Smart Forecasting for a Smarter Grid

Most successful smart metering programs focus on business process management to maximize benefits (e.g., operational improvements) enabled by an Advanced Metering Infrastructure (AMI). Improved data visibility, frequency, and quality enabled by AMI technologies, can enhance future-state business processes and allow the utility to truly reach the expectations of the large investment they are making. Smart metering is much more than the limited implementation of AMI networks and advanced, digital two-way communicating meters. Smart metering not only provides the ability to change what is done, but also, how, when, and where process tasks and activities are performed to support existing end-to-end business process execution and the implementation of entirely new smart metering processes. AMI technologies should not in themselves define business processes. Instead, AMI should be leveraged as one piece of an overall smart grid program aimed to ultimately improve the business (AMI is an important component of the solution, not the solution itself).

Any smart grid vision will not emerge from a single rollout of revolutionary technology; the transformation will rather evolve through incremental investments that knit together this new intelligent infrastructure. The term "smart grid" is dynamic and likely defined differently by each utility or system operator (RTO/ISO) based on their scope of services, market structures, load profiles, and power portfolio composition. Smarter grid management capabilities, process management, and technologies that leverage key data from near or real-time monitoring, sensing, and decision support systems can enhance current state capabilities related to:

• Substation and distribution automation
• Renewables integration
• Transmission line loss optimization
• Outage management
• Distributed energy storage and generation

The objective of this article is to examine how pervasive consumption and usage data made available in volume, frequency, and quality as a result of an AMI/smart metering solution implementation may impact forecasting processes and tools.

There is great potential for vast amounts of AMI-enabled consumption data to potentially replace estimated load profiles, however, many back office applications currently used to support forecasting processes are not capable of handling and aggregating these huge data volumes efficiently to meet operational, regulatory, and customer requirements across the various departments within the utility. The ability to process these huge volumes of data combined with the time required to do so and costs involved may not provide a compelling value proposition to switch from sample population load profiles.

Utilities and ISOs may deploy different applications, processes, and related technologies to support current state energy forecasting and settlement operations. Regardless of the particular approach currently in place, we believe these organizations would be in a better position to evaluate the specific impacts of AMI-enabled data on their operations by understanding the general impacts presented below.

Forecasting and Load Profiles

Forecasting can be thought of as the set of processes, activities, and toolsets used to create predictions to support operational decision making. Across the electric value chain, forecasts are used to identify how much electricity will be needed in the future, to predict monthly sales revenue and unbilled consumption, and to support asset management, loading analysis, and predictive maintenance. A primary objective of forecasting is to produce the most accurate forecast by optimizing key variables and methodologies.

In many organizations, one department is centrally responsible for gathering and measuring key system load data and studying the characteristics of that data to make critical decisions; maintaining forecasting models; identifying consumption pattern trends and key attributes; and managing forecasting processes and load profiles. Conversely, these responsibilities may be distributed across multiple departments.

Important characteristics of forecasts include:

• Forecasts vary across season, by day, and within the day
• Customer class-level forecasts are impacted by unique variables that may be different from those variables impacting system-wide forecasting processes
• Some very large commercial and industrial customers may be analyzed separately and result in customer-specific forecasts
• Forecasts should include reserve margin requirements to ensure system reliability and also to ensure the ability to meet peak system demand - reliability can be managed by optimizing cost of conformance to the cost of non-conformance
• Effective practices include forecasting by customer class as well as end use; utilization of advanced forecasting models; and utilization of 3-point estimates (pessimistic, likely, optimistic) versus a single forecast

From a statistical perspective, forecasts are mathematical prediction models where a dependent variable, the forecast, is calculated using various independent variables that are typically the most highly correlated to the dependent variable. The term `forecast' implies an estimate, and in those cases where the forecasted variable contains input measurements that are based on a subset of a larger population, load profiles are often used. A load profile can be envisioned as a graph of the variation in the electrical load over time. It is important to note that load profiles can be estimated or actual, depending on the availability of interval data used to support the calculation of the load profile.

