Big Data : New opportunities for EM&V of energy efficiency programs Annika Todd, Erin Hult April 2014
Data explosion in other fields Pedabytes of data collected by Google, Netflix, Amazon Development of new analytical techniques Pull useful and valuable insights from the data Example: Google tracks the number of flu-related searches to provide early warning signs of an outbreak 2
Similar data explosion in energy Smart meters, thermostats, appliances, cars Linked to other time and location-specific information (weather data, census data, satellite images ) Provide vast, constantly growing streams of rich data Sometimes called big data 3
This data enables new types of analysis What can we do with all of this data? Many possibilities! These data have the potential to provide tremendous value to a wide range of energy policies. 4
What can this data do for Evaluation, Measurement & Verification (EM&V) of energy efficiency (EE) programs? These data have the potential to enable: Quicker and cheaper answers More accurate savings estimates Allows new kinds of measurement, new kinds of programs 5
Outline Annika: examples of opportunities for residential programs Erin: examples of opportunities for commercial buildings 6
Outline Annika: examples of opportunities for residential programs Erin: examples of opportunities for commercial buildings 7
Outline Annika: examples of opportunities for residential programs Currently evaluated using a mix of deemed savings, engineering models Assumptions about baseline energy, installed measure performance and usage Erin: examples of opportunities for commercial buildings 8
Use easily accessible data to tell you what is actually happening Example: an EE air conditioning rebate program Using only hourly energy and temperature data, detect households that installed efficient AC, and measure the actual efficiency gain Scale this up to all of the households in a utility territory This household installed a 30% more efficient AC kwh kwh temperature temperature Based on actual performance and usage, rather than relying on assumptions Potentially easier, more accurate, and faster than collecting survey data, on-site data Especially helpful for upstream programs, where program participants aren t tracked Can also use this method to target specific households (e.g., those with the most inefficient AC) 9
More accurately determine net program savings by comparing program participants to a well-matched comparison group Program participant households Households in the comparison group Program begins Net savings kwh More data allows a better comparison group - a group of households matched along several characteristics that provide a counterfactual or baseline for the program participants (what would have happened without the program?) Hours (over four days) The ability to examine changes immediately after a program begins helps attribute savings to a specific program More data and more specific data allows more precise savings estimates - the savings don t get lost in the noise 10
Get quicker answers: determine if the program is working and make improvements in real time Example: Examine savings for two different types of households (e.g., households in two different geographic areas, or households with different characteristics) Estimated savings Savings from Type 1 households Savings from Type 2 households time Quickly determine if program is working Detect areas of improvement Fix problems in real time - improve marketing and outreach to Type 2 households 11
What big data can t do Useful insights require good data, not big data! Good data = data with a good underlying design, where we can see how things change between customers and over time in a way that allows us to tease out the causal relationships of interest If the underlying program is poorly designed, so that there is a fundamental bias in the savings estimates, no amount of data can help. Good data and good evaluation is driven by good program design. Designing programs up-front is very important! E.g., randomized controlled trials (the gold-standard), or good quasi-experimental methods 12
Conclusions Big-data and analysis methods provide many opportunities: Get quicker answers on program effectiveness Make improvements in real time Determine what s actually happening rather than relying on assumptions More accurately estimate program savings But requires good program design to get good data Contact: atodd@lbl.gov 13
Outline Annika: examples of opportunities for residential programs Erin: examples of opportunities for commercial buildings 14
Data & tools are changing energy management in commercial buildings In-house energy management Northwrite JCI Panoptix Fault detection & diagnostics Automated system optimization Remote Audit Tools Target large portfolios Retroficiency First Fuel Gridium Low-cost energy tracking Noesis WegoWise FirstView EnergyAI 15
How well can we predict energy use? M&V Quantifying load shed (DR) Near future load forecasting Energy baseline Fault detection & diagnosis Cost savings Detect waste in real time 16
Streamlined M&V is an emerging capability Emerging tools can conduct M&V at dramatically lower cost, with comparable or improved accuracy Offered in more advanced building information technologies, analytical software tools Baselines automatically created Historic interval meter data, weather data feeds, production data Regression models most common, also NN, Bin models User enters the date of ECM implementation, and savings are automatically calculated 17
New approaches to performance monitoring Behavior over time and for peer systems can inform performance Fault detection System optimization Savings estimation Tools increasingly incorporating multiple capabilities EcoFactor Example: HVAC performance diagnostics using web-linked thermostat 18
Automation can reduce cost & time required for M&V Reduce expert judgment required for M&V Real-time feedback Interval data models require fewer months of training data to predict future energy use NorthWrite s Energy Worksite provides monitoring & benchmarking 19
New tools allow new program types Streamlined M&V enables: Whole building savings estimation (IPMVP Option C) Multiple measures Interactive effects Building IQ Example: Automated HVAC system optimization tool Also includes M&V functionality following IPMVP Option C: whole building savings estimation 20
Interest to quantify accuracy of new M&V tools Pacific Gas & Electric Emerging Technology Program: How do we objectively assess vendors for consideration to include in programs? Consortium for Energy Efficiency (CEE): Where are the tests we can use to ensure that acceptance criteria and minimum performance levels are being met for M&V for whole building-focused programs? California Public Utilities Commission Error in baseline projection leads directly to error in estimated savings 21
How do we enable use of these emerging tools? LBNL Research Objectives: Develop standard test methods Remove key barriers questions of accuracy, transparency Outcomes: Testing methodology & protocols using cross validation & large test data sets Performance benchmark of ~5% median error for commonly used baseline models Used to prequalify tools for inclusion in 2013-2014 PG&E Whole Building pilot Training Prediction Outdoor Temp Metered Modeled Funding from PG&E and DOE Building Technologies Office For detailed results & more information, please contact: Erin Hult ELHult@lbl.gov or Jessica Granderson JGranderson@lbl.gov 22
Conclusions Emerging tools use use big data to monitor energy use and equipment performance, increasing M&V capability included Streamlined M&V methods can dramatically reduce costs at comparable or potentially improved accuracy Testing methodology provides a means to compare tool accuracy For detailed results & more information, please contact: Erin Hult ELHult@lbl.gov or Jessica Granderson JGranderson@lbl.gov 23