If you run a power utility, guessing wrong hurts. Underestimate demand and you end up scrambling for expensive short‑term purchases—or worse, you impose outages. Overestimate it and you’re stuck paying for capacity you don’t need. That’s why a careful, data‑driven demand forecast isn’t a “nice to have”; it’s the backbone of reliable, affordable electricity. The study summarized here lays out a sensible, transparent approach to projecting Bihar’s electricity needs over the next decade and, just as importantly, explains the assumptions behind the numbers.

Why demand forecasting matters (now more than ever)

Load forecasting touches every major decision a distribution utility makes: long‑term power procurement, network augmentation, tariff design, and day‑to‑day dispatch. Because new power plants can take 3–5 years (gas) to 7–12 years (thermal/hydro) to come online, we don’t have the luxury of “waiting to see” where demand goes. Planning must look 10–20 years ahead. At the same time, short‑term forecasts shape tariff setting and help minimize purchases from volatile spot markets.

The risks of getting it wrong are not abstract. Historically, Indian forecasts have missed in both directions. In some years, total energy requirement forecasts show errors approaching 19.8%, with peak demand errors up to 13.24%. Errors on the low side mean shortages and costly buys; errors on the high side mean stranded assets and unnecessary costs borne by consumers. In a reforming sector—unbundled services, evolving tariffs, growing private participation—realistic, defensible forecasts are essential to keep the grid reliable and the finances healthy.

Why precision is especially critical for Bihar

Bihar’s demand story is tightly linked to its broader socio‑economic trajectory. The study shows strong correlations between electricity demand and variables such as per‑capita availability, GSDP (state output), and population. In fact, decadal growth in both electricity demand (≈9.09%) and per‑capita availability (≈11.69%) is the highest in India, underscoring how quickly the state’s consumption base is evolving. Notably, in the COVID‑19 year (FY 2020‑21), Bihar posted positive demand and GSDP growth while many other states contracted—a sign of underlying momentum.

The economic mix is shifting toward services, with Banking & Insurance emerging as a notable growth engine (≈12.21% growth). That matters because services can have distinct load profiles—think daytime commercial use, IT/ITES hubs, and expanding digital infrastructure—compared with traditional industrial loads. Add in Bihar’s faster‑than‑national population growth and rising urbanization, and you have a recipe for sustained upward pressure on electricity consumption. The study’s takeaway is straightforward: if you want an accurate forecast for Bihar, you must reflect these structural changes, not just extend a straight line from yesterday’s kWh.

Methods

An honest forecast combines what the economy is doing (causal drivers) with what the data itself says (time‑series patterns). This study uses both.

1) Trend and univariate time‑series

Sometimes the best signal is in the series itself. Univariate methods look only at past electricity consumption to predict the future:

• Exponential smoothing gives more weight to recent observations. The study tests several flavors:
  • Simple exponential smoothing (no trend/seasonality).
  • Brown’s double smoothing (captures trend explicitly).
  • Holt’s method (separate smoothing for level and trend).
  • Holt‑Winters (additive or multiplicative) to model both trend and seasonality when patterns repeat, such as monthly peaks. Demand Forecasting Study
• ARIMA (Auto‑Regressive Integrated Moving Average) blends three ideas: autoregression (past values predict today), differencing (to remove drift and make the series stationary), and moving-average of past errors (to learn from misses). ARIMA(p, d, q) models are selected based on how many lags (p), differences (d), and past errors (q) best explain the data. Demand Forecasting Study

The study also references an Expert Modeler—software that tries many model combinations (e.g., multiple smoothing setups and ARIMA variants) and picks the best fit without hand‑tuning every knob. Demand Forecasting Study

2) Multivariate (econometric) models

When you know why demand should rise—more people, higher incomes, sectoral shifts, weather—you can encode those relationships in a regression:

ED = f(Rainfall, Population, Weather, Income/GSDP, …)

This approach estimates how sensitive electricity demand (ED) is to each driver. Because independent variables can be correlated with each other (e.g., income and urbanization move together), the study uses Variance Inflation Factor (VIF) checks to ensure the model isn’t double‑counting the same effect. Demand Forecasting Study

