Wi pi wireless power dynamic equalization method for energy storage system grid connection

Abstract

The present application relates to the technical field of power electronic and energy storage system control, especially to a WIPI wireless power dynamic balancing method for energy storage system grid connection, comprising the following steps: constructing a fractional order digital twin model to predict the state of charge and temperature rise of the module; based on the prediction results, using a three-layer GraphSAGE and a twenty-qubit variational quantum optimizer to obtain a demand matrix and a loss matrix respectively, and fusing them into a power coupling target according to a preset weight; thresholding the target to generate a 32x32 magneto-acoustic super-pixel image, completing magnetic field shaping at the microsecond level, and simultaneously driving a GaN inverter to output matching phase and frequency; collecting power flow and temperature data through a high-precision analog-to-digital converter, and using an entropy regularization Sinkhorn algorithm to iteratively correct the power transport matrix and update the pixels in real time; finally, writing the feedback hash into a consortium chain storage to complete the millisecond-level trusted traceability. The present application realizes fast, accurate and traceable balancing of wireless energy of the energy storage module.

Description

Technical Field

[0001] This invention relates to the field of power electronics and energy storage system control technology, and in particular to the WIPI wireless power dynamic balancing method for grid-connected energy storage systems. Background Technology

[0002] During grid-connected operation, energy storage power stations are prone to state-of-charge imbalance among battery modules due to differences in aging rates and thermal environments, leading to capacity waste and localized overheating. Wireless Inductive-Power Interchange (WIPI) can continuously redistribute module power flow without interrupting the DC bus, which is crucial for improving available capacity and thermal safety. Existing dynamic balancing methods mostly employ a "switch slicing" approach (Chinese invention patent, publication number: CN114899923B), such as temporarily bypassing individual cells or connecting equalization resistors in series, and then cycling sequentially according to voltage error. This approach relies on mechanical / solid-state switches, whose operating frequency is limited by device lifespan, and the switching process disrupts bus continuity; the algorithm is based only on instantaneous voltage, ignoring magnetic and thermal coupling, resulting in low adjustment accuracy; as the system voltage level increases, the number of switches and the risk of failure increase exponentially, significantly raising maintenance costs. Summary of the Invention

[0003] To address the numerous problems existing in the prior art, this invention provides a WIPI wireless power dynamic balancing method for grid-connected energy storage systems. This invention predicts the future charge and temperature rise trends of batteries based on a fractional-order digital twin model. A graph neural network outputs a demand-oriented matrix, and a quantum optimizer outputs a loss-optimized matrix. These two matrices are fused with fixed weights to generate a power coupling target. The target is then used to control a 32×32 magneto-acoustic meta-array via threshold mapping and synchronously drive a GaN inverter, achieving uninterrupted energy redistribution. Real-time power and temperature are sampled at high resolution and fed into a Sinkhorn loop to quickly obtain the optimal transmission gradient and update pixels, forming a closed loop. This solution significantly reduces the charge difference between modules, suppresses bus inrush current, and mitigates inverter harmonics.

[0004] A method for dynamic WIPI wireless power balancing in grid-connected energy storage systems, the method comprising:

[0005] A digital twin model is established to integrate system parameters characterizing coil geometry, battery module electrochemical impedance, and coil coupling into a unified model, which is used to describe the electromagnetic topology and electrochemical state of the energy storage system.

[0006] Electrical and magnetic flux data are collected and synchronized. Fractional dynamic predictions are performed based on the digital twin model to obtain state of charge data, temperature constraint data, and control optimization initialization data.

[0007] Based on the prediction results and initialization data, power coupling control target data is generated, and magnetic-acoustic superstructure tuning data and execution control command data are generated accordingly.

[0008] The inverter is driven according to the execution control command data, and the magnetic-acoustic metastructure is tuned according to the magnetic-acoustic metastructure tuning data to achieve wireless power coupling. Power flow data and temperature data are measured in real time. The optimal transmission algorithm is used to calculate the coupling correction gradient to update the magnetic-acoustic metastructure. The power flow data, temperature data and execution control command data are formed into feedback data and returned to the power coupling control target data generation step, thereby realizing wireless power dynamic balance between battery modules within the closed loop.

[0009] Preferably, the system parameters consist of the following data: coil center coordinates, coil normal vector, coil effective area, battery module zero-frequency internal resistance, battery module polarization resistance, and coupling coefficients between each pair of coils.

[0010] Preferably, the digital twin model is established through the following steps:

[0011] In the electromagnetic part, the coil coupling impedance matrix is ​​calculated using a finite element mesh with a unit size of no more than 0.5 mm. In the electrochemical part, the battery module impedance curve is fitted using a fractional-order equivalent circuit containing constant-phase elements. With the goal of minimizing the mean square error between the model output and the measured impedance spectrum, a sparse identification kinetic algorithm with a penalty coefficient of is used to determine the fractional-order order and circuit parameters.

[0012] Preferably, the synchronization of electrical data and magnetic flux data is accomplished using a time-sensitive fiber optic network, with a synchronization jitter of no more than 50 nanoseconds.

[0013] Preferably, the order of the fractional-order dynamic prediction is a value between 0.80 and 0.90, which is determined by minimizing the mean square error of the state of charge within a 10-second prediction window.

[0014] Preferably, the power coupling control target data is generated by linearly fusing the coupling weight matrix output by the graph neural network and the optimal coupling matrix output by the variable quantum optimizer with weight coefficients of 0.60 and 0.40, respectively. The graph neural network adopts a 3-layer GraphSAGE structure with 256 hidden nodes in each layer, and the variable quantum optimizer uses 20 qubits and a parameterized quantum circuit with a depth of 6 layers.

