Electrifying a fleet turns refuelling into an energy-management problem: dozens or hundreds of vehicles must be full at departure, but the grid connection is limited, tariffs vary by the quarter hour and a single peak can set the demand charge for a year. Smart charging combines forecasts of each vehicle's energy need with a solver that decides when and how fast each one charges, lowering energy cost and avoiding costly grid upgrades while protecting the hard constraint: ready on time.
Problem & value
- Today: many depots use "plug in and charge at full power" or static timetables. This creates evening peaks, trips connection capacity, and wastes cheap or PV-rich hours. Dispatchers often assign vehicles to chargers by hand and discover low charge only at departure.
- Why distinct from stationary storage (OPT-011): the "battery" here is mobile and has a duty schedule. Charging couples energy optimisation with route planning (OPT-001), telematics and a hard readiness deadline per vehicle.
- AI value: (1) predict the energy each vehicle will need (route, load, weather, driver), (2) solve a constrained schedule under connection, tariff and charger limits, (3) re-plan in real time when vehicles arrive late, chargers fail or prices change.
- Benefits: vendors and operators report lower energy cost and peak-power reduction, and the ability to host more vehicles on an existing grid connection. Figures differ strongly by tariff, fleet and site; measure your own baseline rather than relying on marketing numbers [1][2].
- Regulatory driver: the Clean Vehicles Directive sets minimum shares of zero-emission buses in public procurement; AFIR mandates public charging build-out and standardised interfaces; CO2 standards for heavy-duty vehicles push logistics firms to electrify [3][4].
Variants
- Municipal bus depots: overnight charging, fixed timetables, opportunity charging at terminals.
- Delivery vans and e-trucks: strict morning departures; for trucks, very high power and shared hubs.
- Company cars / workplace (SMB): load management behind a shared building connection, often with PV.
- Car sharing, taxi, rental: stochastic arrivals instead of fixed duty plans.
- V2G / bidirectional charging: feeding back power or selling flexibility (ISO 15118-20), gated by warranty.
- Private use: home charging on dynamic tariffs, see OPT-032.
How it works
- duty schedule + telematics (SoC, route, weather)
- energy-need forecast per vehicle (departure SoC target)
- site model: grid limit, building load, PV forecast, tariff/price forecast
- optimisation (MILP / MPC rolling horizon; RL or heuristics for large fleets)
- charger setpoints via OCPP smart-charging profiles / charge-point management system
- monitoring: arrival deviations, charger faults, price changes
- re-plan every few minutes
- dispatcher overrides + reporting
- Forecasting: gradient-boosted trees or neural time-series models predict consumption per route and arrival SoC; price and PV forecasts come from PRED-007/PRED-008.
- Optimisation core: a mixed-integer or linear program with constraints for connection limit, charger and vehicle power curves, readiness deadlines and charger-sharing/assignment, solved in a rolling horizon. The AI part is mostly the forecast and the robustness to uncertainty; solvers remain exact methods.
- Control: OCPP 1.6/2.0.1 smart-charging profiles set power limits per connector; ISO 15118 enables plug-and-charge and communicates departure time and energy demand from the vehicle.
- Human-in-the-loop: dispatchers set priorities (VIP routes, buffer targets), approve plan deviations and can always force full-power charging; alarms when readiness is at risk.
Data preparation
Needed: duty schedules and departure times (dispatch or fleet software, often Excel or paper rosters), telematics data (state of charge, consumption per route), charger status via OCPP, tariffs and grid fees (energy contracts, PDF invoices) and site load (smart meter, building management). The data sits in several systems that rarely talk to each other.
Typical gaps: duty plans that exist only as spreadsheets or in dispatchers' heads, telematics gaps or inconsistent vehicle IDs, chargers without OCPP or with proprietary protocols, tariffs only as PDF, no 15-minute site load profile, and missing records of actual departures.
Preparation steps: (1) inventory sources and owners (PREP-001, PREP-004); (2) export data from fleet, ERP and telematics systems via API or file export (PREP-005); (3) move duty plans from spreadsheets into tidy tables with one vehicle key (PREP-011); (4) digitise tariff and contract terms from PDF (PREP-006, DOC-014); (5) collect 3-6 months of history to build the consumption forecast and a test set (PREP-018). Tools: Python/pandas, a time-series database, the OCPP backend of the CSMS.
Rough share of effort: about 25-35 % of the project, mostly in getting reliable duty plans and telematics. See GUIDE-003 for the readiness check.
