictly bounded to the semi-empirical characteristics of lithium iron phosphate chemistry and do not currently account for physical communication latencies, sensor noise, or analog-to-digital conversion jitter. While the average algorithm execution time of 0.62 milliseconds indicates computational feasibility for standard digital signal processors, subsequent hardware-in-the-loop experimental validation remains necessary to bridge these modeling limitations and confirm the projected system performance in physical field deployments.

Links and sources

• PDF from original source

Comments