Federated triangulation architecture for extraterrestrial dc grid stabilisation

Abstract

A system for stabilising extraterrestrial DC power networks 10 comprises orbital nodes, static-surface nodes, and static-mobile nodes 14a-14n, each equipped with a central processing unit (CPU) and a

Description

Field of the Invention

[0001] The invention relates to distributed computing, power-management, and control systems for extraterrestrial electrical networks. More particularly, it concerns a physics-supervised, federated learning and control architecture implemented across orbital, static surface, and mobile surface nodes equipped with central processing units (CPUs) and parallel-processing units (PPUs). The architecture enables stabilisation and optimisation of direct-current (DC) power networks powered by solar-generation modules, small modular reactors (SMRs), or hybrid combinations thereof. Through physics-constrained causal learning and coordinated node-level actuation, the system provides autonomous regulation, energy stabilisation, and adaptive model updating for extraterrestrial installations. Background of the Invention

[0002] Future lunar, Martian, and orbital installations will rely on hybrid direct-current (DC) power networks combining solar-generation arrays with small modular reactor (SMR) systems. Solar sources exhibit variable output dependent on illumination angle, eclipse cycles, and dust accumulation, whereas SMRs provide steady base load but require careful thermal management and controlled startup sequencing. The interaction of these heterogeneous sources produces complex, time-varying electrical and thermal dynamics that cannot be managed effectively by conventional centralised controllers.

[0003] Traditional supervisory and rule-based control systems assume continuous communication, deterministic timing, and fixed network topology—conditions that are rarely satisfied in extraterrestrial environments subject to propagation delays, intermittent line-of-sight, radiation-induced faults, and dynamically changing asset configuration. Under such conditions, centralised architectures lose synchronisation and may induce instability in DC-bus regulation, power routing, or thermal balance. A distributed control approach is therefore required, wherein each compute node—whether in orbit, on the planetary surface, or integrated into mobile assets—can operate semi-independently while contributing to a shared stabilisation objective.

[0004] To coordinate these tiers, a federated-triangulation framework is employed: orbital nodes supply global timing and positional reference; static surface nodes act as aggregation and dispatch hubs; and mobile surface nodes perform high-frequency sensing and actuation. This geometric distribution provides redundancy, resilience to communication loss, and the ability to maintain local grid stability through physics-supervised model adaptation even when higher-tier connectivity is degraded or unavailable.

[0005] Existing spacecraft power systems rely on deterministic feedback loops and preprogrammed fault responses that cannot adapt to unforeseen interactions or update control behaviour from real-time physical conditions. Digital-twin and simulation tools offer designphase insight but operate offline and do not exert causal influence on the physical power network during operation. They therefore cannot address non-linearities, harmonics, oscillations, or emergent behaviours generated by hybrid extraterrestrial grids.

[0006] Accordingly, there is a need for a distributed, physics-supervised learning and control architecture capable of coordinating hybrid solar-SMR DC networks through federated triangulation of orbital, static, and mobile nodes. The Large Energy Model for Space (LEM-Space) disclosed herein meets this need by embedding a physics-constrained adaptive-control layer directly into the power network, enabling stable, scalable, and autonomous operation under extraterrestrial conditions. Summary of the Invention

[0007] The present invention provides a distributed, physics-supervised control and modeladaptation system for stabilising extraterrestrial direct-current (DC) power networks. The system integrates orbital, static-surface, and mobile nodes equipped with central processing units (CPUs) and parallel-processing units (PPUs) coordinated under a federated-triangulation architecture. Each node acquires local electrical, thermal, and environmental measurements from hybrid DC grids powered by solar-generation arrays and small modular reactor (SMR) sources. Telemetry and model parameters are exchanged asynchronously through a disruption-tolerant communication layer, enabling continued operation despite latency, intermittent connectivity, or partial node loss. An orchestration controller, referred to herein as PAVLINA (Predictive Autonomous Voltage and Load Intelligence Network Architecture), governs compute load, energy modulation, and federated-model aggregation using physics-supervised causal-update algorithms constrained by conservation laws of energy and charge.

[0008] The architecture provides real-time, distributed stabilisation of extraterrestrial DC-bus voltage, power flow, and thermal behaviour without reliance on continuous ground contact. By coordinating variable solar inputs with steady SMR output inside a unified, physics-supervised control layer, the system harmonises heterogeneous power sources and maintains grid equilibrium under dynamic and unpredictable operating conditions.

