Environmental Quality Monitoring Using Kalman Filter Algorithm in Urban Air Pollution Management
Annotatsiya
Urban air quality monitoring networks built on low-cost IoT sensor arrays offer unprecedented spatial density and real-time coverage relative to regulatory reference instruments, but at a fundamental signal quality cost: low-cost electrochemical sensors for nitrogen oxides (NOx) and carbon monoxide (CO), optical particle counters for PM2.5 and PM10, and non-dispersive infrared (NDIR) sensors for CO2 exhibit measurement noise levels 8-18 times higher than reference-grade instruments, cross-sensitivity interference between pollutant species, and progressive calibration drift that accumulates at rates of 2.4-4.8% per month in urban field deployments. The Kalman Filter (KF), introduced by Kalman (1960) as the minimum-variance linear recursive state estimator for systems with Gaussian process and measurement noise, provides the theoretically optimal framework for fusing these noisy sensor observations with a physics-based atmospheric dispersion prediction model to produce real-time PM2.5, PM10, CO2, NOx, O3, and SO2 concentration estimates whose uncertainty is fully characterised at each timestep by the state error covariance matrix P. This paper deploys a six-dimensional linear Kalman Filter across 15 IoT monitoring stations in three cities (London, United Kingdom: 5 stations; Delhi, India: 5 stations; Beijing, China: 5 stations) over 36 months (January 2022-December 2024), processing 394,560 hourly sensor observations (15 stations × 26,304 hours). The KF achieves PM2.5 RMSE of$2.84 \mu g/m^{3}$on the held-out evaluation set (final 6 months per station), outperforming Moving Average$\left({8. 4 8} {\mu} {g} / {m}^{{3}}\right)$, ARIMA$\left({5. 8 4} {\mu} {g} / {m}^{{3}}\right)$, and Extended Kalman Filter$({3. 4 8} {\mu} {g} / {m}^{{3}})$, with an${R}^{{2}}$of 0.924-a 42.6% improvement over the Moving Average baseline (0.648
Hali tarjima qilinmagan