Spike-Driven Neuromorphic Sensing for Energy-Proportional Indoor Air Quality Monitoring in Multi-Zone IoT-Enabled...

Area of Science:

• Neuromorphic Engineering
• Environmental Sensing
• Smart Building Technology

Background:

• Indoor Air Quality (IAQ) monitoring in smart buildings faces challenges due to high computational and energy demands of continuous sensor processing.
• Event-driven methods offer a more efficient alternative by scaling computational cost with environmental changes, aligning with energy proportionality principles.

Purpose of the Study:

• To present a spike-driven neuromorphic sensing framework for decentralized IAQ monitoring in multi-zone smart buildings.
• To evaluate the energy efficiency and computational reduction of this framework compared to conventional methods.

Main Methods:

• Developed a framework combining adaptive Kalman filter preprocessing, dynamic threshold-based asynchronous spike encoding, and a Leaky Integrate-and-Fire neural network with Spike-Timing-Dependent Plasticity (STDP) learning.
• Collected multi-parameter IAQ data (PM1, PM2.5, PM10, CO2, CO, TVOCs, O3) from nine zones in an educational building.
• Validated the model against multiple baselines including LSTM, GRU, Temporal CNN, XGBoost, and Logistic Regression.

Main Results:

• Achieved a mean Sparse Firing Ratio of 10.94% and a Mean Response Time of 10.62 timesteps.
• Demonstrated an 8.9-fold computational reduction (approx. 89% fewer FLOPs) compared to LSTM inference.
• Confirmed stable, energy-proportional behavior across all zones with robust performance and repeatability.

Conclusions:

• Spike-based neuromorphic computation offers an energy-efficient and scalable solution for decentralized smart-building IAQ monitoring.
• The proposed framework significantly reduces computational requirements, paving the way for more sustainable smart building management.
• Further hardware-level validation is recommended for absolute power saving verification.