DistrictHeat+ is a research-driven software platform developed at the SDU Center for Energy Informatics for operational district-heating intelligence. The platform is designed to help utilities and engineers work more effectively with fragmented district-heating data by bringing asset information, maps, uploaded files, analytical workflows, and AI-assisted interpretation into one unified local environment. At the core of DistrictHeat+ is LLM EnergyAdvisor, which enables multimodal and evidence-linked interaction with district-heating data while keeping the entire system fully local and under organizational control, without relying on external LLM APIs.
Key features:
DistrictHeat+ combines several interconnected modules that support both operational overview and engineering analysis:
- Dashboard: Provides a map-based network overview for fast inspection of the district-heating system and its spatial asset context.
- Heat Assets: Supports asset-level exploration of pipe-network information, filtering, and visual inspection of district-heating infrastructure.
- AI Assistant: Delivers conversational support for multimodal district-heating analysis through LLM EnergyAdvisor, including questions over documents, alarms, images, engineering files, and operational data.
- Admin panel: Supports user management, document and knowledge management, and configuration of the platform’s local AI operation modes.
The platform also supports a broad set of public-facing AI capabilities, including engineering analytics and plotting, geospatial pipe-network visualization, retrieval across uploaded engineering files and persistent domain knowledge, image and screenshot understanding, technical summarization, and report-oriented output generation. Its current workflow includes both Advanced Intelligence Mode for deeper multimodal reasoning and Fast Assistance Mode for quicker responses in practical engineering tasks.
From the current implementation and example workflows, DistrictHeat+ supports use cases such as alarm summarization and troubleshooting, pipe-network plotting and spatial asset exploration, uploaded CSV and log-file trend analysis, screenshot- or image-based task understanding, and vulnerability-oriented asset calculations. The platform also exposes a visible reasoning or Thinking Process view, which makes the agentic workflow more transparent for operators during complex tasks.
Theories and background behind the tool:
DistrictHeat+ is grounded in recent advances in multimodal large language models, retrieval-augmented generation, and agentic AI design for engineering decision support. Its central concept is that district-heating operators often work across heterogeneous and fragmented information sources, including GIS layers, pipe data, operational records, manuals, screenshots, and engineering documents. Rather than treating these sources separately, the platform turns them into a searchable and analyzable operational workspace.
A key methodological foundation is the ReAct-style agentic design, where the assistant does not only generate text, but iteratively reasons, retrieves, analyzes, and executes the next appropriate step before responding. In the underlying architecture, the system combines multimodal ingestion, hybrid retrieval, image understanding, and executable analytics so that the assistant can inspect retrieved evidence, analyze visual inputs, and perform computational tasks when needed. This design is especially relevant in district heating, where useful answers often depend on combining documents, maps, files, and calculations rather than relying on text generation alone.
The retrieval layer draws on hybrid search principles, combining semantic retrieval with exact and keyword-aware matching so that both engineering terminology and specific identifiers can be handled robustly. The platform also reflects explainability principles by linking answers to retrieved evidence and by exposing a visible reasoning process in the user interface, which supports transparency for operational users.
How DistrictHeat+ advances the state of the art:
DistrictHeat+ advances the state of the art by moving beyond isolated dashboards, standalone GIS tools, document viewers, or generic AI chat interfaces. Instead, it provides a single district-heating-oriented platform in which multimodal inspection, retrieval, analytics, mapping, and operator support are orchestrated through one AI-centered workflow. This is particularly important in district-heating practice, where engineers often need to move repeatedly between maps, tables, uploaded files, screenshots, technical documents, and ad hoc calculations.
A second distinguishing contribution is its fully local deployment model. The platform is designed so that sensitive infrastructure data remain inside the deployment environment, which is highly relevant for critical energy infrastructure where privacy, governance, and operational control are essential. Rather than depending on externally hosted AI services, DistrictHeat+ is built as a local full-stack system with locally hosted multimodal intelligence.
A third contribution is the role of LLM EnergyAdvisor as the core engine of the platform. Through its agentic workflow, multimodal reasoning, visible thinking process, and evidence-grounded responses, the system supports a more practical transition from fragmented district-heating data toward operator-oriented decision support. In this sense, DistrictHeat+ is not only a chat interface, but a district-heating asset intelligence platform that enables inspection, summarization, interpretation, and engineering action within one coherent environment.
Related publications:
LLM EnergyAdvisor: An Explainable ReAct Agent for Local Multimodal District Heating Interpretation
Principal investigator: Hamid Reza Shaker
Development team: Hamid Mirshekali, Mohammad Reza Shadi
Tool link: Book a Demo (send email to hrsh@mmmi.sdu.dk)