Source - Energiforsk 2026-1168 AI-modeller Prognostisering Efterfrågan El (2026)

Full citation: Sütfeld, L. & Popoff, A. (RISE Research Institutes of Sweden). AI-modeller för prognostisering av efterfrågan på el. Energiforsk Report 2026:1168. Stockholm: Energiforsk AB, 2026. Program: Elnätens hållbara teknikutveckling och digitalisering. Sub-report 1 of the PREDATOR project (Predicting Energy demand and forecasting changes in local grid loads).

Open access: Published by Energiforsk. RISE Research Institutes of Sweden is Sweden’s largest research and technology organization (statligt forskningsinstitut).

Summary

This is the stakeholder needs analysis report for the PREDATOR project — a participatory investigation into what Swedish DSOs actually need from AI and data-driven load forecasting tools. Through four workshops with eight Swedish DSOs (November 2023 – April 2024), the report documents the current state of DSO demand forecasting capabilities, identifies key gaps and requirements, and proposes an “analyze-aggregate-extrapolate” framework as a conceptual architecture for a holistic forecasting solution.

The report’s primary finding is stark: many Swedish DSOs lack internal data-driven demand/load forecasting models entirely, and those that have some capability do not have integrated end-to-end solutions. DSOs want holistic tools — not partial components — and the barriers to adoption are as much organizational (internal data management capacity) as technical.

Sub-report 2 (planned) will expand the framework with survey data (Eurobarometer, ESS, SCB) and economic indicators for the extrapolation step. A larger follow-up project, DESGRID, is planned based on the PREDATOR findings.

Key claims

Workshop design

• 4 workshops: November 2023, January 2024, March 2024, April 2024
• 8 participating DSOs:
  • Ellevio (large; ~1M customers; national coverage)
  • Göteborg Energi (large; Gothenburg)
  • Mölndal Energi (medium; Mölndal, co-operator of Effekthandel Väst)
  • Trollhättan Energi (medium; Trollhättan)
  • Jönköping Energi (medium; Jönköping, participated in PREDATOR workshops and SWITCH)
  • Umeå Energi (medium; Umeå)
  • C+ Energi (small-medium)
  • Karlstads Kommun (municipal; Karlstad)

This sample covers DSOs from three of Sweden’s four electricity areas (SE3/SE4 dominated; Umeå suggests SE2) and spans from very large to small-municipal scale — a reasonably representative cross-section of Swedish DSO diversity.

Core finding: DSOs want holistic solutions

The primary insight from workshops: DSOs request complete, integrated forecasting solutions, not partial or specialized tools. Specifically:

• DSOs do not want a tool that only processes one data type, covers one load category, or handles one step in the forecasting workflow
• DSOs need help with the entire chain: from data management through model building through interpretation and planning integration
• Partial tools create integration burden that DSOs — especially smaller ones — cannot absorb
• This finding directly mirrors Source - Energiforsk 2026-1157 Nationell Metod Effekt och Kapacitetsprognoser (2026)‘s finding that 32% of DSOs lack documented methodology: the demand is not for more sophisticated components but for more accessible complete solutions

Underlying cause: Many DSOs lack the internal staff with both domain knowledge (power systems) and data science capability to build and operate data-driven models. The organizational barrier is at least as significant as the technical barrier.

Current forecasting gap

• Many of the 8 participating DSOs have no internal data-driven demand forecasting model — planning relies on expert judgment, rule-of-thumb extrapolation, and the Energiforsk lathund (simple template-based tool)
• Even DSOs that have some quantitative forecasting do not have integrated models that combine:
  • Current DER inventory (EVs, solar PV, heat pumps, batteries)
  • Behavioral and socioeconomic drivers
  • Future growth trajectories

This gap is significant given that EV penetration and distributed solar installation are now the dominant uncertainty in short-to-medium-term distribution load forecasting.