Load profiles can be classified as static or dynamic:

• Static profiling - used in those situations where demand is relatively constant; may be based on engineering load calculations; may refer to unmetered loads such as public lighting
• Dynamic profiling - a load profile whose hourly load levels are assigned no less frequently than daily based on actual conditions; loads are less constant across time and can vary significantly across different customer classes, with a dependence on temperature, weekday and holiday schedules, seasonal fluctuations, and economic conditions

How Forecasting and Load Profiles Are Applied

Forecasts and load profiles are used for a variety of purposes supporting local distribution as well as regional grid operations.

Local Distribution Operations

• Load forecasting (supplier scheduling, demand planning)
• Financial planning (budgeting, revenue, unbilled collections, simulation)
• Asset management / Distribution planning (transformer sizing)
• Marketing (demand analysis, program segmentation)
• Pricing (cost allocation, rate analysis, rate design)

Since electricity is, for the most part, a commodity that must be used as it is generated, it is critical to balance demand with supply. Local distribution companies that purchase energy through competitive markets or through single/preferred supplier contracts provide forecasts of their anticipated energy requirements. Forecasts can be used to predict annual system peak load requirements (the highest load the system should be prepared to serve at any one time) and periodic system energy requirements (predicted total consumption) per period. Load profiles support distribution planning operations from the perspective of energy efficiency and reliability and assist utilities in determining if distribution assets can handle system demand requirements for tactical load balancing, preventive maintenance, and long term asset and circuit planning. As an example, transformers are rated for optimal and maximum load patterns and are designated by specifications ranging from average and maximum load factor, utilization, and cycle rating. Load profiles help to ensure these critical and expensive assets are not overloaded and efficiently serve their purpose.

Regional Grid Operations

• Load forecasting (supplier scheduling, demand planning)
• Distribution planning (switching, power quality)
• Capacity planning (transmission and generation investment)
• Market Settlement (delivery reconciliation)

ISOs and RTOs rely upon load forecasts for various purposes including maintaining grid reliability and managing settlement operations. Settlement generally refers to the balancing and reconciliation of financial obligations between power producers and retail energy providers. In those situations where a utility may have under-forecast its energy consumption requirements for a specific period, grid operators take on the responsibility of providing for the additional capacity by engaging spinning reserves or purchasing electricity from the spot market. ISOs often reconcile what was scheduled ahead of time to actual loads provided and monitor the accuracy of the settled load compared to the reconciled load. While energy generation, transmission, and distribution are characterized by instantaneous production and consumption, supplier obligations are often settled on an hourly or sub-hourly basis. The utility industry, however, has been traditionally characterized by a limited availability of detailed interval data and the measurement of actual consumption data often occurs on a monthly cycle - significantly higher than the hourly interval timeframes required for energy settlement operations. Utilities have therefore created and adapted processes based on these limitations. Load profiles are often used to convert monthly total consumption into estimates of hourly consumption. For each hour or required time interval, estimates are determined for all customers belonging to similar customer classes across an energy supplier's service area.

Generating units use load profiles to size their operations to help ensure that a combination of base load and spinning reserve assets can meet customer demand patterns.

In summary, load forecasts are the basis for planning decisions and contribute to strategic and tactical execution of business processes, helping to answer key questions such as:

• How much energy should be procured to support efficient and reliable energy delivery?
• How much capacity is available to meet mid- and long-term energy requirements?
• Will additional generation facilities be required? If so, when?
• How much load can be met through demand-side management techniques such as interruptible demand and demand response?

High Level Potential Impacts with AMI-enabled Data

Many aspects related to forecasting in general, forecasting processes, and the use of load profiles do not change with the availability of high volume, granular interval consumption and demand data provided by AMI systems.

• Estimates, by nature, do not represent actual consumption - they are a representative sample for a larger population. Estimating techniques try to optimize variances as much as possible, however the variances between estimated consumption and actual consumption can result in wasted energy that is not consumed and billed for, to the reverse condition where supply does not meet demand.
• Development and maintenance of forecasting models is expensive and time consuming.
• Core forecasting tasks including budgeting, planning, asset utilization and load analysis will continue to be used.
• Customer billing cycles that are not aligned to monthly financial reporting cycles will continue but they may not be route based.
• Load profiles are likely to be required at the asset level, customer class level, within a customer class, and aggregated at system level.
• Forecasting completed at different levels of the energy value chain are still subject to the bullwhip effect - a situation whereby higher level forecasts based on lower level forecasts results in a compounded variance. Lower level load profiles based on actual data can theoretically reduce expected variance to zero.