[0015] Preferably, the magneto-acoustic superstructure tuning data is a 32-row, 32-column pixel matrix, wherein when the element value in the target data of the coupling matrix is ​​greater than 0.35, the corresponding pixel is set to the on state, and the remaining pixels are set to the off state, and the pixel switching completion time is no more than 60 microseconds.

[0016] Preferably, power flow data is acquired at a rate of 1,000,000 samples per second via an 18-bit analog-to-digital converter, and temperature data is acquired at a rate of 200,000 samples per second via a 16-bit analog-to-digital converter.

[0017] Preferably, the optimal transmission algorithm adopts the entropy-regularized Sinkhorn algorithm with an entropy regularization coefficient of 0.05 and 30 iterations. The iteration stops when the difference between the L2 norms of the power transport matrices obtained from two consecutive iterations is less than 0.

[0018] Preferably, after generating feedback data, the SHA-256 hash algorithm is used to calculate the hash value and write it into the blockchain storage. The blockchain node completes the transaction confirmation containing the hash value and the Coordinated Universal Time timestamp within 200 milliseconds.

[0019] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0020] This invention achieves instantaneous power routing without physical switching by shaping magnetic flux through a magneto-acoustic meta-device array, thus eliminating high-frequency switching stress.

[0021] This invention achieves forward prediction and global optimal allocation of charge-temperature rise-loss through fractional-order digital twin fusion optimization with GraphSAGE-VQO.

[0022] This invention achieves adaptive power flow correction within a 100Hz closed loop through rapid iteration of entropy regularization Sinkhorn.

[0023] This invention achieves millisecond-level trusted data traceability through blockchain hash storage, which facilitates operation, maintenance, and compliance. Attached Figure Description

[0024] Figure 1 This is a schematic flowchart of the method of the present invention;

[0025] Figure 2 This is a schematic diagram of the digital twin model in this invention;

[0026] Figure 3 This is a schematic diagram of the control optimization fusion architecture in this invention. Detailed Implementation

[0027] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0028] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0029] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0030] like Figure 1 As shown, a WIPI wireless power dynamic balancing method for grid-connected energy storage systems is described, the method comprising:

[0031] like Figure 2 As shown, a digital twin model is established to integrate the system parameters characterizing coil geometry, battery module electrochemical impedance and coil coupling into a unified model, which is used to describe the electromagnetic topology and electrochemical state of the energy storage system.

[0032] Coil-to-coil coupling systems include:

[0033] Coil: A planar induction coil mounted on the outside of each battery module. It can be a PCB spiral wire, Litz cable, or three-dimensional irregular winding. It is used as a wireless power transmission port and is not located inside the battery cell.

[0034] Coil geometric parameters: center coordinates, normal vector, effective area, number of turns, and compensation capacitance value, used to calculate self-inductance and radiation distribution.

[0035] Coil coupling system: consisting of mutual inductance M of all coils ij Coupling coefficient It consists of a compensation network and can be abstracted as a multi-port magnetic coupling transmission network, which determines the instantaneous flow of energy between modules and is the core of electromagnetic topology.

[0036] Electromagnetic topology-electrochemical state unified digital twin model, equivalent impedance matrix of external induction network:

[0037]

[0038] L represents the coil self-inductance matrix (H), M represents the coil mutual inductance matrix (H), C represents the compensation capacitor matrix (F), C -1 Its inverse matrix, R Cu Let ω represent the coil copper loss resistance vector (Ω), and ω represent the angular frequency (rad·s⁻¹).

[0039] Fractional-order equivalent electrochemical impedance of the battery module:

[0040]

[0041] R0 represents the zero-frequency internal resistance (Ω); R t Let represent the polarization resistance (Ω); τ represent the diffusion time constant (s); and j represent the imaginary unit, satisfying j 2 =-1, used to represent the phase information of AC impedance; α represents the fractional order (0<α≤1), together yielding a unified fractional order state equation:

[0042]

[0043] Describes the fractional derivative of Caputo form, order β.

[0044] x(t) represents the state vector, which includes: coil current i ij Module State of Charge (SOC) i polarization potential η i Module temperature T i Coupling coefficient κ ij .

[0045] A and B represent the system matrices obtained from online sparse identification, where A is the system matrix and B is the output matrix; u(t) represents the inverter drive voltage amplitude-phase vector.

[0046] A – The system matrix, which consists of four parts:

[0047]

[0048] A mm Describe the coupling between coil currents (by L, M, R) Cu (Derivation); A ee Describe the diffusion-polarization kinetics between electrochemical states (by R0, R t (Derivation of τ,α); Cross-block A me A em The thermo-electric feedback between current, temperature rise, and polarization is coupled together.

[0049] If the system contains N battery modules and N coupling coils, then:

[0050]

[0051] i(t) is the coil current vector of length N. SOC(t) is the state-of-charge vector of length N. η(t) is the polarization potential vector of length N. T(t) is the module temperature vector of length N. Mutual inductance κ ijIt is not presented as a separate state, but rather as a feature within A. mm and A me In the coefficients; when the mutual inductance of the magneto-acoustic pixel matrix changes, these two blocks are refreshed online synchronously.

[0052] This model simultaneously characterizes electromagnetic topology (via Z...) em ) and electrochemical state (via Z) ec (and fractional terms) to achieve integrated digital twin.

[0053] Electrochemical states can be deduced from electromagnetic and chemical impedance, including:

[0054] Online frequency sweep injection involves superimposing a small-amplitude frequency sweep signal onto the inverter outside its operating frequency to collect the total impedance Z. tot .

[0055] Magnetic deconvolution, using known Z em Numerical removal of coupling terms yields

[0056] Sparse identification dynamics, using SINDy-f to minimize the mean square error to obtain R0, R t ,τ,α, real-time updates of the electrochemical sub-model.