[0009] Each node is engineered for resilience through multilayer radiation and micrometeoroid shielding, error-correction memory, and checkpoint-restart logic to ensure functionality in harsh extraterrestrial environments. Physical context—including illumination angle, dust concentration, temperature, and thermal-cycling state—is embedded within each causal tuple of state, action, effect, and environment, enabling adaptive recalibration of control parameters driven by observed physical behaviour rather than pre-defined rules. Through asynchronous, trust-weighted federated-learning cycles, the system maintains local autonomy during communication outages, while aggregated updates restore global model consistency when connectivity returns.

[0010] Model adaptation may be initialised or fine-tuned on a pre-existing Large Energy Model (LEM) trained on terrestrial grid data, enabling rapid transfer of Earth-derived physics representations to extraterrestrial DC systems. This reduces commissioning time, improves early-stage stability, and supports progressive refinement as additional extraterrestrial telemetry becomes available. By distributing computation and control across heterogeneous processing hardware—including CPUs, GPUs, TPUs, FPGAs, ASICs, and neuromorphic processors—the architecture scales from individual base-station microgrids to coordinated planetary-surface and orbital power constellations.

[0011] Overall, the invention transforms an extraterrestrial power grid into a self-correcting computational layer that performs continuous, physics-supervised regulation of electrical and thermal energy flow. It delivers long-term stability, adaptability, and energy efficiency for hybrid solar-SMR power networks, enabling sustainable and autonomous operation of infrastructure in lunar, Martian, and orbital environments. Detailed Description of the Invention Overview

[0012] The present invention relates to distributed computing and power-management systems for extraterrestrial applications. More particularly, it concerns a network of orbital, static-surface, and mobile nodes equipped with parallel-processing hardware and coordinated under a physics-supervised causal-learning framework to stabilise direct-current (DC) electrical grids powered by solar-generation and small modular reactor (SMR) sources.

[0013] As shown in FIG. 1, the system comprises an extraterrestrial DC-power network (10) interconnected through a federated orchestration controller, PAVLINA (12), and a plurality of distributed nodes (14a-14n). Each node participates in a physics-supervised learning process managed by a spatiotemporal causal-learning framework referred to as the Large Energy Model for Space (LEM-Space) (18). Telemetry is exchanged through a disruption-tolerant communication layer (16), and all nodes are enclosed within protective shielding engineered for radiation, thermal, and micrometeoroid resilience.

[0014] As illustrated in FIG. 2, the federated-triangulation architecture comprises static surface nodes, mobile surface nodes, and orbital nodes forming a coordinated, geometry-aware control hierarchy. System Architecture

[0015] Each distributed node (14a-14n) contains a central processing unit (CPU) for supervisory control and a parallel-processing unit (PPU)—which may be a graphics processing unit (GPU), tensor processing unit (TPU), field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), neuromorphic processor, or other high-throughput device—responsible for real-time computation and causal-learning operations.

[0016] A sensor suite within each node measures electrical, thermal, and environmental parameters including DC-bus voltage, current, temperature, dust concentration, illumination angle, and thermal-cycling state. These measurements are encoded into causal tuples comprising state, action, effect, and context for use by the physics-supervised learning framework (18).

[0017] Nodes are deployed in three tiers: 1. Orbital nodes, providing global timing, positional reference, and downlink coordination. 2. Static surface nodes, functioning as aggregation servers and local energy-management hubs. 3. Mobile surface nodes, such as rovers or deployable tools, performing fine-grained sensing, actuation, and adaptive control.

[0018] Collectively, these tiers form a federated-triangulation architecture that provides redundancy, positional accuracy, and system-wide synchronisation even under degraded or intermittent connectivity. Federated Learning and Control

[0019] The orchestration controller (12), designated PAVLINA (Predictive Autonomous Voltage and Load Intelligence Network Architecture), coordinates all nodes through asynchronous, disruption-tolerant communication links (16). Each node transmits encrypted telemetry and receives updated parameters during available communication windows. Federated-learning cycles incorporate delayed-gradient correction to accommodate intermittent latency.

[0020] Within each learning round, nodes perform local physics-supervised training using real-time electrical and environmental data. Model updates are cryptographically signed and aggregated by PAVLINA using Byzantine-robust consensus algorithms, with trust weights assigned according to node-health metrics, radiation exposure, and timing confidence.

