Data, Controls, and Optimization:

Size: px

Start display at page:

Download "Data, Controls, and Optimization:"

Felix Eaton
5 years ago
Views:

1 Data, Controls, and Optimization: Studies in Building Energy Management Scott Moura Assistant Professor ecal Director Civil & Environmental Engineering University of California, Berkeley Center for the Built Environment Brown Bag Lunch Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 1

2 Contributors to this Talk... UCB Students UCB Post-docs Visiting Ph.D. Visiting Scholar Colleagues Eric Burger Xiaosong Hu Chao Sun [Beijing Inst. of Tech.] Xiaohua Wu [Xihua University] Fengchun Sun [Beijing Inst. Tech.], Jae Wan Park [UC Davis] Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 2

3 Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 3

4 Vision for Future Energy Infrastructure Energy Resources and Generation Transmission & Distribution Building Systems Exhaustible Resources Distributed Generation Electrified Transportation Energy Storage Renewable Resources Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 4

5 Cyber Physical Systems Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 5

6 Outline 1 Forecasting Building Electric Demand 2 Residential Buildings with Solar & Storage 3 Integrating PEV Energy Storage with Buildings 4 Open Thoughts & Shameless Advertisements Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 6

7 The Electricity Demand Forecasting Problem Needs: A generalized framework for forecasting electricity across a diversity of buildings, with only hourly consumption and meteorological data Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 7

8 The Electricity Demand Forecasting Problem Needs: A generalized framework for forecasting electricity across a diversity of buildings, with only hourly consumption and meteorological data Reality: Buildings are highly diverse in design, size, type, use, etc. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 7

9 The Electricity Demand Forecasting Problem Needs: A generalized framework for forecasting electricity across a diversity of buildings, with only hourly consumption and meteorological data Reality: Buildings are highly diverse in design, size, type, use, etc. Some Motivating Facts Emit Consume 48% of carbon emissions 39% of total energy 71% of electricity 54% of natural gas Netflix Prize 1M USD Award for Best Algorithm Predict Subscriber Movie Ratings (1 to 5 stars) Competing algorithms converged on similar concepts Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 7

10 The Electricity Demand Forecasting Problem Needs: A generalized framework for forecasting electricity across a diversity of buildings, with only hourly consumption and meteorological data Reality: Buildings are highly diverse in design, size, type, use, etc. Some Motivating Facts Emit Consume 48% of carbon emissions 39% of total energy 71% of electricity 54% of natural gas Netflix Prize 1M USD Award for Best Algorithm Predict Subscriber Movie Ratings (1 to 5 stars) Competing algorithms converged on similar concepts Punchline Apply lessons from open competition to building electricity forecasting Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 7

11 Electricity Demand Forecasting Power Demand (kw) Actual Demand 6 Hour Forecast 60 8/6/13 0h 8/6/13 12h 8/7/13 0h 8/7/13 12h 8/8/13 0h 8/8/13 12h Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 8

12 Stacked Ensemble Learning inspired by Netflix prize Suppose you have already constructed M forecasting models, with prediction output ŷ s R m, s = 1,, M. Consider a weighted sum of these models M ŷ Σ = θ s ŷ s with θ s R the weighting coefficient of sub-model s, for s = 1,, M. s=1 Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 9

13 Regression Employ Least Squares with L 2 regularization (a.k.a. Ridge regression for scikit-learn users) to learn weights θ s for s = 1,, M min θ ( N y i i=1 ) 2 M M θ s ŷ s,i + λ s=1 s=1 θ 2 s θ s R : weights for sub-model s, θ = [θ s ] s=1,,m y i R m : i th observed electricity demand ŷ s,i R m : i th predicted electricity demand for sub-model s i = 1,, N : where N is the number of data samples λ : Weight for regularization penalty Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 10

14 Regression Employ Least Squares with L 2 regularization (a.k.a. Ridge regression for scikit-learn users) to learn weights θ s for s = 1,, M min θ ( N y i i=1 ) 2 M M θ s ŷ s,i + λ s=1 s=1 θ 2 s θ s R : weights for sub-model s, θ = [θ s ] s=1,,m y i R m : i th observed electricity demand ŷ s,i R m : i th predicted electricity demand for sub-model s i = 1,, N : where N is the number of data samples λ : Weight for regularization penalty Consider fitting a model of Wurster on July 15 and September 15. Will the results be different? Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 10