[0057] Fractional-order observers estimate the new parameters and write them into A and B. The SOC is then inverted online using a fractional-order Kalman-like observer. i ,η i ,T i .

[0058] This process can reconstruct the internal electrochemical state based on externally measurable voltage, current, and magnetic flux, eliminating the need for sensors inside the cell and ensuring safety and maintainability.

[0059] Coil geometry vector, C for each coil i Describe using the following triplet: center coordinates normal vector Equivalent cross-sectional area S i >0. Measurement method: Blue light 3D scanner, point cloud density 1000 points / mm. -2 The geometric reconstruction error is ≤0.05mm.

[0060] Electrochemical impedance parameters were obtained by collecting impedance spectra of each battery module in the 0.1Hz–5kHz frequency band and fitting them using a fractional-order equivalent circuit containing a constant-phase element (CPE).

[0061]

[0062] Among them, R 0,i R represents the zero-frequency internal resistance;t,i Indicates polarization resistance; τ i α represents the time constant; i Indicates the fractional order. Fitting method: Genetic algorithm initialization + least squares iteration; mean square error ≤ 4%.

[0063] The coil coupling coefficient was calculated using the finite element method (element side length 0.5mm) to obtain the mutual inductance matrix M, which was then obtained after actual measurement and calibration.

[0064]

[0065] Where L i With L j These are the respective inductance values. The relative error between the simulated and measured mutual inductance is ≤3%.

[0066] A unified state-space structure is adopted, using Caputo form and fractional derivatives starting at 0, to establish the state equations:

[0067]

[0068] y(t)=Cx(t)

[0069] in, This represents the Caputo fractional derivative operator with a starting point of 0 and an order of β. β represents the fractional order, with a measured range of 0.80–0.90. Comparisons show that β = 0.85 yields the best overall performance. x(t) represents the state vector, containing the state of charge (SOC) of each battery module. i Temperature T i and coil coupling coefficient κ ij u(t) represents the inverter drive voltage amplitude-phase vector, derived from the inverter output voltage amplitude V. k With phase φ k Composition. y(t) represents the output vector, which is composed of the DC bus current I. bus Bus power P bus With the maximum temperature difference ΔT max Composition: A represents the system matrix. B represents the output matrix, calculated from the coil topology and inverter interface parameters. C represents the output matrix, which is a set of row vectors selecting the corresponding state components. A, B, and C are determined using a sparse identification dynamics algorithm (penalty coefficient 1×10). -4 It updates online every 1 second.

[0070] Model calibration and verifiability, impedance spectrum fitting error:

[0071]

[0072] The mean square error of the state of charge in the ten-second rolling window is 0.9%, which is 42% lower than that of the integer-order model.

[0073] Example 1 was deployed in an energy storage station with a capacity of 1MWh and a bus voltage of 1200V. After importing data from 46 pairs of coils, an A-dimensional model of 94×94 was generated. After 24 hours of continuous operation, the standard deviation of the State of Charge (SOC) decreased from 3.5% to 1.4%.

[0074] The effect of adjusting the fractional order β in Example 2 based on Example 1 is shown in Table 1:

[0075] Table 1

[0076] β Prediction Mean Square Error Inverter peak current (A) 0.80 1.5％ 132 0.85 0.9％ 118 0.90 1.1％ 125

[0077] Example 3 reduces the number of hidden nodes in GraphSAGE from 256 to 64 compared to Example 1, increases the prediction error to 1.3%, and reduces memory usage by 62%.

[0078] The unified digital twin model of this invention controls the impedance spectrum fitting error to within 5% and the state-of-charge prediction error to ≤1%. Compared with the integer-order model, the prediction error is reduced by 42% and the inverter peak current is reduced by 11%. The online update cycle is 23ms, which meets the 1s control cycle requirement and verifies the real-time availability of the model.

[0079] Preferably, the system parameters consist of the following data: coil center coordinates, coil normal vector, coil effective area, battery module zero-frequency internal resistance, battery module polarization resistance, and coupling coefficients between each pair of coils.

[0080] This invention explicitly defines the system parameters as: coil center coordinate p i , coil normal vector n i Effective coil area S i Battery module zero-frequency internal resistance R 0,i Battery module polarization resistance value R t,i and coil-coil coupling coefficient k ij These parameters collectively drive the digital twin kernel for dynamic wireless power equalization. The specific acquisition, fusion, and verification process is as follows.

[0081] Electromagnetic geometric parameters, coordinates, and normal vectors were obtained. A blue light 3D scanner (0.02mm accuracy) was used to collect point clouds of all coils within the busbar slot. The center p of the coil was reconstructed using Poisson-Surface. i With normal vector n i Critical value test: When the point cloud density drops to 200 points / mm -2 At that time, the reconstruction error increased to 0.18 mm, which led to a 6% increase in the downstream mutual inductance calculation error, verifying the necessity of high-density scanning.

[0082] The effective area is calculated by truncating the scanned mesh and projecting it onto the cross-sectional normal vector:

[0083]

[0084] Where r is the cross-sectional boundary vector. The area error is better than 0.6%. The closed boundary curve of the i-th grid cell in the cross section, which is composed of the lines connecting the sampling points, lies in the cross section plane, and the curve direction is conventionally counterclockwise; dr represents the direction along... The differential displacement vector is in the same direction as the tangent vector of the curve, and its length tends to zero;

[0085] Electrochemical impedance spectroscopy parameters, wideband EIS sampling: 0.1Hz–5kHz, amplitude 5mA, four-point probe to eliminate wire impedance. The rapid fitting process includes:

[0086] 1. Generate candidate [R] using a genetic algorithm 0,i ,R t,i ,τ i ,α i ].

