Power generation planning and economic load dispatch are the two most important decision-making processes in power generation. There, big data analytics enables the energy utility sector to optimise power generation and planning. But this is not always a smooth sailing practice.
Electricity grid modernisation initiatives and deployment of demand-side management (DSM) programmes increasingly depend on deriving actionable insights from the large volumes of data collected on the smart grid. The volume of data available to the utility industry is increasing exponentially in both number and type.
However, the lack of standardised and secure access to data and for analytical methods to extract actionable information from data is limiting innovation in DSM programme design and deployment as well as grid modernisation. Operations and performance data from power plants and renewable generation resources, electric power grid transmission and distribution system data, and smart meter data on electricity usage are all rich sources of information.
In this part-one article, using the findings of a paper titled Big-Data Analytics for Electric Grid and Demand-Side Management, big-data types and sources are explored along with the use of advanced analytics for big data related to the smart grid, with a focus on demand response (DR). There are three types of data analytics, namely:
• Descriptive analytics mine and aggregate data to provide insight into the past (‘what happened’).
• Predictive analytics utilise a variety of statistical, modelling, data-mining, and machine-learning techniques to study recent and historical data as a basis for forecasting the future.
• Prescriptive analytics use optimisation and simulation algorithms to suggest possible outcomes and recommend the best course of action for any pre-specified outcome.
For this paper, the research team focused on descriptive and predictive analytics for consumer-based DR programmes.
Electricity grid stakeholders and product vendors can leverage the findings from this study to identify and prioritise development of technologies for big-data and analytics-related applications to maximise DSM programme benefits and accelerate grid modernisation.
Data analytics for demand-response applications
DR programmes are widely recognised as essential tools for utility companies. Key benefits include peak-load shifting and potential elimination of costly spot-market energy purchases or capital investment in additional generation capacity.
Historically, consumption was calculated at an aggregated level and could not be easily apportioned across the customer base. Now, smart meters provide granular consumption data for the whole customer base. This data can be used to predict loadshedding from DR events.
DR-related predictive analytics (at varying time scales)
Increasing adoption of smart connected devices –such as thermostats; heating, ventilation, and air- conditioning (HVAC) systems, advanced lighting – is influencing the design of DR programmes, especially for thermostat-based DR.
In the residential sector, thermostat-based DR is a new business model for utilities, involving millions of clients each using far less power individually than the large commercial and industrial facilities that are the traditional targets of DR programmes.
Even though utilities want big (megawatt) DR resources, smart thermostats are already penetrating small and medium-sized commercial facilities to manage HVAC systems for efficiency and comfort and therefore offer DR potential.
The research team developed two predictive-analytics models for HVAC/thermostat-based DR. The two HVAC control strategies considered are: (1) shut down HVAC system, and (2) adjust HVAC system thermostat set points. The model predicts the possible load reduction (kW capacity) based on predicted building load, which is, in turn, based on historical meter data and current third-party weather forecasts (hour-ahead or day-ahead).
A previous Lawrence Berkeley National Laboratory (LBNL) project presented a method for fast, accurate prediction of kW capacity reduction using a physical (EnergyPlus) model. This model was improved recently to integrate a data-driven approach (using meter interval and weather data) as shown in Figure 1.
Figure 2 shows an example of analytics to predict DR capacity at a certain time of day, taking into account the weather forecast. The x-axis in Figure 2 (a) shows the estimated DR capacity of all commercial customers in a specific area. The y-axis refers to the ratio of the estimated DR capacity to whole-building power. The y-axis in Figure 2 (b) shows the distribution of each customer’s kW load-shed quantity in a specific area.
Results indicate that a majority of commercial customers have less than 10kW of DR capacity. Using a data analytical framework such as this, utilities can dispatch DR capacity at each location using the most cost-effective resources. In this study, ‘DR capacity’ refers to the potential kW shed from a building’s HVAC system during a peak four-hour event (e.g., 2PM-6PM).
DR post analysis (different time scales)
DR M&V quantifies DR performance in terms of the following metrics: total DR (kW shed during DR event hours), DR per building square foot or meter, and DR percentage of whole-building power (%WBP). DR-related post analysis includes the settlement of the load reductions achieved by each customer and at the programme level.
Different M&V methods are used for DR settlement based on DR resource characteristics such as load variability, weather sensitivity, etc. These baselines can also be used to estimate the large-scale potential of DR, assess the impact of the DR programme, and plan and operate DR programmes.
In the data set used, each customer’s smart meter measured energy use at 15-minute intervals. Generally, baseline loads are calculated using two models: (a) simple average over the previous 10 recent baseline days – which are normal operation days, excluding weekends, holidays, a DR event day, and any operational off day – (5/10 baseline) or the highest 3 or 5 out of 10 (3/5 out of 10 baseline), with and without morning adjustment; and (b) weather regression model baseline. These models are described below.