15 Online/Real-time Regression Answer: YES! Building behavior evolves over time! So should the models! Compute time-varying weights using a retrospective moving horizon optimization min θ t ( T y t i i=1 ) 2 M M θ t,s ŷ s,t i + λ s=1 s=1 θ 2 t,s θ t,s R : weights at time-step t, for sub-model s, θ t = [θ t,s ] s=1,,m y t i R m : observed electricity demand at time step t i ŷ s,t i R m : predicted electricity demand for sub-model s, at time step t i i = 1,, T : where T is the length of the retrospective time horizon λ : Weight for regularization penalty Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 11

16 Electricity Demand Forecasting Train (offline): 8 buildings w/ 8 models each; data from 1/2013 to 6/2014 Test (online): Generate 6 hour forecasts; data from 7/2014 to 12/2014 MAPE = 100% 1 N N y i ŷ i,σ i=1 y i Birge Davis East Asian Library Hearst Mining Dwinelle Offices Sproul University Wheeler Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 12

17 Electricity Demand Forecasting Dwinelle Offices Actual Demand Ensemble Forecast, ŷ Σ,t Power Demand (kw) /5/13 12h 8/6/13 0h 8/6/13 12h 8/7/13 0h 8/7/13 12h Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 13

18 Electricity Demand Forecasting Time-Varying θ t,s Ridge Weights k-nn Weights θ 0.2 t,5 0.1 θ t,6 0.0 θ t,7 0.1 θ 0.2 t, /2013 8/2013 9/ / /2013 θ t,1 θ t,2 θ t,3 θ t,4 Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 14

19 Outline 1 Forecasting Building Electric Demand 2 Residential Buildings with Solar & Storage 3 Integrating PEV Energy Storage with Buildings 4 Open Thoughts & Shameless Advertisements Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 15

20 Solar+Storage - An Emergent Market Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 16

21 The Building Solar+Storage Problem Needs: Optimally manage energy flow between loads, solar, and storage. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 17

22 The Building Solar+Storage Problem Needs: Optimally manage energy flow between loads, solar, and storage. Reality: Current controls are mostly heuristic - no models, no data. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 17

23 The Building Solar+Storage Problem Needs: Optimally manage energy flow between loads, solar, and storage. Reality: Current controls are mostly heuristic - no models, no data. Some Motivating Facts Policy 50% renewables in CA by 2030, 100% in Hawaii by 2045 Climate 2011 Tsunami in Japan energy security and reliability Costs Li-ion battery pack costs decreasing toward 350 USD/kWh Data Over 50M (43%) of US homes have smart meters Hybrid Vehicles Photovoltaics/Grid Engine Home Demand Driver Power Demand Battery Storage Battery Storage Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 17

24 The Building Solar+Storage Problem Needs: Optimally manage energy flow between loads, solar, and storage. Reality: Current controls are mostly heuristic - no models, no data. Some Motivating Facts Policy 50% renewables in CA by 2030, 100% in Hawaii by 2045 Climate 2011 Tsunami in Japan energy security and reliability Costs Li-ion battery pack costs decreasing toward 350 USD/kWh Data Over 50M (43%) of US homes have smart meters Hybrid Vehicles Photovoltaics/Grid Engine Home Demand Driver Power Demand Battery Storage Battery Storage Punchline Apply & Extend 10 years of HEV Energy Management Control Research to Building Solar+Storage Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 17

25 Features of Published Energy Management Frameworks for Buildings Features [1] [2] [3] [4] [5] [6] [7] Our Paper Problem Formulation Lin Lin? Lin NL? Lin Lin thermal circuit Solver LP DP MILP GA MILP MILP LP? DP Battery Model Lin Lin (Int) Model Predictive Control Load Forecast Sensitivity on Load Forecast Error PV Generation Forecast Carbon Emission Reduction Gauss Noise Gauss Noise Lin Lin (Int) Lin (Int) Lin (Int) Lin (Int) Known Known Known Known Y (ANN) Known Y? Gauss Y (ANN) Known Battery Aging Y Y Y Blank = feature not used or not mentioned Y NL NL Y Y Y (API) Y WattTime Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 18

26 Residential Buildings with Solar & Storage Photovoltaic Arrays Cloud with information, algorithms... Load Demand Controller & Converters Utility Grid (a) Battery Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 19