[0087] 2. Least squares iteration converges after 20 rounds, mean square error:

[0088]

[0089] Keep it below 0.04.

[0090] 3. Online update cycle: Quick EIS (6s frequency sweep) calibration every 30 minutes to resolve temperature-lifetime drift.

[0091] Coupling coefficient κ ij Finite element mutual inductance matrix:

[0092]

[0093] Φ ij C j Flow path L j At C i The magnetic flux of the link on I. j Indicates coil C j The excitation current (A) in the figure is the reference current amplitude used to calculate the mutual inductance.

[0094] The simulation-experiment closed loop uses 46 pairs of coils as a benchmark, and the average deviation between the measured mutual inductance and the simulated value is 3%. If κ... ij If the error exceeds 5%, the system triggers adaptive magnetic field shaping at the control layer to ensure coupling accuracy.

[0095] A unified mapping to a fractional-order state space, with the following core relationships:

[0096]

[0097] y(t)=Cx(t))

[0098] This represents the Caputo fractional derivative with a starting point of 0 and an order of β.

[0099] x(t) represents the state vector, which includes: coil current i ij Module State of Charge (SOC) i polarization potential η i Module temperature T i Coupling coefficient κ ij ;

[0100] u(t) includes the inverter output amplitude V k With phase φ k ;

[0101] y(t) includes bus current I bus Power P bus Maximum temperature difference ΔT max ;

[0102] A, B, and C are analyzed on a GPU using a sparse identification dynamics algorithm (penalty factor 1×10). -4 Updated every 1 second.

[0103] Quantitative verification of the technical effects: After 24 hours of deployment at a 1MWh power plant, the SOC standard deviation decreased from 3.5% to 1.4%. Compared to the linear integer-order model of wireless power equalization, the prediction error was reduced by 42%, and the inverter peak current was reduced by 11%. Online updates took 23ms, meeting the 1s control cycle and leaving a 40% computational margin.

[0104] Example 1 (Basic Example): All system parameters are obtained according to the above process, and A dimension 94×94 is generated to achieve the goals of charge balance and thermal balance.

[0105] Example 2 (Extended Example: Change in the Number of Hidden Nodes): Based on Example 1, the number of hidden nodes in GraphSAGE is reduced from 256 to 64. The prediction error is improved to 1.3%, and the memory usage is reduced by 62%, demonstrating that the number of hidden nodes can be flexibly adjusted according to the computing power of edge devices.

[0106] Preferably, the digital twin model is established through the following steps:

[0107] In the electromagnetic part, the coil coupling impedance matrix is ​​calculated using a finite element mesh with a unit size of no more than 0.5 mm. In the electrochemical part, the battery module impedance curve is fitted using a fractional-order equivalent circuit containing constant-phase elements. With the goal of minimizing the mean square error between the model output and the measured impedance spectrum, a sparse identification kinetic algorithm with a penalty coefficient of is used to determine the fractional-order order and circuit parameters.

[0108] The coil coupling impedance matrix was calculated by establishing a three-dimensional finite element model for all coils on both the bus side and the module side, with a mesh element side length of 0.5 mm. The mutual inductance matrix M could be solved in a single run on an Nvidia A40 GPU in 28 seconds. Critical value tests showed that when the element side length increased to 1 mm, the mutual inductance error rose from 3% to 7%, and the inverter phase compensation increased by 4°, proving that 0.5 mm is a feasible upper limit.

[0109] In the coil resistance R i Mutual inductance M ij Given the given conditions, the coupling impedance matrix can be written as:

[0110]

[0111] L represents the self-inductance diagonal matrix, and C represents the compensation capacitor diagonal matrix. The calculation results serve as the known input for subsequent sparse identification.

[0112] Fractional-order equivalent circuit, impedance spectrum sampling frequency band 0.1Hz–5kHz, amplitude 5mA. The original curves are filtered by Savitzky–Golay filtering (11-point window, polynomial order 3) to suppress measurement noise. R0||R t -CPE||Topology, impedance expression:

[0113]

[0114] Where R0 is the zero-frequency internal resistance, R t Let τ be the polarization resistance, τ be the time constant, and α be the fractional order.

[0115] Sparse identification dynamics algorithm (SINDy-f), objective function:

[0116]

[0117] Let λ be the parameter vector to be identified, and λ = 1 × 10 -4 This is the sparsity penalty coefficient.

[0118] Initially, 50 candidate Θ values ​​were generated using a differential evolution algorithm. The L-BFGS algorithm was iterated for 40 rounds on a GPU, with 8 frequency data points in each round for gradient batching; the average convergence time was 3.1 seconds. Constraints were set: 0 < α < 1, R0 t >0, τ>0.

[0119] Model coupling and online refreshing, after identification is completed, Z... em With Z ecThe merged mapping is applied to a fractional-order state matrix A. This matrix is ​​updated online with temperature drift and aging, with a refresh cycle of 1 second. A single update takes 23 ms on an RTX A5000 GPU, meeting the 1-second scheduling window requirement of the control layer. The implementation results are shown in Table 2.

[0120] Table 2

[0121] index This invention No fractional order comparison <![CDATA[Impedance spectrum fitting error ε Z > ≤4％ ≈9％ 10-second state of charge mean square error 0.9％ 1.6％ Inverter peak current 118A 132A

[0122] Technical effect verification: The measured data comes from a 1MWh prototype, proving that after fractional modeling and sparse identification are combined, the prediction accuracy is improved by 44% and the device stress is reduced by 11%, which directly translates into narrower bus voltage ripple and lower switching loss.

[0123] Basic Example: Running the above three steps with 46 coil pairs and a 0.5mm grid yielded α = 0.85. The SOC standard deviation decreased from 3.5% to 1.4%.

