Multi-objective analysis to control ozone exposure C. Carnevale, G. Finzi, E. Pisoni, M. Volta Dipartimento di Elettronica per l’Automazione Università degli Studi di Brescia, Italy DEA - Università degli Studi di Brescia Research aim To develop a secondary pollution control plan: • Multi-objective optimization: – Objective 1: Air Quality Index (AQI) – Objective 2: Internal Costs (C) – Objective 2: External Costs (ExC) • for a mesoscale domain – Milan CityDelta domain (Northern Italy) DEA - Università degli Studi di Brescia Problem formulation: objective 1 the Air Quality Indicator (AQI) min( AQI ) min D [ AOT 40 i , j (Nis, j (d ) (1 rsN ),Vi s, j (d ) (1 rsV ))] i , jd 1 Nis, j ( d ),Vi s, j ( d ) daily cell NOx and VOC emissions in the reference case for CORINAIR sector s; ( rsN , rsV )s 1,...,11 decision variable set: CORINAIR sector precursor emission reductions; DEA - Università degli Studi di Brescia Problem formulation : objective 2 the emission reduction cost (C) 11 min( C ) min ( N s (1 rsN ) csN ( rsN ) V s (1 rsV ) csV ( rsV )) s 1 csN ( rsN ), csV ( rsV ) unit costs related respectively to NOx and VOC emission reduction; ( rsN , rsV )s 1,...,11 decision variable set: CORINAIR sector precursor emission reductions; DEA - Università degli Studi di Brescia Study domain 300x300km2 5150 SONDRIO TRENTO 5100 (m) 3000 VARESE 2800 BERGAMO 2600 5050 BRESCIA NOVARA 2400 VERONA MILANO 2200 2000 5000 1800 TORINO 1600 PIACENZA 1400 ALESSANDRIA 1200 PARMA 1000 4950 MODENA 800 600 400 GENOVA Milan domain 200 4900 0 400 450 500 550 600 650 DEA - Università degli Studi di Brescia AQI model identification • Pollutant concentration are computed by 3D deterministic chemical transport multiphase modelling system – Time consuming • Identification of source-receptor models (Neural Networks), describing the nonlinear relation between decision variables (emission reduction) and air quality objective, processing the simulations of TCAM DEA - Università degli Studi di Brescia TCAM model • Gas phase chemical mechanisms: SAPRC90, SAPRC97, COCOH97, CBIV • 21 aerosol chemical species • 10 Size classes – Size varying during the simulation – Fixed-Moving approach • Processes involved: – Condensation/Evaporation – Nucleation – Aqueous Chemistry Shell Core DEA - Università degli Studi di Brescia TCAM simulations • base case simulation: – – – – – – 300 x 300 km2, 60 x 60 cells, cell resolution: 5x5 km2 11 vertical layers emission and meteorological fields: JRC (CityDelta Project) initial and boundary conditions: EMEP the run of such a simulation takes about 12 days of CPU time simulation period: 1999 april to september • alternative scenario simulations: – CLE: current legislation – MFR: most feasible reduction O3 precursor CLE % MFR % NOx -29.79 -44.50 VOC -38.16 -58.74 DEA - Università degli Studi di Brescia Source-receptor models (NN) • Elman NN architecture: – Nodes of input layer: 2 – Nodes of output layer: 1 – Nodes of hidden layer: 8 an-1 • One neural network for each group of 2x2 (10x10 km2) domain cells vn • Input data: daily NOx and VOC emissions FW IW + Delay [MxM] AF1 an [MxQ] 1 b [Mx1] OW [LxM] 1 f(vn) + AF2 g [Lx1] • Target data: cell AOT40 daily values computed by the GAMES system DEA - Università degli Studi di Brescia Source-receptor models (NN) • Identification and validation dataset: – 3 TCAM seasonal simulations • Base Case; • Current LEgislation; • Most Feasible Reduction. • Validation dataset (126 values): – Third week of each month. • Identification dataset (423 values): – Remaining patterns DEA - Università degli Studi di Brescia Source-receptor models (NN) NBIAS r=0.97 5150 0.25 SONDRIO TRENTO 5100 0.2 VARESE 0.15 BERGAMO 5050 0.1 BRESCIA NOVARA VERONA MILANO 0.05 0 5000 PIACENZA -0.05 ALESSANDRIA PARMA 4950 -0.1 MODENA -0.15 GENOVA 4900 400 450 500 550 600 650 DEA - Università degli Studi di Brescia Cost functions • Cost curves used are estimated on the basis of RAINS-IIASA database (http://www.iiasa.ac.at) • An emission reduction cost curve has been assessed for each CORINAIR sector. • Decision variables = emission reduction for sectors: – VOC: 2, 3, 4 ,5, 6, 7, 8, 9 – NOx: 2, 3, 4, 7, 8 DEA - Università degli Studi di Brescia Cost functions • Fitting the costs of the available technologies: – considering 2nd order polynomial functions – with the constraint of estimating a monotonically increasing and convex function. 