27 Predictive Controller with Load/Weather Forecasting S T Hour of P day dem Day of Week PV Model P PV Load Forecaster P dem Predic0ve Controller P bab SOC Smart Home ngrid Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 20

28 Single-Family Home Energy Patterns in LA ElectricityiLoadiHkWY HourlyiLoad DailyiLoad MonthlyiLoad YeariAverage Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Month Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 21

29 Single-Family Home Energy Patterns in LA Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 21

30 Single-Family Home Energy Patterns in LA LoadLvs.LTemp FitLResult WeeklyLAverageLLoadL(kW) Noise WeeklyLAverageLTemperatureL( C) Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 21

31 Artificial Neural Network (ANN) X T a D w T d L h H(X,C,σ) w w w Future load demand. Y=P dem Input Layer Hidden Layer Output Layer Y = f ANN (X), X = [T a, D w, T d, L h ], Y = [P dem,k+1,, P dem,k+m ] Y = f ANN (X) = N a i H i ( X C i ) i=1 H i ( X C i ) = exp [ 1 ] 2σ X C 2 i 2 i Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 22

32 Short-term Forecast of Home Load 30 ba)-1 30 bb)-1 C AirATemperautreAonA AirATemperautreAonA LoadAbkW) ba)-2 RealALoad ForecastAL HourAofADayAonA Tuesday Hour of Day on Friday Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 23

33 Short-term Forecast of Home Load EmpiricaltCDF 1 0y8 0y6 0y4 0y2 haa RMSEtcdftoftLAtData RMSEtcdftoftBerkeleytData AveragetRMSE 0 0 0y2 0y4 0y6 0y8 1 1y2 RMSE 0y52 0y49 0y46 0y43 hba 0y InputtHistoricaltLoadtLength Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 23

34 Cloud-Enabled Control S T Hour of P day dem Day of Week PV Model P PV Load Forecaster P dem Predic0ve Controller P bab SOC Smart Home ngrid Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 24

35 Battery and Photovoltaic Cell Models battery d dt SOC(t) = I batt(t) Q, P batt (t) = V oc I batt (t) I 2 batt(t)r in, (a) OCV-R photovoltaics V d = V cell + I pv R s, [ ] I = I sc I s e ( qv d AkTpv(t) ) 1 ( ) 3 Tpv (t) I s = I s,r e T r qe bg Ak V d, R p ( 1 1 Tr Tpv(t) ), I sc = [I sc,r + K I (T pv (t) T r )] S pv(t) 1000, P pv (t) = n cell V cell (t)i(t) X. Hu, S. Li, and H. Peng, A comparative study of equivalent circuit models for Li-ion batteries, (b) Journal OCV-R-RC of Power Sources, vol. 198, pp , S I sc + V d (c) Impedance - R p R s I + V cell G. Vachtsevanos and K. Kalaitzakis, A hybrid photovoltaic simulator for utility interactive studies, IEEE Transactions on Energy Conversion, no. 2, pp , Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 25 -

36 Internet-based Data Feeds S T Hour of P day dem Day of Week PV Model P PV Load Forecaster P dem Predic0ve Controller P bab SOC Smart Home ngrid Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 26

37 Internet-based Data Feeds Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 26

38 Internet-based Data Feeds Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 26

39 Cloud-Enabled Control S T Hour of P day dem Day of Week PV Model P PV Load Forecaster P dem Predic0ve Controller P bab SOC Smart Home ngrid Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 27

40 Nonlinear MPC Formulation (k+hp) t min J k = [λ 1 ElecPrice(t)P grid (t) + λ 2 CO 2 (t)p grid (t)] 2 dt, k t Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 28

41 Nonlinear MPC Formulation (k+hp) t min J k = [λ 1 ElecPrice(t)P grid (t) + λ 2 CO 2 (t)p grid (t)] 2 dt, k t s. to SOC= I batt Q, 0=V oc I batt I 2 batt R in P batt, [Battery] 0 = h PV (P pv, S, T), [Photovoltaic] 0 = P grid + η dd η da P pv + η da P batt P dem, [Pwr Balance] Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 28