[0124] Extended Example (1mm Mesh Size): Widen the mesh side length to 1mm, ε Z The error rose to 6.8%, the SOC error increased to 1.2%, and the peak current increased by 7%. This indicates that a fine mesh is key to ensuring the accuracy of the equalization process.

[0125] Through a hierarchical modeling process of finite element-CPE-SINDy-f, this invention achieves accurate digital mapping of electromagnetic-electrochemical coupling behavior in the WIPI wireless power dynamic equalization application. It can be stably refreshed on-site at the second-level clock speed, supporting subsequent fractional-order prediction and real-time closed-loop operation of the controller.

[0126] Electrical and magnetic flux data are collected and synchronized. Fractional dynamic predictions are performed based on the digital twin model to obtain state of charge data, temperature constraint data, and control optimization initialization data.

[0127] The sampled signal consists of:

[0128] 1. Electrical data:

[0129] Module terminal voltage V i With current I i 18-bit ADC, 1MS / s -1 .

[0130] Bus voltage V bus With bus current I bus Similarly, 1MS / s -1 This reflects station-level load disturbances. A swept-frequency injected signal δV(t) is used to superimpose a small-amplitude sinusoidal sequence onto the PWM sideband for online extraction of electrochemical impedance spectroscopy.

[0131] These electrical quantities are timestamped before being used to calculate the instantaneous power P. i =V i I i And generate frequency domain impedance curves to provide data for updating R0 and R t ,τ,α.

[0132] 2. Magnetic flux data:

[0133] Triaxial magnetic induction intensity B i =[B x,i B y,i B z,i ]; Tunneling magnetoresistive array 10kS / s -1 Sampling. Data is used for online inversion of coil mutual inductance M. ij The calibration of the coupling submatrix M ensures that the electromagnetic topology adapts to temperature drift and mechanical vibration.

[0134] The fractional-order dynamic prediction model expression is:

[0135]

[0136] Let β represent the Caputo fractional derivative, with order β = 0.85.

[0137] x(t) represents the state vector, which includes: the instantaneous coil current i ij Module State of Charge (SOC) i Polarization potential η i Module temperature T i .

[0138] A represents the system matrix, consisting of coil geometry L, mutual inductance M, and copper loss R. Cu and electrochemical impedance parameters R0, R t The combination of τ and α. B represents the input matrix, corresponding to the inverter drive voltage vector u(t). u(t) represents the amplitude-phase output of each GaN inverter.

[0139] The derivation process of the state of charge and temperature constraints includes:

[0140] The inverter injects δV(t) into the broadband band and samples V. i ,I i The total impedance Z is obtained tot .

[0141] From Z tot Subtracting the known Z from the middle em residual R0 and R are obtained by SINDy-f. t ,τ,α.

[0142] Write the new parameters into A, and use a fractional-order Kalman observer to fuse the instantaneous Coulomb integral ∫I. i dt, outputs the corrected SOC i .

[0143] Current power P i With the heat network model (C th ,R th By combining the predictions of temperature rise, a temperature constraint vector can be obtained.

[0144] Control optimization initialization includes: SOC with a prediction window length of 10s. i and Concatenate into an initialization vector The second-order approximation H0 is obtained by the automatic differentiation of the Jacobian mentioned above, providing a warm start for the GNN-VQO hybrid optimizer, enabling the power coupling objective to converge within a 0.1s period.

[0145] This invention clarifies, on the one hand, the specific sources of "electrical data" and "magnetic flux data" and their uses in the model; on the other hand, it explains how fractional-order digital twins can use this data to perform online inversion of SOC. i This, along with temperature rise, supports closed-loop control for dynamic equalization of wireless power.

[0146] To achieve second-level dynamic wireless power balancing in grid-connected operation of the WIPI energy storage system, this invention sets up a "synchronous observation-fractional prediction" layer at the control front end. This layer estimates the charge evolution curve and upper temperature rise limit of each module 10 seconds in advance through precisely timed electrical-magnetic flux data streams and the digital twin model constructed in the previous section, and compresses the results into initialization quantities that can be directly used by the optimizer.

[0147] The synchronous observation principle involves using a Time-Sensitive Fiber Network (TSN) to uniformly stamp electrical variables (bus voltage, current, and 46-channel module terminal voltage) and magnetic flux variables (46-channel triaxial magnetic flux density) with a 64-bit PTP timestamp at the hardware layer. A 40ns jitter ensures alignment of the two types of signals within the same time step, providing in-phase input for subsequent coupled prediction.

[0148] The fractional-order kinetic principle is used, and the model describes the multi-domain coupling of "coil coupling-electrochemical impedance-thermal diffusion" using a Caputo fractional-order framework with order β = 0.85. β is minimized by scanning a grid of 0.80–0.90, while the peak inverter current can be suppressed to within 120A.

[0149] Based on the principle of temperature constraint, the thermal behavior of the module is simplified to a first-order RC network, with real-time I... 2 R-loss and switching loss drive the temperature node; if the predicted temperature in a 10s window will exceed 55℃, then the temperature weight w is adjusted. T=4 Write control vectors to prioritize heat dissipation.

[0150] The initialization principle involves concatenating the endpoint SOC of the prediction layer output with the upper temperature limit into a 92-dimensional vector, along with a Hessian approximated by the second-order gradient; together, these two serve as a "hot start" for the graph neuron-quantum hybrid optimizer, significantly shortening the number of convergence steps.

[0151] In practical applications, during sampling and synchronization, the DC bus signal is measured using an 18-bit ADC at a rate of 1 MS / s. -1 The flux array uses 10 kS / s -1 All data were resampled to a step size of 0.1s after TSN time synchronization; experiments showed that if the step size was increased to 0.2s, the charge prediction error increased by 0.4%, so 0.1s was fixed as a compromise value.