[0021] To enhance causal inference, PAVLINA may instruct nodes to perform active microexperiments—controlled modulations of compute load or electrical output executed within defined safety envelopes—to generate additional state-action-effect data for system identification.

[0022] Compute workload on each CPU and PPU is dynamically modulated in proportion to local DC-bus voltage deviation or power-flow imbalance, providing synthetic inertia, harmonic attenuation, and voltage stabilisation for the extraterrestrial DC network (10). Hybrid Power Integration

[0023] The DC-power network (10) includes solar arrays, SMRs, or combinations thereof. The orchestration controller PAVLINA (12) continuously estimates instantaneous power availability and allocates computational and stabilisation duties across these sources to maintain DC-bus equilibrium and thermal balance.

[0024] This hybrid coordination enables the system to manage variable irradiance, eclipse cycles, dust deposition, and SMR thermal transients while sustaining continuous computation and stable electrical performance. Shielding and Fault Tolerance

[0025] Each node (14a-14n) is enclosed within a protective shielding subsystem comprising multilayer radiation barriers, regolith-based berms or other impact-attenuating materials, and integrated thermal-management structures. To mitigate single-event upsets, nodes employ error-correction memory, checkpoint-restart logic, and local fallback control models. In the event of communication loss, a node autonomously reverts to its most recently synchronised control parameters until reconnection, ensuring safe, standalone operation. Earth-Trained Foundational Layer

[0026] In certain embodiments, the LEM-Space framework (18) is initialised or fine-tuned on a pre-existing terrestrial Large Energy Model (LEM). This transfer-learning approach enables extraterrestrial deployments to inherit physics-consistent representations of energy flow, accelerating convergence and improving early-stage stability under novel DC-grid conditions. Operation Summary

[0027] During normal operation: 1. Nodes capture multimodal telemetry via onboard sensors. 2. CPUs execute local control loops while PPUs perform real-time inference and gradient updates. 3. PAVLINA (12) aggregates and distributes model parameters via asynchronous, disruption-tolerant networking (16). 4. Federated-learning updates are constrained by conservation laws of energy and charge. 5. The combined system produces a self-stabilising, physics-supervised adaptive-control layer (18) that maintains electrical and thermal equilibrium across the extraterrestrial DC network (10). Advantages and Technical Effects

[0028] The disclosed system delivers real-time, distributed stabilisation of extraterrestrial DC power networks using heterogeneous computational hardware. It integrates hybrid solargeneration and SMR sources under a unified physics-supervised control architecture that dynamically balances power flow, voltage stability, harmonic content, and thermal behaviour across nodes operating in orbital, surface, or mobile environments. The architecture maintains reliable operation under radiation exposure, micrometeoroid impacts, thermal cycling, and intermittent communication.

[0029] In certain embodiments, the control framework is initialised or fine-tuned using a terrestrial Large Energy Model (LEM), enabling transfer learning and rapid deployment in off-world environments. Through this distributed, physics-constrained architecture, the power network functions as a self-correcting computational layer capable of adaptive energy management and sustained autonomous operation beyond Earth. Key to Figures Reference Numerals 10: Extraterrestrial DC grid 12: Orchestration controller 14a-14n: Orbital nodes, static surface nodes, and mobile surface nodes 16: Encrypted telemetry 18: Large Energy Model for Space (LEM-Space) Figure Captions FIG. 1 Illustrates the overall architecture of the Large Energy Model for Space. FIG. 2 Illustrates the federated triangulation architecture comprising orbital nodes, static surface nodes, and mobile surface nodes. 31 12 25

Claims

26Claims1. A distributed, physics-supervised learning and control system for extraterrestrial direct-current (DC) power networks, comprising:(a) at least one orbital node, at least one static surface node, and at least one mobile surface node, each including a central processing unit (CPU) and a parallel-processing unit (PPU) selected from a graphics processing unit (GPU), tensor processing unit (TPU), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or neuromorphic processor;(b) a sensor suite at each node configured to measure electrical, thermal, and environmental parameters of a local DC-power system comprising one or more solargeneration modules or small modular reactors (SMRs), including measurements of DC-bus voltage and current representing an electrical state of the DC power network;(c) a telemetry and communication module implementing a disruption-tolerant networking protocol enabling asynchronous data exchange and federated model exchange among the orbital, static, and mobile nodes; and(d) an orchestration controller operative across the nodes to coordinate compute load modulation and federated-model aggregation using physics-supervised causal-learning processes based on state-action-effect tuples derived from said electrical, thermal, and environmental measurements,wherein modulation of compute workload at one or more nodes alters electrical power consumption at the node within the DC power network and thereby contributes to stabilisation of DC-bus voltage and power-flow behaviour.