3) Time series, curve fitting, and hybrids

The team supplements econometrics with time‑series models (using lagged values of demand as predictors) and curve fitting (choosing mathematical functions that best fit historical points for visualization or extrapolation). In practice, hybrid modeling—combining causal regressions with time‑series patterns—often performs best, capturing both structure (economics) and signal (data patterns). The final model selection is based on goodness‑of‑fit metrics, including Pearson’s chi‑square tests, and back‑testing on held‑out years. Demand Forecasting Study

What data goes in and how the forecast is built

Good forecasts start with clean foundations. The study uses:

• Category‑wise sales, consumers, and connected load (FY 2018–FY 2022)
• T&D losses (FY 2018–FY 2022)
• GSDP (primary, secondary, tertiary) at constant prices (FY 2012–FY 2020)
• Census population data (2011 & 2021)

This is a category‑specific, district‑level approach, which is exactly what utilities need to plan feeders, substations, and procurement by segment (domestic, commercial, industrial, agriculture, street lighting, water supply, etc.). Demand Forecasting Study

The forecasting pipeline runs through eight clear steps:

1. Unrestricted sales projection: Start by estimating what each category would consume with full supply (i.e., removing curtailment effects).
2. Compile candidate drivers: P‑GSDP, S‑GSDP, T‑GSDP, population, irrigated area, number of consumers, etc.
3. Identify significant variables: Regressions isolate which drivers matter for each category.
4. Project those drivers: Each independent variable gets its own univariate time‑series forecast.
5. Build hybrid models by category: Marry causal and time‑series pieces for the best of both worlds.
6. Scenario analysis: Pessimistic, Realistic, and Optimistic variants reflect different growth assumptions.
7. T&D loss trajectory and load factor: Losses follow regulatory and historical paths; load factor is estimated from limited historical points.
8. Roll up to state‑level energy and peak demand over a 10‑year horizon (FY 2022‑23 to FY 2031‑32). Demand Forecasting Study

“Unrestricted” demand and peak vs non‑peak use

Because supply interruptions can mask true demand, the study reconstructs unrestricted sales using observed supply hours and simple, transparent assumptions:

• Peak hours are assumed to be 6 hours/day; non‑peak 18 hours/day (a 1:3 split).
• Consumption factors: consumption is 1.25× in peak vs 1.00× in non‑peak for both domestic and other categories.
• A straight‑line relation between restricted and unrestricted sales is used, with formulae that scale daily consumption by available supply minutes and the peak/non‑peak split. The idea is to answer, “What would demand have been if supply were 24×7?” Demand Forecasting Study

This normalization step is crucial. Without it, you’d systematically understate demand in areas with frequent curtailment and overstate demand where supply is already robust.

Scenario thinking: planning for a range, not a point

Forecasts are not crystal balls—they are decision tools. The study wisely provides three scenarios:

• Pessimistic: Conservative assumptions on growth and driver variables.
• Realistic: Baseline aligned to stakeholder input and recent trends.
• Optimistic: Higher growth trajectories for key drivers.

For each, the hybrid model generates category‑wise, district‑level consumption and peak demand, then aggregates the results. This allows planners to check procurement and capex resilience across a band of futures rather than anchoring to a single number. Demand Forecasting Study

What could bend the curve: EVs, rooftops, pumps, and efficiency

Electric Vehicles (EVs)
More than half of Indian states now have EV policies, and Bihar is on that list. Government fleets are shifting, charging networks are expanding, and two‑ and three‑wheelers are electrifying. EV adoption raises electricity demand—especially for evening charging—so forecasters must allocate for transport loads explicitly rather than letting them “leak” into commercial or domestic segments. Demand Forecasting Study

Rooftop solar
India’s national target calls for ~500 GW of renewables by 2031‑32, including ~50 GW of rooftop solar. Bihar’s 2017 policy targeted 1,000 MW of rooftop by FY 2021‑22, but only ~30 MW had been installed by that date (per MNRE at the time). As rooftop scales, it will offset grid energy (especially in sunny hours) and reshape the load curve, potentially lowering mid‑day demand while leaving the evening peak intact. The forecast accounts for this offset so we don’t over‑procure energy that rooftop generation will supply. Demand Forecasting Study