Proposed “analyze-aggregate-extrapolate” framework

The report proposes a three-step conceptual architecture for holistic DSO demand forecasting:

Step 1: ANALYZE
  Identify current DER inventory from available data
  → What EVs, solar panels, heat pumps, batteries are 
    connected at each substation/feeder today?
  → Sources: AMI smart meter data, connection register,
    survey data, administrative data (SCB vehicle register)
         │
         ▼
Step 2: AGGREGATE
  Build current-state load profiles per grid segment
  → Combine DER inventory with behavioral profiles
    (charging patterns, solar output profiles, heat pump cycles)
  → Produce realistic substation-level demand curves
  → Validate against actual AMI measurements
         │
         ▼
Step 3: EXTRAPOLATE
  Project DER inventory and load profiles into the future
  → Apply growth scenarios for EV adoption, solar installation,
    heat pump uptake
  → Incorporate macroeconomic and sociodemographic indicators
    (population, income, housing type, industrial activity)
  → Produce scenario-based load forecasts per grid area

Step 3 data sources (planned for Sub-report 2): Eurobarometer consumer surveys; ESS (European Social Survey); SCB (Statistics Sweden) — population, income, housing; economic indicators. These are used to build statistical models of DER adoption rates by household/customer type that can be extrapolated geographically.

DESGRID — planned follow-up project

Based on PREDATOR Sub-report 1 findings, a larger follow-up project DESGRID is planned. DESGRID would implement and validate the analyze-aggregate-extrapolate framework as a concrete tool for Swedish DSOs. Scope and timeline at time of this report’s publication: under development.

The Vinnova Advanced Digitalization funding application for an earlier version of this work failed due to summer scheduling — applications submitted in summer had lower review committee availability, reducing the chances of success. This is a practical note on the Swedish research funding landscape (Vinnova’s call cycle timing matters for proposal success rates).

Relationship to other forecasting initiatives

This project addresses a different aspect of the forecasting challenge than Source - Energiforsk 2026-1157 Nationell Metod Effekt och Kapacitetsprognoser (2026):

Both projects identify the same underlying problem (inadequate DSO forecasting capability) and are complementary solutions at different levels of the system.

The “identify current EV/solar/HP inventory” step in PREDATOR’s “analyze” phase directly corresponds to the RISE AMI classification tool already mentioned in Flexibility Need Assessment › Practical tools for flexibility need quantification — confirming that RISE was developing this capability simultaneously in multiple research programs.

Relevance to the wiki

This source directly informs or strengthens:

• Distribution System Operator — documents the practical AI/data readiness gap at Swedish DSOs; confirms that the lack of data-driven forecasting is a known sector-wide issue, not just a few outliers; the analyze-aggregate-extrapolate framework as a structured conceptual approach to closing the gap
• Flexibility Need Assessment — the “analyze” step (identify current DER inventory) is a prerequisite for accurate flexibility need quantification; PREDATOR addresses the same data foundation gap that the Endre probabilistic tool and RISE AMI classification tool address
• Distribution Network Development Plan — better DER-level load forecasting directly feeds DNDP scenario development (Pillar 1 of the three-pillar planning process); PREDATOR Sub-report 2 data sources (ESS, SCB, Eurobarometer) could provide inputs to DNDP behavioral scenario development
• Aggregation — the “aggregate” step (combining DER inventory into feeder-level profiles) is exactly what Virtual Power Plant and aggregation platforms do from a commercial standpoint; shared research interest in aggregation methods

Internal links to other Energiforsk 2024–2026 program sources:

• Source - Energiforsk 2024-1043 DNDP Analys och Flexibilitet (2024) — the automation index ranks “current DER inventory identification” as high priority; PREDATOR’s “analyze” step is the automation of that sub-task
• Source - Energiforsk 2026-1157 Nationell Metod Effekt och Kapacitetsprognoser (2026) — complementary approach (top-down vs bottom-up); same problem, different level of resolution; both identify that the sector’s forecasting capability is inadequate
• Source - Energiforsk 2026-1151 Effektauktioner med Värmepumpar (2026) — both require accurate per-unit DER profiles; the SVM normalization used in 2026-1151 is a related methodology to PREDATOR’s extrapolation step

Data gaps

• Sub-report 1 is a needs analysis only — no implemented model, validated result, or DSO test deployment
• DESGRID is planned but not funded as of this report’s publication; outcome uncertain
• The 8 participating DSOs are a self-selected group interested in the topic — may not represent the ~120 smaller Swedish DSOs with lower digital maturity
• The Vinnova application failure (summer scheduling) introduces a research continuity risk — if DESGRID funding is not secured, the analyze-aggregate-extrapolate framework remains conceptual