Conversely, there are some expected changes and potential impacts to forecasting and data analysis processes as a result of the availability of AMI-enabled data and the introduction of new system assets such as the meter data management system (MDM).

• Theoretically, estimated load profiles can be abandoned if actual consumption data is available for an entire population.
• Sources of consumption and power quality data may change with AMI implementations. As MDM systems come online, data repositories and applications may switch to the MDM as the system of record and away from custom databases that are in use today.
• Integrations between existing systems may need to be updated or replaced to link to the new data sources and handle increased data payloads. Business models and processes involving multiple value chain partners such as generators, distributors, and retail merchant providers that are linked through complex data integration schemas supporting meter reading execution, consumption billing and invoicing, and T&D infrastructure charges may be affected. These integration points, data flows, data payloads, and frequencies are all subject to evaluation and potential impacts will need to be evaluated and addressed.
• Resources supporting financial settlement and load research operations may require training on new processes and procedures using the data provided from AMI networks.
• Master data synchronization between source and target systems remains a challenge in many industries as part of general systems integration challenges.
• Rebate and re-bill processing can impact actual consumption. Load profiles that do not account for material variances between projected and reconciled consumption can affect future forecast requirements and profitability. Load profiles based upon actual data consumption data across customer classes can reduce the risk associated with over- and under-forecasting and profitability.
• Financial cycle analysis based on near real-time AMI data can provide usage or consumption data that is better aligned with available production and delivery data; projections of unbilled consumption are perfectly known at any point in time and may no longer rely upon estimated load profiles.
• System loss information is more accurately known with generation, delivery, and consumption data being readily available; root causes of energy loss can be more specifically pinpointed and addressed.

Considerations with AMI-enabled Data

As expected, the previously identified high level impacts have relevant costs and benefits associated with each of them. In some cases, it is likely that the availability of AMI-enabled data to support load profile and forecast development is not worth the effort to upgrade systems and processes capable of processing the data. An important clarification here is that AMI is very relevant to achieve many operational processes whose benefits surpass implementation costs but the focus here is on load profile and forecasting activities. Time of Use pricing, real time pricing, demand response, and outage management cannot achieve their potential without an AMI infrastructure and supporting smart meter processes.

Three primary challenges exist related to the use of AMI-enabled data to support load profile development and forecasting processes include:

• Back office system capability / performance
• Compressed time requirements to provide and track specific energy profiles per regulations
• Marginal cost vs. ROI

An MDM system is specifically designed to handle the high volumes of data collection, transmission, validation, aggregation, and storage as AMI meters are installed in large quantities across service territories. This is the purpose that an MDM is built for, and suited for. However, this does not mean that back office systems can efficiently process the same volumes of data to create `actual' load profiles vs. estimated load profiles and existing data integrations may not be scoped to process the required bandwidth.

Some MDM validation processes require up to 24 hours before the previous day's consumption data is available at the interval level. This time lag may not facilitate processes requiring hourly load profile data based on actual vs. sample estimates.

The incremental benefit(s) of using actual data rather than class-based load profiles may not justify the incremental costs of processing the data. An industry standard for acceptable forecast variance is about 8%. A utility should evaluate whether the use of actual consumption data that might be available for complete customer classes is worthwhile compared to the costs of acquiring and aggregating that same data. In other words, will the benefits of increased percent accuracy or variance reduction improve system capacity planning significantly? Some MDM systems include load forecasting capabilities as extensions to their core functionality. This is potentially a criterion to consider when evaluating different MDM platforms.

Realistic Opportunities within Local and Regional Grid Management

The following examples illustrate practical improvements to load profile utilization and energy forecasting processes that can be implemented by utilities as well as ISO/RTO organizations. These application areas include capacity offset forecasting, infrastructure investment, cost allocation, asset management, and settlement.

