1. Introduction

Electricity demand estimation is crucial to enable optimal power generation, transmission and distribution grid expansion. In India, the Central Electricity Authority (CEA) is the apex body which regularly undertakes the exercise of medium- and long-term demand estimation for states and discoms through the Electric Power Survey (EPS). The EPS forecasts annual electricity demand for five years on an annual basis using the partial end-use method. The Electricity Regulatory Commission (ERC) and distribution companies are guided by the EPS estimates. Yet, in the past¹ it has been observed that there are significant variations in the EPS projections and actual electricity demand. CEA has recently published guidelines for medium- and long-term demand estimation in utilities with the aim to bring uniformity in approach.

Electricity distribution companies (Discoms) on the other hand are also required to undertake rigorous task of demand estimation which forms the basis for resource planning i.e., power procurement and network expansion. Discoms undertake demand i.e., sales and loss estimation for short and medium term as a process of the annual revenue requirement (ARR) and multi-year (MYT) tariff process as mandated by respective State ERCs.² It is observed that there is significant variation in the method and data used by discoms for the process of sales estimations. In this article, we present an overview of different approaches prescribed by state regulations and practices adopted by discoms for sales estimation. We hope this comparative analysis provides direction for adoption of better sales estimation techniques across states.

The article covers notified MYT regulations for 28 states (including Delhi)³ & 5 UTs, ARR Petitions and Tariff Orders of 62 discoms. The terms Sales estimation- Sales forecast- Sales Projection have been used interchangeably across MYT regulations and tariff petitions. Sales estimation methods for a few union territories and Sikkim state have not been covered owing to unavailability of recent documents. The article presents details of the different methods suggested by MYT regulations, an overview of the methods and data used by discoms and best practices adopted by some of them.

2. Sales Estimation Approach suggested in Regulations

Most MYT Regulations provide direction on the approach, factors, time period and methods that shall be used by discoms for sales estimation. Table 1 provides an overview of the different factors included in these regulations. Excerpts from MYT regulations of Maharashtra Electricity Regulatory Commission, Joint Electricity Regulatory Commission (JERC) and Andhra Pradesh are included here to provide a quick snapshot of the persuasive directions included in the regulations.

MERC MYT Regulations, 2019 (82.1) “The Distribution Licensee shall submit a month-wise forecast of the expected sales of electricity to each Tariff category/sub-category and to each Tariff slab within such Tariff category/sub-category to the Commission.”

JERC MYT Regulations, 2021 (8.6 (a)) “The Distribution Licensee shall forecast sales for each Consumer category and subcategories, at different voltage levels, for each Year of the Control Period.”

APSERC MYT Regulations, 2018 (10.10) (4) “The forecast for the ensuing year shall be on monthly basis to properly capture the seasonality in demand.”

Table 1: Summary of the sales projections approach defined by SERCs & JERC⁴.

This overview shows that many elements which typically influence future sales have been included in several state MYT regulations.  However, many regulations lack the ability to provide details of methodology and data that should be used by discoms to arrive at the forecasts.  For example, some commissions direct discoms to consider the impacts of factors such as seasonal and weather fluctuation, implementation of demand side management measures, energy efficiency and energy conservation policies, transmission and distribution loss reduction initiatives, open access (OA) transactions, social-economic and technological changes on sales. However, detailed data and methods to be used, to assess impact of these factors on sales estimation are not provided.

3. Sales Estimations Approach adopted by Discoms

Discoms are required to submit year-ahead sales projections for determination and approval of their annual revenue requirement (ARR) to respective Commissions. In this study, we reviewed the tariff petitions filed by discoms and tariff orders (issued by Commissions) to better understand details of the sales forecasting approach and methodology adopted by them. For the states of Andhra Pradesh and Telangana, we also reviewed the business plans and resource plans filed by discoms as they include granular data which has been used for demand estimation. This data has also been used to perform sales estimation analysis using different methods, presented later in this article.

