Top-down · System view
Economic activity, population, weather, and technology adoption are often best observed at the system or region level. Top-down forecasting trains native load models at that level, layers the load modifiers that apply at scale, and produces an 8760-hour probabilistic scenario for the whole system before anything is allocated.
Top-down · Allocation
The system scenario is allocated to planning areas, substations, transformers, and feeders using explicit rules: historical proportion, forecast proportion, or spatial allocation driven by GIS, customer, and growth data. Each rule is a named, versioned choice — never an implicit average.
Top-down · Coherence
After allocation, the children sum back to the parent within a defined tolerance, at every hour. When a downstream asset has its own forecast, the platform reports where the two disagree instead of overwriting either.
Top-down · When to use
Top-down is the natural view for system operators and transmission planners, for long-horizon scenarios where local history is thin, and for any study that starts from a policy or growth assumption defined system-wide.
Bottom-up · Local models
Every feeder, transformer, and substation with measured load gets its own native load model, trained on its own history, its local weather, and its local growth context. Assets with similar behaviour can be clustered and trained together so thin histories still produce stable models.
Bottom-up · Local modifiers
EV charging, rooftop PV, heat pumps, and known large loads are applied at the feeder level, allocated from regional adoption outlooks through GIS, customer, and land-use data. A neighbourhood with early EV uptake looks different from one without — because it is.
Bottom-up · Roll-up
Local scenarios are summed up the hierarchy — feeder to substation to service area to system — hour by hour, preserving the diversity and coincidence between assets rather than adding peaks.
Bottom-up · Comparison
A bottom-up total and a top-down system forecast rarely match. The platform puts them side by side and traces the gap to its cause — a growth assumption, a modifier, a data gap — rather than forcing one to win.
