Enabling Privacy-Preserving Natural Language Access to Industrial Smart Meter Data
Organizations: Valencian Research Institute for Artificial Intelligence (VRAIN), Universitat Politècnica de València, Valencia, Spain · Know Center Research GmbH, Graz, Austria
Abstract
Energy management officers and grid operators increasingly rely on smart meter data for anomaly detection, performance monitoring, and strategic decision-making. However, extracting actionable insights from large-scale, heterogeneous telemetry databases requires advanced SQL and data engineering expertise, creating a critical bottleneck for non-technical personnel. Furthermore, the strict data privacy regulations and security constraints inherent to critical energy infrastructure often prohibit the routing of sensitive consumer data through cloud-based Large Language Models (LLMs). To address these challenges, we present OpenMeter, a schema-agnostic Multi-Agent System (MAS) designed to translate natural language queries into executable code and human-readable insights. This architecture lowers the technical barrier for day-to-day utility operations by empowering non-expert personnel to interact with complex databases through intuitive natural language dialogue, transforming raw telemetry into a highly valuable tool for routine grid workflows and strategic decisions. Multi-scenario ablation studies over this environment demonstrate that full multi-agent orchestration shows its largest observed accuracy gains on resource-constrained local hardware (+6.1 pp EX-R over the strongest single-agent configuration); at frontier scale, individual components retain value but should be enabled selectively, indicating that the cost of full orchestration is justified specifically where privacy demands local execution.
Figures & tables
| Setup 1 (Local) | Setup 2 (Cloud) | |
| Model (all agents) | Qwen3.5 | Qwen3.5 |
| Parameters | 9B | 397B-A17B MoE |
| GPU | 1 RTX A5000 (24 GB) | Ollama Cloud |
| System RAM | 25 GB | — |
| vCPUs | 9 | — |
| Inference backend | Ollama (local) | Ollama Cloud |
| Dataset | Meters | Rows | Res. | Schema | Key Measurements |
| EMSify | 19 | 3.5M | 15m | Long/EAV | , , PF, THD, (28 types) |
| LP3 | 74 | 3.0M | 10m | Wide | (voltage only) |
| LPR | 39 | 1.0M | 15m | Wide | , load profile registers |
| Tier 1 (Local, 9 B) | Tier 2 (Frontier, 397 B) | ||||||||||||||
| Execution | Structural | Runtime | Execution | Structural | Runtime | ||||||||||
| Arch. | Scen. | EX-S | EX-P | EX-R | EM | CM | ESR | VER | EX-S | EX-P | EX-R | EM | CM | ESR | VER |
| Single | S-Base | 24.5 ±3.2 | 25.3 ±2.3 | 32.7 ±4.1 | 3.7 ±1.7 | 0.273 ±0.018 | 53.5 ±3.0 | 27.6 ±4.2 | 36.7 ±1.4 | 37.1 ±2.2 | 54.7 ±1.7 | 15.5 ±2.3 | 0.422 ±0.010 | 98.8 ±1.1 | 90.3 ±2.9 |
| S+SC | 26.5 ±4.8 | 28.2 ±4.7 | 36.3 ±7.8 | 10.6 ±1.7 | 0.309 ±0.020 | 68.6 ±16.8 | 49.1 ±7.1 | 35.5 ±4.2 | 35.5 ±4.2 | 53.1 ±4.8 | 12.7 ±2.7 | 0.403 ±0.021 | 99.6 ±0.9 | 94.5 ±3.1 | |
| Multi | M-Base | 28.2 ±2.7 | 29.8 ±2.3 | 35.5 ±2.3 | 15.1 ±1.8 | 0.335 ±0.019 | 57.6 ±3.9 | 28.3 ±9.9 | 34.7 ±3.8 | 35.5 ±3.7 | 50.6 ±3.7 | 11.0 ±1.8 | 0.362 ±0.045 | 95.1 ±4.0 | 88.3 ±5.2 |
| M+SC | 28.6 ±2.9 | 30.2 ±2.7 | 36.7 ±2.5 | 15.9 ±3.9 | 0.359 ±0.032 | 77.6 ±5.2 | 56.6 ±7.2 | 34.7 ±2.5 | 35.1 ±2.7 | 49.0 ±2.0 | 10.6 ±3.9 | 0.365 ±0.024 | 99.6 ±0.9 | 96.6 ±3.4 | |
| Tier 1 (Local, 9 B) | Tier 2 (Frontier, 397 B) | |||||
| Arch. | Scen. | Eval | Rate | Rate | ||
| Single | S-Base | LLM | 0/8 | 0.0% | 3/19 | 15.8% |
| S+SC | LLM | 1/20 | 5.0% | 7/24 | 29.2% | |
| HumEv | 6/20 | 30.0% | 9/24 | 37.5% | ||
| Multi | M-Base | LLM | 3/9 | 33.3% | 3/22 | 13.6% |
| M+SC | LLM | 2/17 | 11.8% | 4/24 | 16.7% | |
| Tier 1 (Local, 9 B) | Tier 2 (Frontier, 397 B) | |||||||||||
| Interpretation | Visualization | Interpretation | Visualization | |||||||||
| Arch. | Scen. | Eval | Grd | Rel | Coh | Align | Pres | Grd | Rel | Coh | Align | Pres |
| Single | S-Base | LLM | 3.53 | 4.45 | 4.56 | 3.30 | 3.02 | 4.03 | 4.65 | 4.86 | 3.87 | 3.41 |
| S+SC | LLM | 3.67 | 4.54 | 4.72 | 3.47 | 3.05 | 4.14 | 4.68 | 4.93 | 4.05 | 3.40 | |
| HumEv | 4.41 | 3.80 | 4.30 | 4.15 | 3.27 | 4.57 | 3.91 | 4.14 | 3.27 | 3.00 | ||
| Multi | M-Base | LLM | 4.51 | 4.66 | 4.74 | 3.67 | 3.26 | 4.64 | 4.80 | 4.91 | 3.95 | 3.67 |
| Tier 1 (Local, 9 B) | Tier 2 (Frontier, 397 B) | ||||||||
| Arch. | Scen. | Thpt | Tok | Tok | Thpt | Tok | Tok | ||
| Single | S-Base | 22.5 | 57.1 | 27 796 | 1 280 | 27.6 | 73.5 | 30 315 | 1 965 |
| S+SC | 31.9 | 60.2 | 34 340 | 1 914 | 28.2 | 72.4 | 30 332 | 1 926 | |
| Multi | M-Base | 88.5 | 59.4 | 8 425 | 5 088 | 132.9 | 54.4 | 8 911 | 6 837 |
| M+SC | 135.2 | 66.1 | 11 953 | 8 424 | 128.7 | 61.5 | 8 956 | 7 416 | |
| M+KB | 144.4 | 46.3 | 49 780 | 6 494 | 474.0 | 20.7 | 49 289 | 10 208 | |
Data Analysis Agents
Data analysis agents use language models and tools to explore datasets, run analyses, and produce answers or reports grounded in evidence.
See all 117 papers