As discussed in the previous section, MYT regulations recommend that the sales forecasting method should consider the impact of multiple factors that can potentially influence demand in the future. At the same time, it is also expected that granular data should be used by discoms to develop an accurate sales estimation. The overview of existing tariff petitions and orders shows that discoms are not able to adhere to all suggestions provided by such regulations. Table 2, below provides a summary of the methods adopted and factors considered by different discoms for sales forecasting through the review of ARR /Tariff Petitions and Business/Resource Adequacy Plans.

Table 2: Summary of the approach and methodology adopted by discoms for Sales Projections⁶.

This overview also shows that among different forecasting techniques, CAGR and trend analysis are used by approximately 75% of the discoms. Trend Analysis method is a non-causal forecasting technique assuming that the factors follow the same historical trend as past in future and focus on identifying the pattern and direction (increasing, decreasing or stable) of growth over time. Whereas, CAGR measures a smoothed annual growth rate while eliminating any fluctuations of data assuming a steady growth for a specified time period longer than one year. The historical actual or audited sales, number of consumers, and connected load data are used to arrive at the CAGR. Although, the number of years used for determining the CAGR varies between 7 years to 3 years most common being 5 years, and some discoms also use annual or sub-annual data.

It is observed that in states such as Gujarat, Madhya Pradesh, and Uttar Pradesh all discoms use a uniform method for sales forecasting. However, in some other states like Andhra Pradesh, APEPDCL uses linear regression method for sales forecasting whereas APCPDCL uses monthly linear trend method and APSPDCL uses CAGR method for sales forecasting.

The Joint Electricity Regulatory Commission (JERC) and 9 other SERCs have also directed discoms to submit month-wise sales projections. But our analysis shows that only discoms of Andhra Pradesh and Telangana include category-wise monthly sales projections in their tariff petitions. Another interesting observation from the Department of Power, Arunachal Pradesh mentioned⁷, “As seen from the category-wise monthly sale for the FY 2022-23 there is no effect on seasonal change, hence annual basis forecast has been adapted instead of a monthly basis.”

4. Good practices included in regulations and adopted by discoms

In the previous section, we observed that most discoms adopt a consumer category-wise CAGR method for estimating future sales. These numbers may not necessarily be representative of the variations in consumption at circle or division level across consumer categories, for example, Some circles may observe higher growth in demand in comparison to discom-wide CAGR for particular consumer categories. This information will be helpful for more accurate power purchase planning and distribution network expansion. Some discoms have incorporated more granular approach in their sales estimation and data collection practices. Table 3 presents a brief overview of good practices adopted by discoms and included in regulations.

Table 3: Summary of good practices adopted in sales projections methods by SERC and discoms.

It is interesting to observe that 3 state regulations have directed discoms to develop robust database of consumer demand to aid better future demand estimation. MYT regulation by Odisha Electricity Regulatory Commission (OERC) is the only regulation which suggests discoms to use load duration curve for sales forecasting.

Another interesting practice by Andhra Pradesh discoms, especially Central discom (APCPDCL) is that they have included consumer categories, such as Township, Religious places, Showrooms, etc. within HT category. They include category-wise monthly projections for each of these categories in sales estimation. The industrial sales during peak and off-peak time-of-day are also estimated for future by APCPDCL.

5. Recent Changes in Regulations

The states of Gujarat and Maharashtra have issued new draft MYT regulations in 2023 and 2024. Also, Karnataka ERC has recently notified new MYT regulations on 24^th June 2024. A review of these regulations has been captured in the table below. These regulations are the first to emphasise the need for long-term and short-term demand estimation, using granular data sets for all consumer categories. They also recommend discoms to utilise advanced estimation techniques for sales forecasting.

Table 4: Review of draft MYT Regulations for Gujarat, Maharashtra and Karnataka.

These regulations suggest inclusion of better sales estimation practices like use of scenario-based analysis, inclusion of load forecasting studies, use of granular data and development of consumer database.  But it needs to be seen if future tariff petitions of discoms in these states incorporate suggested methods for sales estimation.

