EnglishDeutsch/api/energyGitHub · iceccarelli
Grimaldi EngineeringCluster control · Frankfurt
Send a SKU list
Energy Intelligence · Cluster 1

Research frontier

Research survives only if it improves a product, creates defensible IP or produces credible scientific output. Nothing is called state of the art without a published benchmark against a classical baseline.

A benchmark record contains

datasetbaselinemetricconfidence intervalhardwarelatencyfailure casesgeneralizationablation

Evidence matrix — benchmark fields per topicreportedmissing
datasetbaselinemetricresultCIhardwarelatencyfailuresreproPINNs for AC/DC power flowPINNs for AC/DC power flow · dataset: IEEE 9-busPINNs for AC/DC power flow · baseline: Newton-Raphson (pandapower)PINNs for AC/DC power flow · metric: Voltage-angle RMSEPINNs for AC/DC power flow · result: DC 0.0124° · AC 0.0035° (owner-stated)PINNs for AC/DC power flow · CI: not reportedPINNs for AC/DC power flow · hardware: not reportedPINNs for AC/DC power flow · latency: not reportedPINNs for AC/DC power flow · failures: not reportedPINNs for AC/DC power flow · repro: falseMILP dispatch vs learned dispatchMILP dispatch vs learned dispatch · dataset: no benchmarkMILP dispatch vs learned dispatch · baseline: no benchmarkMILP dispatch vs learned dispatch · metric: no benchmarkMILP dispatch vs learned dispatch · result: no benchmarkMILP dispatch vs learned dispatch · CI: no benchmarkMILP dispatch vs learned dispatch · hardware: no benchmarkMILP dispatch vs learned dispatch · latency: no benchmarkMILP dispatch vs learned dispatch · failures: no benchmarkMILP dispatch vs learned dispatch · repro: no benchmarkProbabilistic forecasting (load, PV, price)Probabilistic forecasting (load, PV, price) · dataset: no benchmarkProbabilistic forecasting (load, PV, price) · baseline: no benchmarkProbabilistic forecasting (load, PV, price) · metric: no benchmarkProbabilistic forecasting (load, PV, price) · result: no benchmarkProbabilistic forecasting (load, PV, price) · CI: no benchmarkProbabilistic forecasting (load, PV, price) · hardware: no benchmarkProbabilistic forecasting (load, PV, price) · latency: no benchmarkProbabilistic forecasting (load, PV, price) · failures: no benchmarkProbabilistic forecasting (load, PV, price) · repro: no benchmarkInertia / RoCoF with grid-forming invertersInertia / RoCoF with grid-forming inverters · dataset: Book appendix figures (chapter 1)Inertia / RoCoF with grid-forming inverters · baseline: Swing equationInertia / RoCoF with grid-forming inverters · metric: Test pass against stated figuresInertia / RoCoF with grid-forming inverters · result: 48 tests passInertia / RoCoF with grid-forming inverters · CI: n/aInertia / RoCoF with grid-forming inverters · hardware: n/aInertia / RoCoF with grid-forming inverters · latency: n/aInertia / RoCoF with grid-forming inverters · failures: n/aInertia / RoCoF with grid-forming inverters · repro: trueSafe / constrained RL for dispatchSafe / constrained RL for dispatch · dataset: no benchmarkSafe / constrained RL for dispatch · baseline: no benchmarkSafe / constrained RL for dispatch · metric: no benchmarkSafe / constrained RL for dispatch · result: no benchmarkSafe / constrained RL for dispatch · CI: no benchmarkSafe / constrained RL for dispatch · hardware: no benchmarkSafe / constrained RL for dispatch · latency: no benchmarkSafe / constrained RL for dispatch · failures: no benchmarkSafe / constrained RL for dispatch · repro: no benchmarkGNNs / neural operators for grid stateGNNs / neural operators for grid state · dataset: no benchmarkGNNs / neural operators for grid state · baseline: no benchmarkGNNs / neural operators for grid state · metric: no benchmarkGNNs / neural operators for grid state · result: no benchmarkGNNs / neural operators for grid state · CI: no benchmarkGNNs / neural operators for grid state · hardware: no benchmarkGNNs / neural operators for grid state · latency: no benchmarkGNNs / neural operators for grid state · failures: no benchmarkGNNs / neural operators for grid state · repro: no benchmarkRuntime assurance & agent verificationRuntime assurance & agent verification · dataset: no benchmarkRuntime assurance & agent verification · baseline: no benchmarkRuntime assurance & agent verification · metric: no benchmarkRuntime assurance & agent verification · result: no benchmarkRuntime assurance & agent verification · CI: no benchmarkRuntime assurance & agent verification · hardware: no benchmarkRuntime assurance & agent verification · latency: no benchmarkRuntime assurance & agent verification · failures: no benchmarkRuntime assurance & agent verification · repro: no benchmarkUncertainty quantification & distribution s…Uncertainty quantification & distribution shift · dataset: no benchmarkUncertainty quantification & distribution shift · baseline: no benchmarkUncertainty quantification & distribution shift · metric: no benchmarkUncertainty quantification & distribution shift · result: no benchmarkUncertainty quantification & distribution shift · CI: no benchmarkUncertainty quantification & distribution shift · hardware: no benchmarkUncertainty quantification & distribution shift · latency: no benchmarkUncertainty quantification & distribution shift · failures: no benchmarkUncertainty quantification & distribution shift · repro: no benchmark