2500 NOx, sector 3: unit cost (K€) 2000 y = 11419x2 - 182,13x + 380,88 1500 1000 500 0 0% 10% 20% 30% 40% DEA - Università degli Studi di Brescia Optimization problem solution • Weighted Sum Method min( AQI( ) (1 ) C ( )) • Constraints 1. Maximum Feasible Reductions RsN S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 0 0.39 0.33 0.80 0 0 0.28 0.25 0 0 0 0 0.68 0.60 0.32 0.33 0.27 0.47 0.67 0.06 0 0 RsV 2. Technologies reducing both precursors DEA - Università degli Studi di Brescia Results • Pareto boundaries 100% 2,9E+07 90% AOT40 reduction (% max) AOT reduction (ppm) Utopia 3,1E+07 2,7E+07 2,5E+07 2,3E+07 2,1E+07 1,9E+07 1,7E+07 1,5E+07 0,E+00 80% 70% 60% 50% 40% 30% 20% 10% 0% 8,E+04 2,E+05 2,E+05 Cost reduction (Keuro) 3,E+05 4,E+05 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cost reduction (% max) DEA - Università degli Studi di Brescia Results NOx reductions VOC reductions 100% S2 S4 S5 S6 S7 S8 S9 90% 80% 70% 60% 50% 40% NOx emission reduction (%) VOC emission reduction (%) 100% 30% 20% 10% 90% 80% 70% 60% 50% S2 S3 S4 S7 S8 40% 30% 20% 10% 0% 0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% AOT reduction (% max) AOT reduction (% max) DEA - Università degli Studi di Brescia Results NOx emissions VOC emissions 180000 2 4 5 6 7 8 9 250000 200000 150000 100000 50000 0 NOX emissions (ton/year) VOC emissions (ton/year) 300000 160000 140000 2 3 4 7 8 120000 100000 80000 60000 40000 20000 0 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100 % 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100 % AOT40 reduction (% max) AOT40 reduction (% max) DEA - Università degli Studi di Brescia Conclusions – A procedure to formulate a multi-objective analysis to control ozone exposure has been presented – The procedure implements Elman neural networks tuned by the outputs of a deterministic 3D modelling system – The methodology has been applied over Milan CityDelta domain (Northern Italy): a strong reduction of ozone exposure (60% of the maximum air quality improvement) can be attained with a small fraction of the emission reduction technology costs (about 12%) DEA - Università degli Studi di Brescia Current activities – Uncertainty analysis: • • • Source-receptor models Cost curves VOC/NOx reduction functions for transport sectors – CityDeltaIII simulations to extend source-receptor model calibration and validation sets; – source-receptor models for SOMO35, AOT60, max8h, mean PM10 and PM2.5 concentrations; – PM10 and PM2.5 precursor (NOx, VOC, primary PM10, NH3, SO2) cost curves; – PM10 and PM2.5 two-objective optimization DEA - Università degli Studi di Brescia Thanks to… • This research has been partially supported by MIUR (Italian Ministry of University and Research). • The authors are grateful to the CityDelta community. • The work has been developed in the frame of NoE ACCENT. DEA - Università degli Studi di Brescia References • Finzi, G., Guariso, G., 1992. Optimal air pollution control strategies: a case study. Ecological Modelling 64, 221–239. • Barazzetta, S., Corani, G., Guariso, G., 2002. A neural emission-receptor model for ozone reduction planning. In: Proc. iEMSs 2002. • Volta, M. 2003. Neuro-fuzzy models for air quality planing. The case study of ozone in Northern Italy. European Control Conference. • Guariso, G., Pirovano, G., Volta, M., 2004. Multi-objective analysis of ground level ozone concentration control. Journal of Environmental Management 71, 25–33. • Carnevale C., Finzi G., Pisoni E., Volta M., 2006. Identification of source-receptor models for secondary tropospheric pollution control. 14th IFAC Symposium on System Identification. 2931 march, 2006 (pp. 762-767). IFAC Ed., CDROM published by Causal Productions. • M Carnevale C., Finzi G., Pisoni E., Volta M., 2006. Multi-objective analysis to control ozone exposure, 28th ITM-NATO. DEA - Università degli Studi di Brescia Constraints • Maximum feasible reductions allowed by technologies for macrosector s: 0 rsN RsN 0 rsV RsV • Technologies reducing both NOx and VOC emissions DEA - Università degli Studi di Brescia Optimization problem solution Constraints (2): technologies reducing both precursors macrosector 8 VOC reduction VOC reduction macrosector 7 NOx reduction NOx reduction] DEA - Università degli Studi di Brescia scenario A Utopia 100% AOT40 reduction (% max) 90% 80% 70% 60% A 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cost reduction (% max) DEA - Università degli Studi di Brescia basecase emission scenario NOx emissions VOC emissions s10 0,0% s9 0,2% s8 9,0% s10 0,0% s1 s11 1,6% 0,7% s2 1,8% s3 0,0% s11 0,4% s9 1,7% s4 5,0% s5 5,5% s1 20,8% s8 14,8% s2 4,0% s6 29,6% s3 7,4% s7 35,2% s4 15,8% s7 46,5% s5 0,0% s6 0,0% DEA - Università degli Studi di Brescia AOT40 scenarios source-receptor model simulations basecase Scenario A 30 30 ppb*h 65000 5150 60000 25 5150 25 55000 SONDRIO TRENTO 5100 50000 SONDRIO TRENTO 5100 20 20 45000 VARESE VARESE BERGAMO BERGAMO 5050 40000 BRESCIA NOVARA 15 VERONA MILANO 35000 5050 BRESCIA NOVARA 15 VERONA MILANO 30000 5000 5000 PIACENZA 10 25000 MODENA 5 GENOVA 4950 10000 5 10 500 MODENA GENOVA 4900 450 PARMA 15000 5000 5 ALESSANDRIA 20000 PARMA 4950 400 PIACENZA 10 ALESSANDRIA 15 550 20 600 25 650 30 0 4900 400 5 450 10 500 15 550 20 600 25 DEA - Università degli Studi di Brescia 650 30 scenario A: emissions s2 s3 s4 s5 s6 s7 s8 s9 0% -10% 2 -20% -30% -40% 3 2 NOx -50% -60% control priorities 1 1 VOC s2 s3 s4 s5 s6 s7 s8 s9 0 3 -20000 -40000 emission reductions (ton/year) -60000 2 2 1 -80000 -100000 NOx -120000 -140000 1 VOC DEA - Università degli Studi di Brescia