Implementation paths
Which path for whom
- smb: Path A, because mature OCPP charge-point management systems with smart-charging exist at SMB prices (many EU-hosted) and a depot with a few dozen chargers is a standard case.
- enterprise: Path B, because large fleets need integration with own dispatch, telematics and energy-contract data, which off-the-shelf tools cover only partly.
- developer: Path C, because OCPP stacks (e.g. EVerest) and open solvers (OR-Tools, HiGHS) run on HW1 and give full control.
- public: Path C, because municipal depots (bus operators, Stadtwerke) value sovereignty and one platform can serve several depots and sites.
Path A — Buy (recommended for most operators)
- Fleet depot software: Siemens Depot360 (EU), The Mobility House ChargePilot (EU), Ampcontrol, Driivz, AMPECO, Monta, ChargePoint fleet. Public transport: Optibus, IVU.suite (EU), Hacon/Init (EU).
- Charge-point management systems (CPMS) with load management cover small sites; check OCPP 2.0.1 support, ISO 15118, telematics integrations, tariff handling and EU hosting.
- Choose A when: < several hundred vehicles, standard chargers and telematics exist; contracts often bundle hardware, software and energy.
Path B — Build on cloud APIs
- Data: telematics via OEM APIs or aggregators (e.g. Smartcar, High Mobility (EU), Mercedes/Volvo/Daimler Truck fleet APIs), FMS standard interfaces, ENTSO-E Transparency Platform for day-ahead prices, Open-Meteo or DWD for weather, Forecast.Solar for PV.
- Optimisation: OR-Tools, Gurobi, HiGHS or cloud solvers (Azure, AWS) for MILP; forecasting with Nixtla (TimeGPT, StatsForecast), Chronos or Amazon/Google time-series services in EU regions.
- LLM role (optional): natural-language what-if and explanations for dispatchers; see OPT-018. The LLM must never set power limits directly.
- Choose B when: you have unusual constraints (mixed fleet, shared hub, own tariff logic) and a software team.
Path C — Self-host (open source)
- Charging back-end: SteVe or CitrineOS (OCPP), EVerest (Linux Foundation Energy, charger firmware stack), OCPP-compatible open CSMS projects; MQTT/Kafka for streaming.
- Optimisation: OR-Tools, HiGHS, Pyomo, PuLP, CVXPY; EV scheduling toolboxes such as ACN-Sim/SmartCharging research code (Caltech ACN) for simulation and baselines.
- Forecasting: LightGBM, Darts, Chronos or TimesFM (open weights); no GPU needed for most depots.
- Hardware: HW1 (industrial PC at the depot or small server); an on-site controller keeps charging safe when the cloud link drops.
Path D — Custom training
Train consumption models per vehicle type/route class once you have months of telematics; consider RL or learned surrogate solvers only for very large fleets with stochastic arrivals. Data: SoC traces, route profiles, charger logs, tariffs. Start with a rule-based plus MILP baseline and benchmark any learned policy against it in simulation.
State of the art (2026)
- Rolling-horizon MILP/MPC with probabilistic forecasts is the production standard; robust and chance-constrained variants protect departure readiness against forecast error.
- Bidirectional charging and V2G is moving from pilots to early commercial offers (standardised in ISO 15118-20; mandated-readiness debates in the EU); battery-warranty and degradation costs remain the main business-case limit.
- Grid-aware charging: DSO signals, flexible connection agreements and German section 14a EnWG control of controllable consumption make depots participants in congestion management (OPT-022).
- Limits and failure modes: wrong energy-need forecast (winter, hills, payload) leaves vehicles short; telematics gaps; charger/vehicle incompatibility and flaky OCPP implementations; real charge curves differ from nameplate; aggressive cost optimisation can accelerate battery wear; uncertain arrival times in non-fixed fleets. Always keep a safety buffer and a manual override.
Shared resources & combinations
- Same stack (optimization-solver + time-series): OPT-001 routing, OPT-019 dispatch, OPT-007 building energy, OPT-011 battery dispatch (depot buffer battery), OPT-032 home charging.
- Forecasts: PRED-007 load/PV, PRED-008 price, PRED-019 mobility demand.
- Grid and markets: OPT-022 congestion management, OPT-039 flexibility marketing and VPP.
- Shared infrastructure: one edge controller and OCPP back-end serve charging, building and PV control.
Effort, cost & KPIs
- Path A: weeks to a few months incl. charger commissioning; software typically priced per charger or vehicle per month, sometimes bundled with energy contracts.