A full row is the condition for the words "state of the art". Today no row is full.

Topics

TopicProduct useModuleRepositoryStateNext
PINNs for AC/DC power flowSurrogate power flow for fast candidate screeningphysicsphysics-informedpartialPublish the full record with CI, latency vs Newton, and behaviour on 39-bus / 118-bus (generalization).
MILP dispatch vs learned dispatchWedge A — BESS / community dispatchoptimizeGridOSunbenchmarkedBenchmark MILP against the customer's rule-based schedule on real prices; learned policies only after MILP is deployed.
Probabilistic forecasting (load, PV, price)Forecast inputs to dispatch; forecast-error KPIforecastGridOS (MPC loop)unbenchmarkedPersistence and seasonal-naive baselines first; pinball loss on open data (e.g. ENTSO-E Transparency, Open Power System Data).
Inertia / RoCoF with grid-forming invertersPhysics constraint checks; credibilityphysicsRenewables_Migration_*_Proof_EnginepartialSeparate book-consistency tests from physical validation against a public dynamics case.
Safe / constrained RL for dispatchNot before MILP is in productionoptimize—watchWatch only. Revisit when there is customer data and a deployed baseline.
GNNs / neural operators for grid stateState estimation with sparse measurementstwin—watchWatch only.
Runtime assurance & agent verificationNeuralBridge designneuralbridgeNeuralBridge (unlocated) · mcp-foundryunbenchmarkedSpecify the simplex-architecture style monitor; formalize the envelope check; property tests before any code.
Uncertainty quantification & distribution shiftForecast intervals; refusal to act under shiftforecast · planning—watchConformal intervals on the forecasting module once it has a backtest.
Benchmark record: PINNs for AC/DC power flow
datasetbaselinemetricresultCIhardwarelatencyfailure casesreproducible
IEEE 9-busNewton-Raphson (pandapower)Voltage-angle RMSEDC 0.0124° · AC 0.0035° (owner-stated)not reportednot reportednot reportednot reportedfalse
Benchmark record: Inertia / RoCoF with grid-forming inverters
datasetbaselinemetricresultCIhardwarelatencyfailure casesreproducible
Book appendix figures (chapter 1)Swing equationTest pass against stated figures48 tests passn/an/an/an/atrue

Frontier watched

OPFAC/DC power flowMPCGNNsneural operatorsPINNsdifferentiable simulationsafe RLconstrained RLuncertainty quantificationdistribution shiftcausal inferenceprobabilistic forecastingdigital twinsmulti-agent optimizationformal methodsruntime assuranceagent verification

As of 2026-09-05Machine-readable: /api/energy/researchAll resources: /api/energy/index

Written and maintained by Vincenzo Ceccarelli Grimaldi, Frankfurt am Main.