Energy Router: Dynamic Control and Conversion

Updated 4 July 2026

Energy router is a control and conversion node that treats energy as an explicitly routable quantity, integrating power-electronic ports and internal buffering.
It leverages multi-port interfaces and software-defined control to manage hybrid AC–DC grids, power-packet systems, and renewable integration efficiently.
Its applications span smart grids, urban renewable networks, and mobile infrastructures, enhancing efficiency, traceability, and dynamic energy management.

Searching arXiv for papers on “energy router” to ground the article in current literature. arXiv search query: energy router power packet grid energy router hybrid AC DC

An energy router is a control and conversion node that manages energy flow between sources, storage, loads, and network interfaces, typically through power-electronic ports, internal buffering, and software-defined control. Recent literature uses the term in several overlapping senses: as a multi-port device for hybrid AC–DC distribution grids, as a software-defined gateway with galvanic separation and a DC backplane, as a power-packet router that forwards energy units with embedded information tags, and, in some communications and computing contexts, as a controller that routes traffic or workloads according to energy availability or energy cost (Asadi et al., 9 Nov 2025).

1. Conceptual scope and terminological range

The literature suggests that “energy router” is not a single canonical architecture but a family of constructs unified by one principle: energy flow is treated as an explicitly routable quantity rather than an uncontrolled consequence of network impedances. In hybrid AC–DC grid work, the energy router is a controllable multiport node that connects multiple AC and DC feeders, routes power bidirectionally, and independently controls active power $P$ , reactive power $Q$ , and voltages (Asadi et al., 9 Nov 2025). In EnergyNet, it is defined as “a power‑electronics device designed to manage energy flows dynamically, software‑defined, and safely,” with galvanic separation, a DC backplane, and software-controlled energy flows via the Energy Router Operating System and Energy Protocol (Birgersson et al., 9 Sep 2025). In power-packet systems, it is a router in the literal packet-switching sense: it reads an information tag attached to a power payload and selectively forwards that payload over wired or wireless channels (Mamiya et al., 2023).

A related but broader usage appears in routing and network control. In heterogeneous wireless networks, a routing protocol can be interpreted as an “energy router” when it minimizes energy consumption while maximizing network lifetime and accepts a slight compromise in end-to-end delay (Fedrizzi et al., 2012). In Integrated Access and Backhaul with O-RAN, the control plane becomes a logical energy router by jointly selecting routes and active nodes to minimize the number of active IAB nodes while maintaining per-UE capacity (Gemmi et al., 2023). In renewable-aware mobile infrastructure, a central energy router coordinates a power packet grid among base stations using Gaussian Processes and Model Predictive Control (Gambin et al., 2018).

Meaning in the literature	Characteristic mechanism	Representative paper
Multi-port grid node	Hybrid AC–DC interfacing, $P/Q/V$ control	(Asadi et al., 9 Nov 2025)
Software-defined gateway	Galvanic separation, DC backplane, EROS/EP	(Birgersson et al., 9 Sep 2025)
Power-packet router	Header-aware store-and-forward energy transfer	(Mamiya et al., 2023)
Line-switching router	Crossbar of bidirectional semiconductor switches	(Mochiyama et al., 30 Jun 2025)
Hierarchical grid energy router	R/F/B/U/M layered architecture	(Chen et al., 2020)
Energy-aware routing controller	Route/path selection driven by energy metrics	(Fedrizzi et al., 2012)

2. Device architectures and internal organization

Several architectures recur. A prominent grid-oriented design is the partial-power energy router for hybrid AC–DC grids. It comprises a common DC bus powered by an Active Front-End, Dual Active Bridges supplying floating low-voltage links, and modular low-voltage high-current series modules inserted in AC or DC feeders (Asadi et al., 9 Nov 2025). Its central claim is partial-power processing: the series module injects only a small fraction of the line voltage while controlling the full line power flow. Table I in that work reports partial-power processing of about $10\%$ , efficiency $>99\%$ , a $43\%$ cost reduction, and a $3.1\times$ reliability improvement for a $200\,\mathrm{kW}$ system relative to full-power solutions (Asadi et al., 9 Nov 2025).

A second architecture is the software-defined EnergyNet router. Its defining elements are galvanic separation at each port boundary, an internal DC backplane, modular AC/DC and DC/DC converter ports, and a dual-supervisor control plane running EROS and the EP-Server under operator-scale management through ENMS (Birgersson et al., 9 Sep 2025). The paper distinguishes four logical port roles: local consumption, traditional grid interconnection, connections to other Energy Routers, and local energy resources such as PV, batteries, and EVs. The DC backplane functions as a shared internal “energy bus” through which software-defined port-to-port routing is enforced (Birgersson et al., 9 Sep 2025).

A third architecture is the line-switching power router for traceable multi-energy management. Here the router is a crossbar switching matrix built from bidirectional semiconductor switches using back-to-back SiC MOSFETs. One experimental implementation uses Wolfspeed C3M0060065 devices rated at $650\,\mathrm{V}$ and $20\,\mathrm{A}$ , with per-port voltage and current sensing and direct electrical circuit formation between selected ports (Mochiyama et al., 30 Jun 2025). This approach does not perform voltage conversion; its distinctive property is physical distinguishability of power paths, which the authors use to support traceability of shared electricity and renewable-origin hydrogen (Mochiyama et al., 2024).