- Path B: 3-9 months for a production system with telematics, tariffs and CPMS integration.
- Path C: 3-9 months; the main cost is integration and operations, not compute.
- Cost drivers: hardware and grid connection (far larger than software), telematics access, integrations.
- KPIs: departure readiness rate (% vehicles at target SoC), energy cost per kWh and per km, peak power vs. connection limit, share of energy charged in cheap/renewable hours, charger utilisation, battery throughput/degradation, number of manual interventions, CO2 per km.
People & change
Involve management (budget, energy and electrification strategy) and the depot or fleet lead (process owner) first; bring in IT and the grid or energy contractor early for charger, network and OCPP integration, and the data owners (dispatch, telematics, energy purchasing) for data access. Dispatchers and drivers are the daily users and should join from the pilot design onward.
Works council: depends. Pure charging optimisation is not a monitoring system, but linking charge sessions to named drivers or vehicles, telematics and driving behaviour can make the system objectively suited to monitor performance, so co-determination under §87(1) no. 6 BetrVG applies; independent of that, §90(1) no. 3 requires timely information when work processes (dispatch, shift and parking planning) are planned with AI. See GUIDE-004. Data protection: pseudonymise vehicle-to-driver links, let the DPO check the purpose and, with driver-level data, whether a DPIA is needed (GUIDE-005).
AI literacy: short training for dispatchers on what the plan does, when it may fail and how to override (GUIDE-006); AI Act Art. 4 requires measures to support staff AI literacy. Start with one depot as pilot group and AI champions among dispatchers (GUIDE-008). Typical concerns: "the vehicle is not full at departure" (answer: readiness alerts, buffer targets, full-power override) and "surveillance of drivers" (answer: no driver-level evaluation, written in a works agreement). Humans stay in control: dispatchers set priorities and can always force charging.
Risks, regulation & compliance
- EU AI Act: typically minimal risk. Charging optimisation is not an Annex III use by itself; however, if the system acts as a safety component in the management and operation of critical digital or energy infrastructure (Annex III point 2) at utility scale, high-risk classification may apply. Check case by case; AI literacy (Art. 4) applies to operators. Not legal advice [5].
- GDPR: vehicle location, driver identity and charging sessions of company cars are personal data when linked to employees; works-council co-determination in Germany (BetrVG section 87), data minimisation, pseudonymised telematics, and clear purpose limitation (no driver performance monitoring through charging data).
- Energy and grid law: AFIR (Regulation (EU) 2023/1804), the EU Electricity Market Design rules on flexibility, German section 14a EnWG, Eichrecht (calibration-law-compliant metering) for billing, grid-connection agreements with the DSO [3][6].
- Cybersecurity: NIS2 where in scope; secure OCPP with TLS and certificates, segment networks.
- Safety: hard limits stay in the charger/BMS, never in the AI layer; respect V2G warranty terms.
- Mitigations: SoC buffers, fallback to static schedule on data loss, simulation before roll-out, audit log of plans and overrides.
Getting started
- Collect baseline data for 4-8 weeks: arrival and departure times, SoC, kWh per route, site load and tariff.
- Check the grid connection, DSO conditions and tariff structure (peak-demand charge, dynamic tariff options).
- Choose OCPP 2.0.1-capable chargers and make sure telematics data is accessible by API.
- Start with static load management plus a simple departure-time-driven schedule; add forecast-based MILP once readiness is stable.
- Simulate the next quarter with historic data (e.g. ACN-Sim) and agree readiness buffers with dispatch before go-live.
Sources
- The Mobility House - ChargePilot fleet charging management - https://www.mobilityhouse.com/int_en/charging-and-energy-management
- Siemens eMobility - eBus Depot360 / depot charging management - https://www.siemens.com/global/en/products/energy/e-mobility.html
- Regulation (EU) 2023/1804 (AFIR) - https://eur-lex.europa.eu/eli/reg/2023/1804/oj
- Directive (EU) 2019/1161 (Clean Vehicles Directive) - https://eur-lex.europa.eu/eli/dir/2019/1161/oj
- Regulation (EU) 2024/1689 (AI Act) - https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- Open Charge Alliance - OCPP specifications - https://openchargealliance.org/protocols/open-charge-point-protocol/
- EVerest project (LF Energy) - https://lfenergy.org/projects/everest/
- Caltech ACN-Sim adaptive charging network simulator - https://github.com/zach401/acnportal