Solar pumps (PM‑KUSUM)
PM‑KUSUM includes three components; for demand forecasting, the study focuses on Component C—solarization of existing grid‑connected agricultural pumps—because it directly reduces grid energy draw in the agri segment. Components A and B add supply or standalone pumps, but Component C changes demand on the grid. Demand Forecasting Study

Demand‑side management (DSM) and efficiency
Efficiency programs—national missions, municipal retrofits, street lighting upgrades, and agricultural initiatives—can shave meaningful energy. BEE’s UNNATEE framework targets ~285 BU of national electric savings by FY 2030‑31; in FY 2020‑21 alone, ~240 BU of savings were estimated. The projection includes these offsets as part of a realistic net‑demand outlook. Demand Forecasting Study

Testing, choosing, and defending the model

A forecast is only as credible as its validation. Here’s how the study handles it:

• Back‑testing: Build models on a truncated history and test whether they correctly “predict” the held‑out years.
• Goodness‑of‑fit checks: Use statistics (including chi‑square) to see how closely modelled values track observed data.
• Collinearity control: Use VIF to avoid inflating coefficients by including highly correlated drivers.
• Category specificity: Because domestic, commercial, industrial, agriculture, street lighting, and water supply behave differently, each gets its own tailored model.
• Loss trajectory and load factor: Integrate realistic T&D loss improvements and load factor estimates so the energy‑to‑peak conversion is grounded. Demand Forecasting Study

Together, these steps make the output defensible—exactly what regulators, boards, and lenders look for.

What this means for planners and regulators

For procurement teams: Use the scenario band (pessimistic to optimistic) to structure long‑term contracts and decide how much flexibility to keep for medium‑term buys. Stress‑test portfolios against higher‑than‑expected EV charging or lower‑than‑expected rooftop uptake. Demand Forecasting Study

For network planners: The category‑ and district‑level granularity is your friend. Align substation upgrades, feeder strengthening, and urban network reinforcement with where growth is projected (e.g., services corridors or new industrial clusters). Factor in day‑night patterns (peak vs non‑peak) that the forecast captures. Demand Forecasting Study

For tariff design: Use the consumer‑category forecasts to refine cost‑reflective tariffs and avoid cross‑subsidies that blow up the utility’s balance sheet. The short‑term forecast component supports annual tariff setting with fewer surprises. Demand Forecasting Study

For policy makers: Track how EV, rooftop, KUSUM, and DSM schemes change the shape—not just the size—of load. Nudging charging to off‑peak hours, for example, can materially improve load factor and reduce peak procurement costs. Demand Forecasting Study

Key takeaways you can act on

• Don’t rely on straight lines. Bihar’s growth is dynamic and sector‑specific. Causal drivers (GSDP mix, population, irrigated area) plus time‑series patterns deliver better accuracy. Demand Forecasting Study
• Normalize for supply constraints. Estimating unrestricted demand avoids systematically under‑measuring fast‑growing districts. The study’s peak/non‑peak framework (6 vs 18 hours; 1.25× vs 1.0× consumption factors) makes this explicit. Demand Forecasting Study
• Model offsets and add‑ons. EVs add load; rooftop and KUSUM C subtract; DSM reduces. Your forecast should net these forces out by category and by hour. Demand Forecasting Study
• Plan with bands, not points. Pessimistic/Realistic/Optimistic scenarios let you manage risk rather than pray for a single “right” number. Demand Forecasting Study
• Document everything. When the assumptions are transparent—data sources, variable choices, VIF checks, fit statistics—regulators and lenders say “yes” faster.

Final word: a living forecast beats a perfect one

No model will capture every twist in technology, policy, or behavior. That’s okay. A living forecast—one you revisit annually with fresh data, revised policy inputs, and recent load patterns—will beat a “perfect” one‑and‑done forecast every time. Bihar’s demand is rising for understandable reasons: expanding services, rising incomes, urbanization, and electrification of transport and agriculture. By combining econometric drivers with time‑series signal, normalizing for supply constraints, and explicitly modelling rooftop/EV/DSM effects, this study offers a practical roadmap that utilities, regulators, and policy makers can use today—and improve tomorrow.