A fourth architecture is the Grid Energy Router. It is organized into five layers: Routing, Forwarding, Buffer, User, and Multi-Carrier Energy (Chen et al., 2020). The Routing layer handles exchange with AC/DC subsystems and neighboring routers; the Forwarding layer supports peer-to-peer forwarding; the Buffer layer contains energy storage; the User layer interfaces distributed generation and loads; and the Multi-Carrier layer connects to an energy hub. Its internal DC-link balance is written as

$Q$ 0

with the Buffer layer used to maintain bus stability (Chen et al., 2020).

3. Control, dispatch, and optimization

Control formulations vary with architectural assumptions. In the Grid Energy Router, the primary dispatch problem is explicitly bi-level. The upper level uses a modified Model Predictive Control formulation over a short-time horizon to optimize Routing-layer powers and Buffer-layer energy, while the lower level uses fuzzy logic control and a distributed fast compensation strategy to track the optimized buffer-energy reference in real time (Chen et al., 2020). The upper-level objective combines penalties on deviation from user-specified port variation ranges with a power-balance term, and the buffer state evolves according to charging and discharging efficiencies: $Q$ 1 Simulation results in that paper show that short-time power variation can be suppressed by sharing energy buffer across multiple GERs (Chen et al., 2020).

In O-RAN-controlled IAB networks, energy routing is expressed as a network optimization. The decision variables include node activation $Q$ 2, per-commodity edge usage $Q$ 3, and aggregate edge activation $Q$ 4, and the objective is

$Q$ 5

which approximates RAN energy minimization by minimizing the number of active gNBs (Gemmi et al., 2023). The formulation enforces flow conservation, tree constraints, and TDMA-coupled wireless capacities, and is transformed from a binary nonlinear program into an equivalent binary linear program solved with Gurobi. On a Milan scenario with 1 donor and 7 IAB nodes, the optimized tree achieved $Q$ 6 gNB-hours versus $Q$ 7 gNB-hours for the always-on baseline, corresponding to a $Q$ 8 reduction in RAN energy consumption while maintaining a minimum per-UE capacity target of $Q$ 9 (Gemmi et al., 2023).

In energy-sustainable mobile networks, the energy router is centralized and predictive. A power packet grid permits energy transfer among base stations, Gaussian Processes forecast harvested energy and traffic, and a Model Predictive Control framework schedules energy allocation and transfer (Gambin et al., 2018). Numerical results using real harvesting and traffic profiles report outage probability equal to zero in most cases and energy purchased from the power grid “more than halved” relative to optimization without GP forecasting and MPC (Gambin et al., 2018).

In heterogeneous multi-hop wireless networks, the router appears as an energy-aware path-selection mechanism rather than a dedicated power-electronic device. The protocol described in the abstract aims at a trade-off between energy consumption and routing delay, accounts for both device energy consumption and link energy costs, defines route-path utility functions to minimize energy consumption while maximizing network lifetime, and is reported to be energy efficient in path selection with a slight compromise in end-to-end delay (Fedrizzi et al., 2012).

4. Packetized power, information, and thermodynamic interpretations

One major branch of the literature models energy routing as packet switching. In wireless power-packet transmission, a power packet contains a payload that transfers real power and an information tag that carries control bits (Mamiya et al., 2023). A simplified packet format in that work fixes the packet at $P/Q/V$ 0 bits, with bits $P/Q/V$ 1– $P/Q/V$ 2 equal to 010 for clock synchronization, bits $P/Q/V$ 3– $P/Q/V$ 4 representing a 4-bit destination address, and bits $P/Q/V$ 5– $P/Q/V$ 6 representing the payload interval (Mamiya et al., 2023). The wireless router combines a class-E inverter, resonant coils, a demodulation branch for header detection, and a class-E rectifier for payload reception, so that only the addressed receiver closes its rectifier and captures the payload.

Power-packet routing can also be cast as a shortest-path problem with a physically derived link metric. In routing optimization for power packet dispatching, the network is modeled as a graph $P/Q/V$ 7, and the edge cost is defined as the ratio of lost energy to received energy for one packet transfer (Mochiyama et al., 2021). Router–router, source–router, and router–load connections are derived from RC circuit models, and the route is selected by minimizing the sum of per-edge costs along the path. Numerical experiments show that the proposed algorithm can allocate distributed sources to load demands and identify the optimal path for power delivery (Mochiyama et al., 2021).

A more radical interpretation appears in power packet networks treated through information thermodynamics. There the router is formalized as a non-equilibrium open system with state $P/Q/V$ 8 and Langevin dynamics

$P/Q/V$ 9

where $10\%$ 0 is the switch state and $10\%$ 1 is noise intensity (Hikihara, 28 Mar 2026). The router’s optimized control effort $10\%$ 2 incurs an exponential information-acquisition cost

$10\%$ 3

and its evaluation function is

$10\%$ 4

That model exhibits a first-order transition at a critical noise level $10\%$ 5, above which the optimal control drops to $10\%$ 6, a regime described as a “strategic abandon of regulation” and interpreted as an information barrier for autonomous energy management (Hikihara, 28 Mar 2026).

