Adding value to storage of smartmeter consumer data. Jan Beyea, Ph.D. Consulting in the Public Interest

Transcription

1 Adding value to storage of smartmeter consumer data. Jan Beyea, Ph.D. Consulting in the Public Interest

2 Scientific research potential Energy analysis Social science Energy economics (e.g., elasticities) Sociology History

3 Health studies EMF epidemiology Covariate data - e.g., confounders Intervention studies - Dietary interventions Science, Vol 328, pages On Ning web site.

4 Impact of possible research use on acceptance of smart meters Both pluses and minuses Could increase privacy concerns among some Use of data by other parties than utility could increase support among others.

5 Without access, no 3 rd party research "Opt in" is ok, although "opt out" for some kinds of studies might fly. instantaneous analysis not usually important.

6 Without data, no data analysis by anyone Consumer resistance is growing. because of concerns about privacy the fear that utilities will use SM to raise bills. and equity issues raised by NGOs

7 Are there extra customer benefits beyond cost, reliability and efficiency? Benefits that will provide incentive to participate in collection and storage?

8 Examples of unanticipated beneficial use Landslide/mudslide tracking. Tang 2010 Smoke detector connection? Preventing electrical fires by early warning of high risks

9 Electrical fires in US More than 40,000 electrical fires per year several hundred deaths /yr thousands injured/yr More than $½ billion damages/yr

10 Preventing electrical fires using smart meters An idea "in progress Arc-fault circuit interrupters rarely in old housing. And, not clear how many fires AFCI will prevent Consider one potential future

11 Future mailings to customers Your odds of an electrical fire in next 5 yrs = x based on smart-meter studies Odds higher if modifications done by inexperienced workers No inspection etc What you can do to reduce your risk: Use newer extension cords Etc

12 How can smart meters reduce risks of electrical fires? Research stage Post-research analysis of smart meter signal to estimate customer risk Communicate individualized risks to customer Assume customer will act if odds great enough Educate customers in general with mailings

13 Research stage Compare pre-fire signals from cases and nonfire signals from controls Link to census & meteorological data Interview occupants Link answers Extract odds ratios by logistic regression For independent variables - as is standard in epidemiologic studies Or use neural nets, genetic algorithms

14 Access to data through utility servers or trusted third party servers No saving of data by researchers. However, data must be saved somewhere Don't know who will have fires.

15 Preliminary hypotheses Fire risk higher if low income. More likely to get incompetent help or do it yourself. Fires correlated with electrical consumption that is high for age of housing and # of occupants high for many appliances and increasing over time Some fires correlated with arcing Sporadic bursts may be detectable by smart meters

16 More initial hypotheses Fire risk correlated with frequent triggering of circuit breakers or fuses That may be detectable with SM. Fire risk correlated with age of housing. Fire risk correlated with geographic area e.g, areas where inspection of work not as common Communication of individualized fire odds is an opportunity for public education

17 Fire-risk warnings should increase support for storing smart-meter data. A potentially reasonable trade off for some increased risk of privacy loss. Should build positive views of using SM data for research.

18 Research challenges What resolution needed? 15-minute resolution may not be ideal - Can software changes help? 1 minute or 1 second data in Europe - Perhaps, study best done there first

19 Sample 1-second data

20 Other research challenges Missing data & imputation Selection bias Uncollected data Separate study identifying sample distribution - Allows better weighting How will missing data affect inferences? A problem that goes beyond research How should missing data be handled? What steps should be taken now to minimize missing data in the future?

21 How long need to wait to get enough data? To get 1600 cases in US When 2% of country has 1-second or 1-minute smart meters Expect 800 fires per year in population Two years of data gives you 1600 cases. These are non-fatal fires

22 Suggestions welcome For improving the concept