10 out of 10 baseline model (10/10)
The average load during the event period calculated from the previous 10 days (excluding weekends, holidays, a DR event day, and any operation off day).
10 out of 10 baseline model with morning adjustment (5/10 MA)
Morning adjustments is a ratio of (a) the average load of the first three of four hours before the DR event to (b) the average load of the same hours from the selected five baseline days. The adjustment factor is limited to ±20% of the customer baseline.
Weather regression baseline model
For the weather regression baseline model, a whole-building power baseline is estimated first, using a regression model that assumes that whole-building power is linearly correlated with the outdoor air temperature (OAT).
Although these data analytics models can help improve operation and performance of DR programmes, the ease of data access and the cost associated with large volumes of data make it challenging to extract the value from the data. The researchers are of the view that there is a need for standards-based data access schemes, which would simplify performance assessment of DR programmes.
[Use] this study to identify and prioritise development of technologies for big-data and analytics-related applications to maximise DSM programme benefits and accelerate grid modernisation.
Standardising data to facilitate demand-response performance assessment
The cost-effectiveness of utilities’ and DR customers’ use of big data to support grid interoperability would be enhanced by data standardisation. Grid interoperability refers to the grid’s ability to interface with disparate DR products, controls, or systems without requiring implementation-specific data translation. According to the US Department of Energy’s grid modernisation plan, “interoperability standards define technical requirements for defining the capability of two or more networks, systems, devices, applications, or components to externally exchange and readily use information securely and effectively”.
A majority of the standardisation principles discussed here are derived from the study team’s research. Prior studies address comprehensive applications for grid and customer transactions. This report focuses only on assessing DR programme performance.
Big data are characterised by an exponential “increase in number of connected devices or systems that can communicate and intelligently act upon information”. Extracting valuable, actionable information from such a deluge of data can be challenging. Realising the benefits of big data analytics required easy, cost-effective access to data from different sources and interoperable data exchange. DR-ready customer products must be able to exchange data and information with the grid; data standardisation will facilitate exchange.
Challenges for standardisation
1. Insufficient adoption of secure, standards-based networks that can sense, collect, and transmit data.
2. Lack of standard support for interoperability.
3. Lack of low-cost integration for fragmented DR systems and services from multiple electricity operators, providers, and vendors.
Standards that allow cost-effective, reliable data exchange among systems would help address these challenges. DR programmes do not exist in isolation but are part of a range of electricity or energy services implemented by customers, energy service providers, markets, and operators. The scope of standardisation for DSM is shown in Figure 3.
In the case of DR, standardisation would apply to data (e.g. a utility might request customer’s facility energy use data) or other information (e.g. a DR customer can request the baseline energy usage information from the service provider). Standardisation can have multiple formats depending on the type of service provided.
Standards for assessing demand-response performance
Table 1 lists some of the key standards and data sources used for DR performance assessment and communication. The standards listed in the table were developed by formal standards development organisations (SDO) unless such a standard does not exist, as noted.
The additional details are classified under, ‘de-jure’ and ‘de-facto’ standards. Here, de-jure refers to standard that is developed by an SDO and adopted by the industry. The de-facto refers to a standard that is not developed by the SDO and is still widely adopted by the industry.
Common methods to analyse a customer’s DR performance require historical and real-time energy usage data. Data analytics are applied to measured energy-usage data to quantify a customer’s load reduction in response to a DR event. Electricity grid and utility managers can benefit from understanding the relationships among the types of analytics and ways to employ various applications. For example, grid-asset and weather data can be used to manage the grid and to trigger DR events.
There is a need for standardsbased data access schemes, which would simplify performance assessment of DR programmes.
Although smart meters and automated metering infrastructure (AMI) are used for DR M&V, standardisation and harmonisation can enable integration among customers, energy service providers and energy markets and their systems and can thereby enhance DR services and performance assessment. For example, GreenButton and OpenADR or SEP standards could be harmonised.
An example from the field is from PG&E’s DR programmes where OpenADR-based management systems were integrated with customer information and meter data management systems to ensure that DR programme signals were dispatched to enrolled customers and to validate DR performance. Standardisation eases data sharing and integration across utility systems, enables many system architectures, and facilitates third-party access to data to help foster DSM technology innovation.
Big data analytics is a complex and yet an evolving exercise. In part two of this article, which will be published in ESI Africa Issue 4 2020, other dynamics of this application including data architecture, technologies, and applications will be explored.
This is article is based on an adaption of a 2019 paper titled Big-Data Analytics for Electric Grid and Demand-Side Management, written by Rongxin Yin, Girish Ghatikar, and Mary Ann Piette – of the Electric Power Research Institute. View online for a full list of references and diagrams. https://www.researchgate.net/publication/336029983