L1 Sensing & execution
Collect metering, environmental and device states. Pass authorized actions and acknowledgments through approved interfaces. Native device protection takes priority over AI.
DCP / Public engineering overview
Compute-power architecture
Bring compute tasks, distributed generation, storage and flexible loads into one resource view. This public overview explains responsibilities, operating boundaries and validation. Protocol support, response times and available functions require project testing.
AI concept · Not a project photograph or product specification01Five roles. Local control.
These layers describe control and service responsibilities, not the physical endpoint, community and regional deployment tiers.
L2 Compute nodes
Run models and queues, reporting power, health and task constraints. Only validated flexible workloads enter the adjustable resource pool.
L3 Local coordination
Check site capacity, equipment state and authority before executing valid policies. Handle disconnection and faults using previously tested procedures.
L4 Operations platform
Aggregate resources, forecast demand, coordinate sites and analyze results. Site safety must not depend on permanent cloud connectivity.
L5 Ecosystem services
Connect energy, compute and operating services where qualifications and interfaces permit, supporting metering, verification and settlement.
02Close the resource loop
Understand state and constraints, formulate a policy, then verify execution.
States & resource models
Combine power, temperature, queues, storage state and device health in timestamped records. Distinguish rated, online and committable capacity.
Integration & authority
Design adapters around site protocols, point-list quality and vendor permissions. Exchange only necessary data; integration does not imply control authority.
Policies & acknowledgments
Specify target equipment, validity and fallback conditions. Compare targets with measured results and record deviations and exceptions.
03Deployment and fault boundaries
Expand from a single site to multiple sites only as recovery capability grows.
Single site to multi-site
Validate data collection, permissions, execution and recovery first. Before expansion, check networks, data location, model compatibility and migration costs.
Cloud or controller failures
Use agreed valid policies and local protection. Stop expired commands. Reconcile states before gradually resuming control.
Node or workload failures
Choose recovery according to checkpointing, migration and deferral capabilities. Critical tasks that cannot move need a separate continuity plan.
DCP / OPERATING LOOP
Compute, power and task coordination
- 01
Observe
Power, queues, device health
- 02
Validate
Permissions, capacity, deadlines
- 03
Schedule
Assign location and time window
- 04
Execute locally
Retain native device protection
- 05
Verify & audit
Results, deviations, traceability
Public logic illustration, not live telemetry. Observations feed validated decisions; execution returns auditable evidence.