Energy research
MOSES - EMS for residential nanogrids
MOSES case study: a residential microgrid with photovoltaics, storage, programmable loads and an interpretable EMS for real-time energy-flow management.
Summary
MOSES turns storage into an energy-management problem.
MOSES (MOdular Smart Energy System) studies modular storage systems suited to fast charge and discharge cycles, for telecom stations, electric-vehicle charging and residential photovoltaic plants. In the residential scenario, photovoltaic energy, battery storage, programmable loads and the grid have to be coordinated by an Energy Management System: not just storage, but real-time decisions on energy flows.
- Local system with PV, ESS, programmable loads and grid connection.
- EMS designed for Demand Response, self-consumption and lower grid stress.
- Three application scenarios: telecom, fast electric-vehicle charging, residential photovoltaics with storage.
Problem
Coordinating intermittent generation, domestic demand and storage.
The growth of distributed renewables, electric vehicles and flexible loads makes grid exchange more variable. A nanogrid has to decide when to use photovoltaic production, when to charge or discharge the ESS, when to draw energy from the grid, and when to delay appliances such as washing machines and dishwashers.
- Intermittent PV production and variable domestic consumption.
- Non-interruptible loads that must be supplied in real time.
- Programmable loads with start windows and user constraints.
- Need for an interpretable decision strategy.
Approach
Formalize energy flows and train an interpretable EMS.
The work starts from the electrical topology of the residential microgrid (rMG) and from a notation for energy flows. This becomes the basis for a data model with training, validation and test sets, then used to synthesize a fuzzy EMS that can be compared with benchmark solutions.
- Energy flows described by node, branch and connected system.
- Residential and PV time series aligned into 15-minute slots.
- Training, validation and test sets separated for modelling and verification.
- Grey-box models selected to keep decisions readable.
System
Three fuzzy systems for grid exchange, battery and programmable loads.
The modelled EMS combines an ANFIS for grid exchange, ESS management and state of charge, plus two Mamdani FIS models to decide the start of programmable loads. Inputs include SoC, load, PV generation, previous grid exchange and pending appliance cycles.
- Central ANFIS for grid connection and ESS management.
- Two FIS models for washing-machine and dishwasher scheduling.
- Input normalization to 0-1 and filtering of acquired signals.
- Action plan tested against the mathematical microgrid model.
Results
Simulations close to benchmarks, with clear limits.
On the test set, the generated EMS models were compared with optimal benchmarks in two scenarios. The distance from the optimal solution is about 10 percentage points in one scenario and about 15 in the other; programmable-load management slightly improves the second scenario and is negligible in the first.
- Results evaluated on normalized cost functions, not economic ROI.
- ANFIS performance considered effective in the simulation context.
- FIS contribution on flexible loads positive but limited.
- Constraints set conservatively and results not to be overclaimed.
Technology
Microgrid, MCIC, data store and signals for energy control.
The system covers the microgrid topology, the modular MCIC converter, the models of the photovoltaic plant, battery and loads, and data management. The database serves both to train the EMS and prediction models and for diagnostics, monitoring, maintenance and retraining in new scenarios.
- Modular MCIC to interface DC/AC sources and DC/AC loads.
- PV system and battery modelled in Simulink with physical constraints.
- Historical series for power converters, ESS and EMS input/output.
- Retraining path when seasons, tariffs, loads or battery sizing change.
Reuse
What carries over to the next projects.
MOSES gave Adeo experience in storage, energy-flow management and interpretable decision models, skills that serve its later projects on energy and electric mobility.
- Separate training, validation and test data to really measure an EMS.
- Prefer interpretable models when decisions affect the grid and the users.
- State the limits: simulated results, on normalised cost functions.
- Keep a historical database to retrain the system when conditions change.
CTA
Do you have a plant or an energy network to manage better?
Adeo can start from the MOSES experience to discuss nanogrids, storage control, datasets and interpretable decision systems.
Discuss energy research