ALLSHIFT

Umbrella repository for Team SHIFT — an AI-driven hydrogen microgrid energy-management system for hospitals.

ALLSHIFT is the central hub tying together the team's models, data, planning documents, and repositories. It links the four clusters — Energy Management, Simulations, Business & Economics, and AI & Control Systems — under one roof.

→ Full documentation index

Mission

Accelerate the transition to resilient, low-carbon energy systems through AI-driven energy management. We develop intelligent control for critical infrastructure that optimizes renewables, storage, hydrogen technologies, and grid interaction — improving reliability, reducing emissions, and increasing efficiency where uninterrupted power matters most.

Founded in 2021 at the TU/e Innovation Space and Neuron (Eindhoven). Case studies: Beth Israel Deaconess Medical Center (BIDMC) and UC San Diego (UCSD).

The challenge

Hospitals need reliable, uninterrupted power while the grid grows more congested, less predictable, and more weather-dependent. With the Dutch "balancing system" being phased out from 2025, energy returned to the grid is only partially compensated. SHIFT answers with an AI-driven hydrogen microgrid that forecasts hospital demand, optimizes solar and battery use, and controls electrolyzers and fuel cells so every component runs at its full potential.

System architecture

Forecasters ──► Simulator (Gym-style env) ──► RL / rule-based controller
 (PV, demand,      wraps physical models       (dispatch policy)
  price)           PV · battery · PEM             │
      ▲            electrolyser · PEM fuel         ▼
      │            cell · H2 tank · grid       KPIs: unmet critical load ≈ 0,
   weather/                                    cost (€), CO₂, curtailment
   market data                                     │
                                                    ▼
                                          Business & Economics analysis

Dispatch priority cascade: PV → battery → hydrogen (fuel cell discharge / electrolyser charge) → external grid. Critical hospital load is always prioritised — "critical load is sacred".

Repository layout

ALLSHIFT/
├── docs/                    ← all current documentation, in markdown
│   ├── 01-project/            vision, work packages, efficiency targets
│   ├── 02-specifications/     simulator contract, RL state/action/reward, forecasting
│   ├── 03-energy-management/  BIDMC & UCSD case studies, datasets, equipment
│   ├── 04-simulations/        Simulink models, controller, PEM models, PowerWorld
│   ├── 05-business-economics/ cost parameter register, SCBA methodology
│   ├── 06-data/               dataset inventory
│   ├── 07-team-and-operations/
│   └── _originals/            the .docx / .pdf sources for root-level reports
├── superseded/              ← outdated docs, each with a documented reason
├── Shift Matlab Drive/      ← MATLAB/Simulink models and raw data (mirrors MATLAB Drive)
├── website/                 ← public site (Astro + Tailwind, Cloudflare Pages)
├── pre_model_01/            ← modelling sandbox
└── .github/                 ← organization profile

Why documents appear twice

Anything that arrived as Word, PDF, Excel, or a MATLAB Live Script has a markdown report in docs/ — readable on the web, diffable in git, greppable. The original binary stays alongside. Each report records its source path and the file's MD5, so you can tell whether the original has moved on since the report was written.

Clusters

• Energy Management — hospital load characterisation, PV/solar data, datasets and manuals.
• Simulations — MATLAB/Simulink physical models: PEM electrolyser, PEM fuel cell, hydrogen storage, battery, gravity storage; the dispatch controller.
• Business & Economics — cost and CO₂ accounting, viability analysis.
• AI & Control Systems — forecasting stack and the RL/rule-based control policy.

Key documents

Simulator I/O Interface	The typed contract between the simulator and everything that talks to it — conventions, static config, per-timestep inputs and outputs, KPIs, priority rules
Forecasting Requirements	Specs for three forecasters — PV generation, hospital demand, day-ahead price — with model stacks, metrics, and targets
RL State · Action · Reward	The reinforcement-learning problem definition
Simulink Model Inventory	Which model is the live one, and how the nine framework copies relate
Dataset Inventory	Every data file — provenance, size, and licence obligations
Cost Analysis	48-parameter economic register with per-row confidence status

Related repositories

Repo	Purpose
.github	Organization profile — mission, vision, public face
website	Public site (Astro + Tailwind, Cloudflare Pages)
pre_model_01	Modeling sandbox — early model approaches
ALLSHIFT	This repo — project hub and documentation index

website/ and pre_model_01/ are also vendored into this repository as directories. The GitHub links above are their standalone homes; this repo's own remote is git.teamshiftenergy.com/pepe/ALLSHIFT.

Status

Early development. Physical models and planning specs are in place; the simulator environment, forecasters, and control policy are being built out.

The honest picture, as of the July 2026 documentation audit:

• Specifications are the most mature layer — but not yet mutually consistent (two variable naming conventions, action spaces of 5 vs 16, reward weights undefined).
• Physical models exist but the PEM electrolyser and fuel cell are still at MathWorks demo parameters; no SHIFT-specific hydrogen sizing has been applied.
• The dispatch controller is a draft that does not yet compile.
• Data covers PV and hospital load well (via simulated proxies), but has no measured hospital load, no electricity price series, and no carbon intensity series.

Every one of these is tracked in Open questions across the project.

Contact

info@shifthydrogen.nl