5. Application domains

The most direct application domain is hybrid AC–DC distribution. The partial-power router targets low- and medium-voltage grids with high shares of electronically interfaced renewables, loads with internal DC links, and meshed or ring topologies that require active control of power flows and voltages (Asadi et al., 9 Nov 2025). Real-time hardware-in-the-loop and prototype measurements in that work validate power sharing, voltage regulation, reactive power control, ripple mitigation, and virtual inertia across AC and DC feeders (Asadi et al., 9 Nov 2025).

A second domain is software-defined local and regional energy networks. EnergyNet uses Energy Local Area Networks and Energy Wide Area Networks interconnected through an open Energy Protocol, with Energy Routers providing galvanic separation, a DC backplane, and operator-scale control through EROS, EP-Server, and ENMS (Birgersson et al., 9 Sep 2025). Municipal demonstrators described there include a Lund deployment with two sites linked by a “Freedom Cable,” each site equipped with an $10\%$ 7 DC Energy Router and local PV, battery storage, and EV charging (Birgersson et al., 9 Sep 2025).

A third domain is residential and community-scale multi-energy management with electricity and hydrogen. In the line-switching power-router system, households share PV, stationary batteries, vehicle-mounted batteries, and a common hydrogen facility, while the router physically distinguishes which source feeds which sink (Mochiyama et al., 2024). Experimental verification of a bidirectional power router using a commercially available stationary battery shows dynamic change of bidirectional power flow and a switching algorithm based on power-flow monitoring that permits reconfiguration without disturbing smooth and stable power-system operation (Mochiyama et al., 30 Jun 2025).

A fourth domain is communication and mobile infrastructure. Renewable-aware ICN routes content requests along routers with the highest renewable energy using a distributed gradient-based protocol in which the green ratio

$10\%$ 8

is combined with hop distance (Mineraud et al., 2014). In mobile access networks, the O-RAN non-RT RIC can act as a centralized logical energy router, periodically measuring, optimizing, and reconfiguring IAB topology and node power states (Gemmi et al., 2023).

A fifth and more recent extension is compute-to-energy routing. XWind treats inference serving at wind farms as a cross-site routing problem in which requests are steered according to site power budgets and real-time telemetry (Reddy et al., 22 May 2026). Its feasibility analysis reports $10\%$ 9 GW of wind capacity within $>99\%$ 0 RTT of Azure data centers and $>99\%$ 1 of that capacity within $>99\%$ 2, while the router reduces P99 end-to-end latency by up to $>99\%$ 3 over the strongest contender and by up to $>99\%$ 4 over power-capping and GPU-idling baselines (Reddy et al., 22 May 2026). This suggests that the term has expanded from routing electrons to routing computation under energy constraints.

6. Misconceptions, unresolved issues, and research directions

A common misconception is that an energy router is simply another inverter. Several papers explicitly distinguish it from narrower device classes. The EnergyNet router is presented as more than a smart inverter because it combines galvanic separation, multiport routing through a DC backplane, and an open control plane (Birgersson et al., 9 Sep 2025). The line-switching router, conversely, is deliberately less than a general converter-based hub: it does not perform AC/DC conversion or voltage transformation, and is chosen precisely because this preserves physical traceability of power origins (Mochiyama et al., 2024).

Another misconception is that packetization is intrinsic to all energy routers. Packetized approaches are central in power-packet networks and wireless power-packet transmission (Mamiya et al., 2023), but other architectures rely on partial-power series injection (Asadi et al., 9 Nov 2025), layered DC-link control (Chen et al., 2020), or software-defined port scheduling (Birgersson et al., 9 Sep 2025). The literature therefore suggests that packetization is one design lineage rather than a universal requirement.

The central unresolved issue is standardization of abstraction level. Some papers treat the energy router as hardware at feeder level (Asadi et al., 9 Nov 2025), some as a cyber-physical gateway with protocol stacks (Birgersson et al., 9 Sep 2025), some as a graph-routing controller (Mochiyama et al., 2021), and some as an optimization function in wireless or mobile networks (Fedrizzi et al., 2012). A plausible implication is that future work will need cleaner separation between port-level conversion, network-layer path selection, market and policy logic, and protocol interoperability.

Current research directions are already visible. Grid-facing work emphasizes hybrid AC–DC interfacing, meshed distribution, buffer sharing, and multi-carrier energy integration (Chen et al., 2020). Packetized-power work emphasizes header design, selective forwarding, and unified wired–wireless power management (Mamiya et al., 2023). Thermodynamic work emphasizes the cost of information itself and critical operating windows under noise (Hikihara, 28 Mar 2026). Software-defined work emphasizes operator-scale control, security, and incremental migration from legacy grids (Birgersson et al., 9 Sep 2025). Across these strands, the recurring design problem is the same: how to couple physical energy flow, buffering, and control information so that routing decisions remain efficient, stable, and auditable under heterogeneous sources, loads, and network conditions.