42 Nonlinear MPC Formulation (k+hp) t min J k = [λ 1 ElecPrice(t)P grid (t) + λ 2 CO 2 (t)p grid (t)] 2 dt, k t s. to SOC= I batt Q, 0=V oc I batt I 2 batt R in P batt, [Battery] 0 = h PV (P pv, S, T), [Photovoltaic] 0 = P grid + η dd η da P pv + η da P batt P dem, [Pwr Balance] SOC min SOC SOC max, I min batt I batt I max batt, P min batt P batt P max batt, P min grid P grid P max grid, Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 28

43 Nonlinear MPC Formulation (k+hp) t min J k = [λ 1 ElecPrice(t)P grid (t) + λ 2 CO 2 (t)p grid (t)] 2 dt, k t s. to SOC= I batt Q, 0=V oc I batt I 2 batt R in P batt, [Battery] 0 = h PV (P pv, S, T), [Photovoltaic] 0 = P grid + η dd η da P pv + η da P batt P dem, [Pwr Balance] SOC min SOC SOC max, I min batt I batt I max batt, P min batt P batt P max batt, P min grid P grid P max grid, ˆd ((k + n) t)=f forecast (d(k t),, d((k H h ) t)), n = 1,, H p Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 28

44 Nonlinear MPC Formulation (k+hp) t min J k = [λ 1 ElecPrice(t)P grid (t) + λ 2 CO 2 (t)p grid (t)] 2 dt, k t s. to SOC= I batt Q, 0=V oc I batt I 2 batt R in P batt, [Battery] 0 = h PV (P pv, S, T), [Photovoltaic] 0 = P grid + η dd η da P pv + η da P batt P dem, [Pwr Balance] SOC min SOC SOC max, I min batt I batt I max batt, P min batt P batt P max batt, P min grid P grid P max grid, ˆd ((k + n) t)=f forecast (d(k t),, d((k H h ) t)), n = 1,, H p state = SOC, control = P grid, disturbance = [P dem, S pv, T pv ] T t = 1 hr, H p = 6 hrs, H h = 6 hrs Solved via Dynamic Programming Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 28

45 Model Predictive Control w/ Cloud-enabled Forecasts Power7ykWG Load7Demand PV7Power Battery7Power Grid7Power SOC Cents&kWh 1 0f8 0f6 0f Electric7Rate7from7PGVE Monf Tuef Wedf Thuf Frif Satf Sunf Optimize for Grid Electricity Cost Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 29

46 Model Predictive Control w/ Cloud-enabled Forecasts PowergGkWb LoadgDemand PVgPower BatterygPower GridgPower SOC kg/kwh 1 0I8 0I6 0I4 0I55 CarbongEmissiongfromgCAISO 0I5 0I45 0I4 MonI TueI WedI ThuI FriI SatI SunI Optimize for Marginal CO 2 Produced from Power Plants Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 29

47 Load Forecasting - how accurate is accurate enough? NormalizediCostip)U DPiOptimal UniformiDistribution RBF NNiResult LoadiDemandiForecastiRMSE (kw) Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 30

48 Battery Health Aware (k+hp) t min J k = [λ ElecPrice(u, t) + (1 λ) Q loss (u)] 2 dt k t Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 31

49 Battery Health Aware Monthly Battery Capacity Loss (%) Utopia λ=1, Electric Cost Emphasis λ=0.89 λ=0.45 Battery Health Emphasis, λ= Monthly Electric Cost (USD) Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 31

50 Battery Health Aware Cell C rate (a) λ=1 λ=0.83 λ=0 Cumulative Grid Power (kw) (b) λ=1 λ=0.83 λ= Battery SOC 0 Off peak On peak Period C. Sun, F. Sun, S. J. Moura, Cloud Enabled Predictive Energy Management of a PV-Battery Smart Home Nanogrid in review. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 31

Smart Home Demonstration Project @ UC Davis Scott Moura

51 Smart Home Demonstration UC Davis Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 32

52 Smart Home Demonstration UC Davis Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 32

53 Outline 1 Forecasting Building Electric Demand 2 Residential Buildings with Solar & Storage 3 Integrating PEV Energy Storage with Buildings 4 Open Thoughts & Shameless Advertisements Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 33

54 Stochastic PEV Energy Storage X. Wu, X. Hu, X. Yin, S. J. Moura, Stochastic Optimal Energy Management of Smart Home with PEV Energy Storage in review. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 34