Capacity Offset Forecasting - Assuming smart meters and associated demand management processes and technologies are implemented, it is logical to conclude that identifiable and achievable load reduction information (available through demand response and interruptible load control programs) can be leveraged to more accurately offset periodic consumption and peak demand forecasts. Many utilities and ISO/RTO organizations use capacity adjustments when determining net demand. Although historically these sources have been limited to large commercial and industrial customer classes that have interval-capable metering installed, as greater saturation levels for deployed smart meters occurs, the availability of detailed consumption information across customer classes provides a larger population from which to draw specific profile data. Grid operators may be able to increase offset estimates with reduced variance and improved confidence once data for the residential customer class becomes prevalent. This is important because they are accountable to possible under-forecasting at the distribution level to maintain grid level supply and demand.

Infrastructure Investment - Better control over demand typically implies better overall utilization and investment. Additional generation sources, whether intended to serve as base load or spinning reserves, are continuously evaluated based on consumption information, population growth, and other economic indicators. Construction, operations, and maintenance costs associated with these investments are typically passed on to customers through rate designs regardless of the utilization of such assets. Avoidance or deferral of these costs is one way utilities and grid operators can maintain existing cost structures rather than increase energy costs - thereby potentially avoiding the cost of additional generation being passed along to consumers.

Cost Allocation - Along with recent legislation mandating carbon accounting and reporting and renewable portfolio standards, new requirements are being implemented to support more precise allocation of commodity and distribution charges to relevant customer classes. This information directly impacts rate case analysis and approvals. As an example, the California Public Utility Commission has in recent years required distribution companies to create and manage load profiles within customer classes. Where load profiles for commercial customers were once good enough on aggregate level by customer, new oversight regulations mandate the need to maintain load profiles for consumers in high rise buildings compared to large footprint acreage, central heating and air conditioning vs. window units and swamp coolers, CFL vs. fluorescent lighting, and distributed generation capability vs. none. More defined rate rates within existing rate classes can be more effectively designed if more granular and specific consumption data is made available.

Asset Management - Improvements appear readily observable for transformer asset management and circuit analysis. An advantage of having smart meters located downstream of transformers facilitates and supports automated monitoring resulting in optimized lifecycle utilization. Smart meters are capable to supply detailed, accurate consumption data as well as demand data for each premise. No longer are estimated profiles required at this level of network analysis since they are easily packaged by the MDM and supplied for all major distribution assets. Not only is actual consumption and demand data available for transformer and circuit assets with AMI, the load data is now available much sooner than possible today when based on load profiles disaggregated from monthly billing cycle data. Real time data for assets mapped to transformers can help to ensure these critical and expensive assets are not overloaded and efficiently serve their purpose.

Settlement - In theory, actual consumption data across customer classes should also facilitate energy settlement transactions, but this too makes the assumption that back office applications can process this data more efficiently than using load profiles. In addition, settlement timeframe constraints may or may not support long processing times. While AMI certainly improves and increases the granularity of measured consumption, the data volumes associated with AMI are staggering.

Conclusion

Many utilities have installed interval-capable metering assets at key strategic locations to support data capture serving both as settlement load profile and distribution load profile process inputs. Advanced mechanical and digital meters capable of capturing hourly (or less) consumption data have been deployed for many medium and large commercial and industrial customers yet these assets have not permeated the residential and small commercial landscapes until recently with the advent and support for smart meters and AMI network implementations. As the cost of smart metering devices has continued to decrease, while functionality has increased, the industry is now at the point where widespread availability of smart metering for residential customers is becoming a reality and this could have significant impacts on load profiling and forecasting expectations - adding new opportunities and capabilities to aid in accurate forecasting.

This publication contains general information only and is based on the experiences and research of Deloitte practitioners. Deloitte is not, by means of this publication, rendering business, financial, investment, or other professional advice or services. This publication is not a substitute for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your business. Before making any decision or taking any action that may affect your business, you should consult a qualified professional advisor. Deloitte, its affiliates, and related entities shall not be responsible for any loss sustained by any person who relies on this publication.

As used in this document, "Deloitte" means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/about for a detailed description of the legal structure of Deloitte LLP and its subsidiaries.

Copyright c 2010 Deloitte Development LLC, All rights reserved.

Want more news and insights? Sign up for a FREE industry newsletter