[0152] Signal purification was performed, using a three-level Daubechies-4 wavelet to retain the second and third-level low-frequency subbands, followed by independent component analysis to remove the 20kHz switching harmonics. Noise power spectral density was reduced from 6 × 10⁻⁶. -9 Reduced to 8×10 -11 V 2 / Hz.

[0153] Fractional numerical progression:

[0154] x k+1 =x k +h 0.85 (Ax k +Bu k )

[0155] h = 0.1s

[0156] Where, x k For the state vector (46 SOCs, 46 temperatures, 1035 coupling coefficients), u k The inverter amplitude-phase vectors A and B are recalculated every second by the digital twin. The RTX A5000 GPU performs 128 batches in parallel, with a 10-step prediction cycle taking 4.2ms.

[0157] Temperature updates, per module in terms of heat capacity C th =4.1kJ K -1 Thermal resistance Rth = 0.23KW -1 Estimate the temperature rise; if the upper limit of the prediction exceeds 55℃, then write it to w. T .

[0158] Initialize the packaging; at step 10, the 46 channels of SOC and the 46 channels of temperature upper limits form χ0, and simultaneously calculate the second-order gradient approximation H0; these two are then compared with w. TBy jointly feeding the control optimizer, the number of iterations was reduced from 52 to 32. The implementation results are shown in Table 3:

[0159] Table 3

[0160] index This invention Integer-order Kalman-SOC 10s rolling forecast MSE 0.9％ 1.6％ Temperature warning lead time 0.8s 0.1s Inverter peak current 118A 132A

[0161] Preferably, the synchronization of electrical data and magnetic flux data is accomplished using a time-sensitive fiber optic network, with a synchronization jitter of no more than 50 nanoseconds.

[0162] In the WIPI wireless power dynamic equalization closed loop, electrical quantities (voltage, current) and magnetic flux (tunnel magnetoresistive triaxial magnetic induction) must be written to the predictor at the "same physical instant." If the time alignment error exceeds 100 nanoseconds, the fractional-order differential kernel will amplify the phase shift error into charge prediction drift. To address this, this invention employs a Time-Sensitive Fiber Network (TSN-FN) to achieve cross-domain synchronization and reduces end-side jitter to within 50 ns.

[0163] TSN-FN uses the IEEE 802.1AS-rev precision clock as its backbone: a single OCXO-stabilized master clock node sends Sync / Follow-Up frames to 12 boundary switching nodes, forming a hierarchical clock tree. Both the electrical measurement board (ADCFPGA) and the flux measurement board (TMRFPGA) embed a hardware timestamp unit HTS-64, which immediately writes a 64-bit PTP timestamp into the sampling FIFO upon receiving a Sync frame, achieving "sampling-time-binding in one step." Because HTS-64 performs timestamp injection at the PHY-MAC boundary without going through the operating system stack, single-hop jitter is limited to within 8ns; the theoretical upper limit of the five-hop link's time base error is... Component information is shown in Table 4.

[0164] Table 4

[0165]

[0166] Time step resampling—all channels are first buffered in a circular buffer, and then sent in packets by FPGA DMA with a fixed step size of 0.1s. If the time standard difference between two consecutive samples exceeds 50ns, a jitter satellite window is triggered, the predictor pauses and waits for the next frame to align.

[0167] Absolute jitter test: The electrical-magnetic flux trigger signal was compared with the dual-channel oscilloscope. The 24-hour statistical results showed that the RMS jitter was 27ns and the peak-to-peak jitter was 48ns, which meets the upper limit of 50ns.

[0168] The impact on prediction accuracy was assessed by manually widening the synchronization jitter to 200ns: the mean square error of the 10s charge prediction increased from 0.9% to 1.3%, and the inverter peak current increased by 6A; the indicators fell back after 50ns of recovery, proving the necessity of strictly controlling jitter.

[0169] Regarding the impact on grid connection stability, a 5kW step discharge test was conducted at a 1MWh power station: the precise synchronization version triggered a temperature warning 0.8s in advance, with a peak bus inrush current of 191A; the wide jitter version reduced the lead time to 0.2s, with the peak value rising to 208A.

[0170] Example 1 (Basic): Complete TSN-FN configuration, jitter 27ns. After 24 hours of continuous operation, the SOC variance decreased from 3.5% to 1.4%, consistent with the main scheme's performance.

[0171] Example 2 (Control: Gigabit copper cable replaced with end fiber): The single-hop jitter of the copper cable link was 65ns, and the total jitter was about 180ns. The SOC variance was reduced to only 1.9%, and the temperature over-limit warning was shortened to 0.3s, confirming that the fiber-to-the-level symmetrical link is the key to ensuring the accuracy of dynamic equalization.

[0172] Upper bound of synchronization error:

[0173]

[0174] Where σ l This indicates the jitter at the l-th physical layer, where N=5 represents the deepest level in the field. This upper bound is directly written into the predictor's fault tolerance threshold; if the limit is exceeded, the system enters a safety buffer mode.

[0175] Preferably, the order of the fractional-order dynamic prediction is a value between 0.80 and 0.90, which is determined by minimizing the mean square error of the state of charge within a 10-second prediction window.

[0176] In the fractional-order dynamics framework, the order β determines the rate at which the model absorbs historical information. A value that is too low will ignore the slow accumulation of battery heat, while a value that is too high will amplify second-level fluctuations into prediction noise. For WIPI wireless power dynamic equalization, this invention limits β to between 0.80 and 0.90, and automatically determines the optimal order β by minimizing the mean square error of the state of charge using a ten-second rolling window. opt The three-point golden ratio iteration is used, with initial values ​​set to β1 = 0.80, β2 = 0.90, and β3 = 0.85. Two iterations are sufficient to converge the search interval to 0.01.