6. CEA Guidelines

Early last year the Central Electricity Authority published guidelines for medium- and long-term power demand forecast (July 2023) for utilities in the Indian Power Sector. The guidelines are intended to provide direction to utilities in order to bring uniformity in the approach of demand forecasting. These guidelines recommend that the annual growth rate for each consumer category can be derived using two simple and statistically appropriate methods, namely ‘least squares method (LSM)’ and ‘weighted average method( WA)’.

The least square method is a mathematical approach for regression analysis. It focuses on finding the best-fitted curve for a set of data points while minimising the sum of the squares of the differences between observed values and predicted values. In the weighted average method, each value in the datasets is assigned a specific weight which is used to calculate the average growth rate. In this method, certain values are assumed more significant and reliable over others. In current sales forecasting practice, we observe that only APEPDCL uses the linear regression method which follows the rule of least square technique to estimate the sales estimation.

Although CEA has recommended the use of specific methods, namely LSM and WA, the review of existing sales forecasting practices conducted by this study shows that CAGR is the most common method adopted by discoms for arriving at annual growth rates. Hence, we performed a simple sales forecasting analysis using these three different methods based on historical sales data for two discoms of Telangana TGNPDCL and TGSPDCL. Figures 1 & 2 present the results of this analysis.

Figure 1: TGNPDCL Projections for 2018-19, 2023-24

Figure 2: TGSPDCL Projection for 2023-24

We observe that for TGNPDCL, the projection for the year 2018-19 using the weighted average method was closer to actual sales for the year in comparison to CAGR and Least Square method. It needs to be seen whether the same will hold true for the projection in 2023-24 when actual sales data is published.

7. Conclusions and Way Forward

To conclude, this study finds that most discoms focus on sales estimation exercises in short and medium term for different consumer categories.  But efforts on loss estimation which is also a contributing factor to total demand is conducted only at the discom level. CAGR is the most commonly used method for sales estimations across discoms. However, the spatial data and time granularity for estimations are low.

Some of the major challenges observed in current sales estimations approaches followed by discoms are:

 1) Limited granularity in consumer types:  Most discoms aggregate consumer types representing different consumption behaviours under single consumer category, and calculate average growth rate for all alike. For example, similar CAGR is used across all residential consumers whether rural, urban, living in high rise or slums potentially leading to over or underestimation of sales which can impact supply and service quality to consumers.

 2) Clarity on assumptions and data values is lacking: No clear explanation or data is provided by discoms on assumptions used to arrive at annual growth rates. Such lack of clear information makes it difficult to also aggregate the sales projections at the country level.

3) Validation of historical sales projection absent: None of the regulatory processes seek to validate past sales projections. Such a process could potentially offer measures to introduce corrections in assumptions used for determination of growth rates for the future.

 4)Variations in time granularity:  Significant variations have been observed in historical time granularity used by discoms. This may have impacts on future estimations especially when demand variations owing to extreme weather changes are increasing and unforeseen events like COVID-19 alter demand patterns.

With increase in renewable (re) energy penetration in the grids, increase in newer electrical loads like EV and induction cookstoves and extreme climatic occurrences the demand for electricity needs to become more flexible. In this context, accurate prediction of demand becomes more important. Adoption of more granular and comprehensive methods of sales forecasting will hence become necessary. Several state Electricity Regulatory Commissions have also emphasized the need for discoms to use better demand estimation techniques through the incoming resource adequacy regulations. However, details on the granularity of data, categorization of consumers to capture spatial differences and objective evaluation of different techniques are missing.

The CEA and ERCs can provide a more detailed and uniform data collection frameworks for discoms to help increase the granularity of data in terms of consumer categories, spatial spread, time duration and list of factors that impact demand. This process can also help in evaluating best techniques for sales forecasting and also test the accuracy of historical projections. At the same time discoms can support this process by collecting better feeder and distribution transformer level data in finer time blocks (15 minutes, hourly, daily), circle or division level consumer category-wise/connected load-wise sales data and provide detailed information on assumptions used for sales estimation. This will help in not just better demand projections but also load projections which are necessary for better power procurement and planning, and improving reliability of electricity infrastructure.

Endnotes