2. The system of claim 1, wherein the orchestration controller operates according to a federated-triangulation architecture in which the orbital node provides global timing and positional reference, the static surface node functions as a local aggregation and distribution hub, and the mobile surface node performs high-frequency data acquisition and actuation.16 03 263. The system of claim 1 or 2, wherein each node comprises a navigation and timing module including one or more of a star tracker, inertial-measurement unit, terrainrelative navigation camera, orbiter-ranging transceiver, pulsar-timing receiver, or planetary GNSS receiver, and wherein federated-learning updates are weighted by positional and timing confidence.

4. The system of any preceding claim, wherein the telemetry and communication module employs asynchronous federated-learning scheduling with delayed-gradient correction to accommodate intermittent connectivity between tiers of the triangulation architecture.

5. The system of any preceding claim, wherein the orchestration controller executes active micro-experiments on one or more surface nodes by modulating compute load or electrical output within predefined safety envelopes to generate causal data for system identification.

6. The system of any preceding claim, wherein each node includes a cryptographicsigning module for authentication of telemetry, control data, and model-update data, and wherein the orchestration controller performs Byzantine-robust aggregation using trust scores derived from node-health and radiation-event telemetry.

7. The system of any preceding claim, wherein compute workload on each CPU and PPU is dynamically adjusted in response to local DC-bus voltage or power-flow variation to provide synthetic inertia and voltage stabilisation within the extraterrestrial electrical network.

8. The system of any preceding claim, wherein each node is further configured to detect and attenuate local harmonic distortion, switching transients, or internal oscillations within its attached power-electronic subsystem by dynamically modulating compute load or control-output waveforms to stabilise the node’s internal DC bus and prevent propagation of distortion into the wider extraterrestrial power network.

9. The system of any preceding claim, wherein environmental context parameters including illumination angle, dust concentration, temperature, and thermal-cycling16 03 26state are embedded within each causal-learning tuple comprising a state, an action, an effect, and a context.

10. The system of any preceding claim, wherein each node is enclosed within a radiation-, thermal-, and micrometeoroid-resistant housing comprising multilayer shielding, regolith-based berms or other impact-attenuating materials, error-correction memory, and checkpoint-restart logic to mitigate single-event upsets and micrometeoroid penetration.

11. The system of any preceding claim, wherein upon loss of communication each surface or mobile node reverts to a local fallback control model derived from most-recent federated parameters to ensure continued safe operation until reconnection.

12. The system of any preceding claim, wherein the static surface node is configured to store and disseminate federated-model parameters and aggregated telemetry to an Earth-based orchestration controller when communication windows permit.

13. The system of any preceding claim, wherein the orchestration controller implements a Large Energy Model for Space (LEM-Space) configured to perform physics-supervised spatiotemporal causal inference across orbital, static, and mobile datasets to predict and stabilise electrical and thermal behaviour of extraterrestrial installations.

14. The system of any preceding claim, wherein the federated-learning process employs trust-weighted gradient aggregation such that contributions from nodes exhibiting elevated radiation flux, thermal stress, or clock drift are proportionally reduced.

15. The system of any preceding claim, wherein the disruption-tolerant networking employs optical or radio-frequency crosslinks between orbiters and surface nodes, and wherein model-synchronisation packets are prioritised using predicted link-quality metrics.

16. The system of any preceding claim, wherein the orchestration controller operates under a physics-supervised causal-learning algorithm that constrains parameter updates according to conservation laws of energy and charge within the extraterrestrial DC-grid topology.

17. The system of any preceding claim, wherein the DC-power network comprises hybrid solar-generation and small-modular-reactor sources, and wherein the orchestration controller dynamically allocates compute load and stabilisation response between the sources to maintain DC-bus stability and thermal balance.

18. The system of any preceding claim, wherein the physics-supervised learning process is initialised or fine-tuned on a pre-existing Large Energy Model trained on terrestrial el ectri cal-grid data to enable transfer learning and adaptation of Earth-derived physics representations to extraterrestrial DC-grid conditions.16 03 26A

Images