Scribd
Upload
4 views

Review of Smart Meter Data Analytics

Uploaded by

Muneebur Rehman
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
4 views

Review of Smart Meter Data Analytics

Uploaded by

Muneebur Rehman
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 25

Review of Smart Meter Data

Analytics:
Applications, Methodologies
and Challenges
Introduction
Problem Statement

 Since the early 2000s number of smart meters have been growing rapidly

 Smart Meters record fine-grained electricity consumption data

 This data can be used for a number of things like load forecasting, outage
detection, energy theft detection etc.

 Several research initiatives have been created to analyse this data in order to
improve forecasting and identify money-saving opportunities

 This research paper provides a comprehensive view on big data collection


projects and their applications, current research trends and future challenges
Current Research

 A search of key terms related to smart meter data analytics was done on the
Web of Science database

 In total 200 research papers were published between 2010 and 2017,
increasing from only 5 in 2010 to 60 in 2017

 A number of review articles have been written on Smart meter data analytics
some focusing on Non Intrusive Load Monitoring (NILM) some focusing solely
on load forecasting or Non Technical Losses (NTL)

 This review differentiates itself by providing a comprehensive review of Data


Analytics application and methodologies.
Datasets
Applications
Load Analysis
Bad Data Detection

 Missing Data or Unusual Patterns in Data

 Caused by failure of data collection, communication or entry

 Will lead to inaccurate models if not removed

 Methods like optimum weight averaging (OWA), low rank matrix methods and
neural networks have been used to detect bad data.

 Methods like clustering and windowing have also been utilised to remove bad
data in real time from the data stream
Energy Theft Detection

 Energy theft is also a form of bad data but appears over a longer time period.

 It can be detected by comparing smart meter data to node data

 Supervised and Unsupervised learning methods have been used to detect theft

 Getting labelled data for supervised learning is a challenge

 Consumers load profile is classified as normal or malicious utilizing a number


of strategies like using SVM classifiers or comparing with historical data.
Load Profiling

 Categorising consumers based on their electricity consumption patterns

 Techniques like K-means and Hierarchical clustering used

 Indirect Techniques like feature extraction and PCA have also been used

 Data sampling resolution of 30 minutes considered adequate

 Variability in load profile data is eliminated using averaging and Markov


models are used to capture dynamic consumption behaviour
Load Forecasting
Forecasting without Smart Meter data

 It is easier to forecast at higher levels by looking at the load profile or in


factories where consumption is driven by the work schedule

 For individual customers weather is by far the biggest factor affecting loads.

 Researchers have used weather data to predict consumer behaviour. Playing


around with the amount and location of weather stations to be included in
the forecast has shown to improve forecast accuracy

 The hierarchy of the power grid has also been used for forecasting starting
from the root node with child nodes forecasted proportionally
Forecasting with Smart Meter Data

 High-resolution smart meter data along with weather data allows for very
accurate short and long term forecasts.

 Techniques like SVM and neural networks have been used to generate
accurate forecasts at the consumer level.

 Consumers have also been clustered into groups with similar load profiles and
different forecasting models have been made for different groups

 Relationships between group size and forecasting accuracy have been


conducted using mean absolute percentage error (MAPE) as the evaluation
metric. Group sizes of 100,000 have been found to be error limiting
Probabilistic Forecasting

 Probabilistic forecasting involves predicting multiple scenarios, each with a


different chance of occurring

 Probabilistic forecasting models account for uncertainties which point


forecasts do not account for

 Generating different temperature scenarios is important especially as


renewable energy generation is highly affected by temperature

 Probabilistic forecasts can be combined with point forecasts to generate more


accurate forecasting models that are more suited to the modern grid
Load Management
Consumer Characterization

 Identifying consumers and grouping them on basis of consumption patterns

 Matching load profiles of consumers to sociodemographic status

 External data such as sociodemographic status can be used to predict load


profiles of consumers without smart meters

 Methodologies like regression and SVM have been used to match


socioeconomic status and standard load profiles

 Deep learning methods like convolutional neural networks have worked very
well on multi-dimensional data
Demand Response

 Demand response includes both marketing and implementation

 Marketing aims to identify consumers amenable to reducing their loads when


required

 Demand response implementation can be price based or incentive based

 Demand response includes dynamic tariffs based on load profiles

 Various clustering methods have been used to identify whether and how
customers can reduce consumption to match the generation
Miscellaneous Applications

Outage Management: Connection Verification:

 Power outage is defined as failure  Connection verification is the


of electric supply to consumers process of ensuring that the
recorded connections match the
physical connections
 Identifying outages quickly and
fixing them is a high priority
application  Voltage and current are used to
estimate topology of the
distribution system and the
 Smart meter data can be used to
impedance of branches
detect power outage and isolate
outage area for quick response
Challenges and Future Research
Data Privacy and Security

 Fine-grained energy data can reveal personal information, raising privacy


concerns among consumers.

 Techniques like data aggregation and PCA can be used to mask individual user
load profiles while still providing valuable insights for energy planning

 Mining and storage of this sensitive data by the utility is also in question as it
can always be hacked by malicious actors

 Using advanced demand response systems opens up the grid to cyber-attacks


where hackers can damage the grid and cause large scale outages
Big Data and Computing
Transition to Renewable Energy

 The transition to renewable energy brings new challenges both in terms of


generation and load profiles as well as the business model of retail energy

 Renewable energy is distributed, intermittent and weather dependent

 Load forecasting and demand response become increasingly important

 Rooftop PV owners also become suppliers of energy when generation is in


excess of consumption

 With distributed battery resources, it is a challenge to coordinate when to


store energy and when to release it back into the grid
Conclusions

 Smart meter data analytics is an important and rapidly evolving field

 Load analysis, forecasting and demand response are the most important
applications of smart meter data analytics

 As the amount of collected data gets bigger, more sophisticated


methodologies will be used to get highly accurate forecasts

 Consumer privacy and cybersecurity are the biggest challenges to solve

 Transitioning to renewable sources and distributed generation will require us


to have extremely accurate forecasting
Thank You

You might also like