[0177] The evaluation metrics within the ten-second prediction window are:

[0178]

[0179] in SOC represents the predicted value at order β. i,meas For synchronous observations, N is the number of samples within the window. The β that minimizes the MSE is taken as β. opt .

[0180] Data and calculation flow: Electrical quantities are processed by an 18-bit ADC at 1 MS / s. -1 Sampling, magnetic flux passes through the tunnel magnetoresistive array at 10 kS / s -1 Sampling. Both types of data were written with 64-bit PTP timestamps via a time-sensitive fiber optic network and resampled to a 0.1s step size. The synchronization sequence was first subjected to three-level Daubechies-4 wavelet denoising, and then 20kHz switching harmonics were removed using independent component analysis. Twenty-one candidate betas were computed in parallel using GPUs, with a complete search taking 6.5ms.

[0181] If the current iteration yields β opt If the value is less than 0.83, the system will increase the temperature weighting. T Half-throttle, to counteract thermal constraint relaxation caused by insufficient memory depth. Full calibration triggers once every five seconds, during which the most recent β value is used. opt Perform fractional-order prediction.

[0182] Using an adaptive order, the mean square error of the 10-second rolling prediction reached 0.9%, while the fixed integer order Kalman-SOC method achieved 1.6%. The inverter peak current decreased from 132A to 118A. After 24 hours of continuous operation of a single 1MWh prototype, the module SOC variance decreased from 3.5% to 1.4%, and the peak bus inrush current was reduced from 215A to 191A during a 5kW step discharge event.

[0183] Basic implementation: The adaptive order process is fully executed to achieve the above-mentioned accuracy and current suppression effect.

[0184] Extended Example: Fixed order β = 0.80. GPU load decreased by 12%, but SOC variance only decreased to 1.8%, and peak current increased by 10A, indicating that dynamically selecting the order is crucial for balancing system performance.

[0185] like Figure 3 As shown, power coupling control target data is generated based on the prediction results and initialization data, and magnetic-acoustic superstructure tuning data and execution control command data are generated accordingly.

[0186] The prediction layer provides a "charge state vector" every 0.1s. "and temperature upper bound vector" "Together, they characterize the future needs of 46 battery modules. The controller first uses a graph neural network (GNN) to map these two sets of information into a 'power coupling weight matrix' W." G Subsequently, a set of "minimum switching loss matrices" W is searched within the feasible region using a variational quantum optimizer (VQO). Q The two matrices are combined into a single objective via a dynamic weight fusion module.

[0187] W T=αW G +(1-α)W Q ,α∈[0.6,0.7]

[0188] Where α is determined by the temperature weight W T Adaptive decision; W T The higher the value, the larger α becomes, thus increasing the priority of thermal equilibrium. Target matrix W T The data is quantized into a 32×32 pixel “magnetic-acoustic meta-device tuning diagram” and translated into an inverter control command sequence (duty cycle, phase, frequency, compensation capacitor).

[0189] In practical applications, the data input and the prediction layer output are... (46 dimensions) and (46-dimensional), the initialization vector χ0 and HessianH0 are entered into the control module.

[0190] The graph neural network uses a three-layer GraphSAGE architecture, with 256 hidden nodes per layer and ReLU activation. The learning rate is 0.001 (below 0.001, the convergence speed decreases by 48%; above 0.01, gradient oscillations occur). The GNN output is coupled with a weight matrix W. G Inference was performed in TensorRT with a latency of 2.3ms.

[0191] A variable quantum optimizer was used, and the quantum hardware simulation employed a 20-qubit, six-layer parameterized circuit. The optimization objective was to minimize switching losses and inverter harmonic distortion. W was obtained through 32 iterations using Adam 0.01. Q It took 1.8ms.

[0192] The dynamic weight fusion module reads the temperature weight w. T ∈{1,2,3,4}, and perform a linear mapping on α: α = 0.55 + 0.05w T If the risk of temperature rise is eliminated in the next cycle, alpha will automatically fall back, ensuring steady-state efficiency.

[0193] Pixel matrix mapping, W T Normalization and thresholding (threshold 0.35) yield a 32×32 pixel matrix. Pixel representation is enabled to activate the magneto-acoustic column unit. The matrix is ​​uploaded to the FPGA LVDS bus to complete the meta-device tuning within 60μs.

[0194] Inverter command generation, based on W T The row and column sums are used to calculate the 46-channel amplitude-phase table, and then the duty cycle is finely adjusted using the single-iteration Newton method in conjunction with HessianH0; the complete instruction frame size is 512B, which is sent to the GaN matrix inverter via Aurora4-lane.

[0195] Preferably, the power coupling control target data is generated by linearly fusing the coupling weight matrix output by the graph neural network and the optimal coupling matrix output by the variable quantum optimizer with weight coefficients of 0.60 and 0.40, respectively. The graph neural network adopts a 3-layer GraphSAGE structure with 256 hidden nodes in each layer, and the variable quantum optimizer uses 20 qubits and a parameterized quantum circuit with a depth of 6 layers.

[0196] Power coupling control target matrix W T It simultaneously undertakes two tasks: first, allocating the "power flow direction and amplitude" among the 46 pairs of coupled coils; and second, taking into account the inverter's own switching losses. To balance these two aspects within a single 0.1s cycle, this invention entrusts the "battery-side demand" to a data-driven graph neural network (GNN) for processing, and the "switching loss compression" to a variable quantum optimizer (VQO) with stronger near-global solution capabilities, achieving unified scheduling through linear fusion.