55 PEV-Home Nanogrid Operating Modes A) Household Electricity Demand B) Household Electricity Demand PEV Battery Utility Grid PEV Battery Utility Grid C) Household Electricity Demand D) Household Electricity Demand PEV Battery Utility Grid PEV Battery Utility Grid Renewable Energy A) Grid-to-Vehicle (G2V); B) Vehicle-to-Home (V2H); C) Vehicle-to-Grid (V2G); D) V2G w/ PV Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 35

56 Stochastic Mobility Needs Probability SOC at Plug out Time SOC at Plug in Time Transition Probability [%] plug-out time plug-in time 0 00:00 04:00 08:00 12:00 16:00 20:00 24:00 Time of Day Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 36

57 Stochastic Dynamic Programming Results 1 SOC Power[kW] Cost[cents] V2G V2H G2V No PEV 00:00 04:00 08:00 12:00 16:00 20:00 24:00 Time of Day X. Wu, X. Hu, X. Yin, S. J. Moura, Stochastic Optimal Energy Management of Smart Home with PEV Energy Storage in review. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 37

58 Stochastic Dynamic Programming Results No PEV V2G V2H G2V Cost(cents) Summer Weekdays X. Wu, X. Hu, X. Yin, S. J. Moura, Stochastic Optimal Energy Management of Smart Home with PEV Energy Storage in review. Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 37

59 Outline 1 Forecasting Building Electric Demand 2 Residential Buildings with Solar & Storage 3 Integrating PEV Energy Storage with Buildings 4 Open Thoughts & Shameless Advertisements Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 38

60 Explicit Consideration of Occupant Comfort + Cost Challenge: How to quantify and learn occupant comfort? (a) Electricity Cost (b) Occupant Comfort Performance + Sensor Radio S, x, u Occupant Survey Microcontroller Plant Dynamics, e.g. 3 room bldg x[t] u[t] Control Algorithm Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 39

software, algorithms, big data, cloud-based computing Topics Include: Energy Management and Power Systems Vehicle-to-Grid and Battery

61 CE 186 DESIGN OF CYBER-PHYSICAL SYSTEMS Fall 2015 in Jacobs Hall (NEW!) Project-based Course on H/W, S/W and Energy Fleet of escooters Indoor environmental sensing node Smart refrigerator Learn hardware, software, algorithms, big data, cloud-based computing Topics Include: Energy Management and Power Systems Vehicle-to-Grid and Battery Models Internet-based Systems Data Collection and Analysis Berkeley Energy and Climate Lectures Curriculum Innovation Award Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 40

Sample Course Projects (S14/S15): Grey-box Model for a Single Room System Identification Temperature Prediction and Optimization Model for Night Flushing Development of a Stochastic Model to Predict

62 Sample Course Projects (S14/S15): Grey-box Model for a Single Room System Identification Temperature Prediction and Optimization Model for Night Flushing Development of a Stochastic Model to Predict Office Plug Load Profiles and Energy Consumption Home Energy Disaggregation Optimal Energy Control in an Residential Building Predictive Energy Demand Side Management for a Residential Photovoltaic and Battery Solution Energy Management in Commercial Buildings Building Materials for Decreased Boundary Losses to Ambient Environment Optimal Refrigeration Control for Soda Vending Machines Empirical Estimation of Electrified Heating Load Curves using Daily Natural Gas Consumption Data CE 295 ENERGY SYSTEMS & CONTROL SOC(t) ˆ = 1 Q I(t)+ (V (t) ˆV (t)) ˆV (t) = OCV ( SOC)+RI(t) ˆ min J = u NX k=0 cfuel(xk,uk)+celec(xk,uk) V (t) = 1 2 xt Qx V (t) apple cv (t) Topics Include: Energy Storage & Renewables Electrified Transportation & Bldg Energy Streaming data analytics Optimal control Spring 2016: 3 units Prof. Scott Moura Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 41

63 VISIT US! Energy, Controls, and Applications Lab (ecal) ecal.berkeley.edu Scott Moura ecal Data, Controls, and Optimization October 19, 2015 Slide 42

Introduction. TBSI Opening Ceremony. Scott Moura

Introduction. TBSI Opening Ceremony. Scott Moura Introduction TBSI Opening Ceremony Scott Moura Assistant Professor ecal Director Tsinghua Berkeley Shenzhen Institute Civil & Environmental Engineering University of California, Berkeley TBSI Opening Ceremony