The Interplay of Data Resolution and Appliance Detection Privacy - PowerPoint PPT Presentation

The Interplay of Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on The Interplay of Data Resolution and Appliance Detection Privacy in Smart Metering Multi-resolution Load Profile Representation Conclusion & Outlook Dominik Engel Appendix (joint work with G¨ unther Eibl) Josef Ressel Center for User-Centric Smart Grid Privacy & Security, Salzburg, Austria Dagstuhl Seminar 16032 -1

The Interplay of Outline Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Smart Meter Privacy Detection Multi-resolution Load Profile Representation The Impact of Resolution on Appliance Detection Conclusion & Outlook Multi-resolution Load Profile Representation Appendix Conclusion & Outlook 0

Diagram source: [9], Smart meter image source: Siemens 1

Diagram source: [9], Smart meter image source: Siemens 1

The Interplay of Non-intrusive Load Monitoring (NILM) Data Resolution and Privacy in Smart Metering Smart Meter Privacy Smart Meters Non-intrusive Load Monitoring User Acceptance The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Conclusion & Outlook Appendix Source: Samsung (upper left), Amazon (upper right), load profile taken from [6] 2

The Interplay of Non-intrusive Load Monitoring (NILM) Data Resolution and Privacy in Smart Metering Smart Meter Privacy Smart Meters Non-intrusive Load Monitoring User Acceptance The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Conclusion & Outlook Appendix Source: Samsung (upper left), Amazon (upper right), load profile taken from [6] 2

The Interplay of Non-intrusive Load Monitoring (NILM) Data Resolution and Privacy in Smart Metering Smart Meter Privacy Smart Meters Non-intrusive Load Monitoring User Acceptance The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Conclusion & Outlook Appendix Source: Samsung (upper left), Amazon (upper right), load profile taken from [6] 2

The Interplay of Non-intrusive Load Monitoring (NILM) Data Resolution and Privacy in Smart Metering Smart Meter Privacy Smart Meters Non-intrusive Load Monitoring User Acceptance The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Conclusion & Outlook Appendix Source: Samsung (upper left), Amazon (upper right), load profile taken from [6] 2

The Interplay of User Acceptance Data Resolution and Privacy in Smart Metering Smart Meter Privacy Smart Meters Non-intrusive Load Monitoring User Acceptance The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 3

The Interplay of User Acceptance Data Resolution and Privacy in Smart Metering Smart Meter Privacy Smart Meters Non-intrusive Load Monitoring User Acceptance The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 3

The Interplay of The Impact of Resolution I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 4

The Interplay of The Impact of Resolution I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 4

The Interplay of The Impact of Resolution I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 4

The Interplay of The Impact of Resolution I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 4

The Interplay of The Impact of Resolution I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 4

The Interplay of The Impact of Resolution I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix 4

The Interplay of The Impact of Resolution on NILM I Data Resolution and Privacy in Smart Metering ◮ Kolter’s REDD dataset [7] Smart Meter ◮ “Low” frequency dataset: ∆ t = 3 s Privacy The Impact of ◮ 6 houses with various appliances Resolution on ◮ Appliances (circuits) & mains Appliance Detection ◮ Focus on one house Resolution in Digital Imagery Resolution in NILM Multi-resolution Load Profile Representation Conclusion & Outlook Appendix Refrigerator Washerdryer 5

The Interplay of The Impact of Resolution on NILM II Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on ◮ Modeled as a classification problem at different Appliance Detection resolutions Resolution in Digital Imagery ◮ Event heights are stable → accuracy remains high Resolution in NILM Multi-resolution ◮ Fewer events are found → recall decreases with Load Profile Representation decreasing resolution Conclusion & ◮ The decrease is more pronounced for appliances with Outlook Appendix shorter on-durations 6

The Interplay of The Impact of Resolution on NILM III Data Resolution and Privacy in Smart Metering bathroom_gfi 0.64 0.62 0.76 0.39 0.00 0.00 0.00 dishwasher 0.83 0.51 0.32 0.28 0.04 0.00 0.00 Smart Meter kitchen_outlets2 0.43 0.49 0.47 0.57 0.47 0.33 0.00 Privacy kitchen_outlets3 0.81 0.68 0.65 0.52 0.00 0.00 0.00 The Impact of kitchen_outlets4 0.42 0.37 0.38 0.00 0.00 0.00 0.00 Resolution on Appliance lighting1 0.81 0.81 0.83 0.76 0.47 0.37 0.30 Detection lighting2 0.86 0.90 0.92 0.92 0.91 0.83 0.43 Resolution in Digital Imagery 0.81 0.89 0.90 0.92 0.92 0.79 0.38 lighting3 Resolution in NILM microwave 0.83 0.65 0.08 0.02 0.00 0.00 0.00 Multi-resolution Load Profile oven1 0.79 0.77 0.35 0.25 0.00 0.00 0.00 Representation oven2 0.94 0.84 0.31 0.21 0.00 0.00 0.00 Conclusion & Outlook refrigerator 0.94 0.91 0.81 0.80 0.29 0.00 0.01 0.77 0.00 0.00 0.00 0.00 0.00 0.00 Appendix stove washer_dryer1 0.72 0.73 0.47 0.35 0.00 0.00 0.00 washer_dryer3 0.99 0.78 0.21 0.12 0.00 0.00 0.00 3s 10s 30s 1m 5m 15m 1h Figure: REDD-Data Set. F-score matrix for simple edge detection. Small values are desirable for privacy. F = 2 pr p + r from [3] 7

The Interplay of Wavelet Transform Data Resolution and Privacy in Smart Metering Resolution Smart Meter Privacy Original Load Data r = 3 The Impact of Low-pass High-pass Resolution on Appliance H 3 r = 2 Detection Multi-resolution Load Profile H 2 r = 1 Representation Wavelet Transform Conditional Access L 0 H 1 r = 0 Open Issues Conclusion & Outlook Appendix from [4] 8

The Interplay of Wavelet Transform & Conditional Access Data Resolution and Privacy in Smart Metering Resolution Smart Meter Privacy Original Load Data r = 3 The Impact of Low-pass High-pass Resolution on Appliance H 3 r = 2 Detection Multi-resolution Load Profile H 2 r = 1 Representation Wavelet Transform Conditional Access L 0 H 1 r = 0 Open Issues Conclusion & Outlook Appendix k 0 k 1 k 2 k 3 L 0 H 1 H 2 H 3 from [4] 9

The Interplay of Open Issues I Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of Resolution on Appliance Detection Multi-resolution Load Profile Representation Wavelet Transform Conditional Access Open Issues Conclusion & Outlook Appendix 10

The Interplay of Open Issues II Data Resolution and Privacy in Smart Metering Smart Meter Privacy The Impact of ◮ Missing “privacy measure” for reduction of personal Resolution on Appliance information by subsampling (and in general) Detection ◮ ... and it’s always only an upper bound Multi-resolution Load Profile ◮ Are we over-engineering? Is there a problem to this Representation Wavelet Transform solution? Conditional Access Open Issues ◮ Number of resolutions needed in the real world? Conclusion & ◮ Which resolution for which use case? Outlook ◮ Communicating this to the end-user? Appendix 11