[0197] The GNN branch takes 46 battery modules as nodes as input graphs, and the edge weights are derived from the instantaneous coupling coefficient κ of the digital twin output. ij A three-layer GraphSAGE model with 256 hidden nodes per layer and ReLU activation function was used. The model was trained offline on 120 hours of historical data with a learning rate of 0.001 and a batch size of 256. After training, the weights were frozen and TensorRT was used for inference, with a single-frame latency of 2.3ms. The output matrix is ​​denoted as W. G The numerical range is compressed to 0–1 by the Sigmoid function; the larger the value, the more power should be tilted towards the coupling channel.

[0198] To compensate for the shortcomings of GNNs in switching harmonic suppression, the second branch uses a 20-qubit, depth 6 parameterized quantum circuit (hardware is an IonQ simulator). The objective function is a weighted sum of the inverter's conduction loss, cutoff loss, and total harmonic distortion coefficient. Optimization using Adam 0.01 for 32 steps yields an approximately globally optimal matrix W. Q The runtime latency is 1.8ms.

[0199] The dynamic weighted fusion module uses temperature weights W. T The fusion coefficient α is dynamically adjusted within the range {1,2,3,4}. When there is a risk of high temperature (w... T When α = 4), it rises to 0.70, making the data-driven part dominant; when the thermal risk is eliminated (w T When α = 1, α decreases to 0.55 to reduce switching losses. Core fusion formula:

[0200] W T =αW G +(1-α)W Q ,α∈[0.55,0.70]

[0201] Symbol explanation: W T Let W be the target matrix. G W is the output matrix of the GNN. Q This is the VQO output matrix, where α is an adjustable fusion coefficient. The default setting is α = 0.60.

[0202] Magnetoacoustic superstructure tuning and command generation, W T Normalized and thresholded (threshold 0.35), the matrix is ​​mapped to a 32×32 pixel matrix; enabling a pixel indicates that the corresponding magnetoacoustic column unit is activated. The matrix is ​​sent to the meta-device FPGA via LVDS within 60μs. The inverter controller determines the appropriate size based on W. T The rows and columns are generated to form a 46-channel amplitude-phase table. The duty cycle is then fine-tuned using a single-step Newton iteration combined with the Hessian provided by the prediction layer to generate a 512B instruction frame, which is then sent to the GaN power board.

[0203] The above GNN+VQO fusion was used to operate continuously for 24 hours at a 1MWh power station.

[0204] The mean square error of the 10-second rolling charge prediction is 0.9%, which is an improvement over both GNN (1.1%) and VQO (1.4%).

[0205] The peak inrush current of the bus is 191A, which is 11% lower than that of the present invention (215A) with a fixed coupling matrix.

[0206] The total harmonic distortion (THD) coefficient was reduced to 3.7%, a 26% improvement over pure GNN.

[0207] Basic implementation: When run with the above parameters (α = 0.60, GraphSAGE 256 nodes, 20 qubits, 6-layer circuitry), the SOC variance decreased from 3.5% to 1.4%.

[0208] Extended Implementation: Reducing the number of hidden nodes to 64 shortens the inference latency by 1.1ms, but the SOC variance only drops to 1.8%, while the peak inrush current increases by 7A, indicating that 256 nodes is the trade-off point between accuracy and computing power.

[0209] By using the above details, the entire process of generating power coupling control targets in real time can be directly reproduced, and the advantages of linear fusion of GNN and VQO in WIPI wireless power dynamic equalization in terms of accuracy, loss and thermal safety can be verified.

[0210] Preferably, the magneto-acoustic superstructure tuning data is a 32-row, 32-column pixel matrix, wherein when the element value in the target data of the coupling matrix is ​​greater than 0.35, the corresponding pixel is set to the on state, and the remaining pixels are set to the off state, and the pixel switching completion time is no more than 60 microseconds.

[0211] The magneto-acoustic meta-device is equivalent to a programmable magnetic field "grating". The system couples the power to the target matrix W. T First, it is normalized to 0–1, then mapped to 32×32 on-chip pixels. Enabling a pixel drives the corresponding magneto-acoustic micropillar to generate a localized "acoustic-magnetic" effect at the 45kHz resonant point, causing real-time shaping of the coil flux distribution. To balance accuracy and hardware lifespan, this invention uses a single threshold triggering strategy: W T Elements greater than 0.35 are marked as 1, otherwise marked as 0; the pixel array completes one write within 60μs, satisfying the 100Hz control cycle.

[0212] The pixel generation process includes: the power coupling controller calculating W T Then, the FPGA performs a step-by-step comparison:

[0213]

[0214] Symbol meaning: P m,n W is a pixel matrix element. T (m,n) represents the corresponding element of the target matrix. The resulting binary matrix is ​​packaged into a 1024-bit LVDS frame and sent in parallel to the superstructure drive array through four 250MHz channels; the pixel excitation level is 3.3V, the drive power is 180mW, and the total switching power consumption is less than 11μJ / frame.

[0215] A 60μs switching guarantee is provided, with the switching latency consisting of two parts: 4ns for LVDS transmission, 12ns for FPGA-SAW driver decoding, and 44μs for micropillar electromechanical response; a 16μs margin is provided for error detection and retransmission. If the CRC check fails, the controller rolls back the settings of the previous frame and retransmits in the next cycle to ensure uninterrupted connection.

[0216] In actual testing, a pixel threshold of 0.35 reduced the peak inrush current to 191A under a 5kW load step change. If the threshold was adjusted to 0.50, the number of active pixels decreased by 28%, but the peak inrush current rose back to 205A. While a threshold below 0.25 could further reduce the peak current by 3A, it increased the pixel switching power by 15%. Therefore, a compromise of 0.35 was chosen. After 24 hours of continuous operation, the cumulative pixel switching was 1.4 × 10⁵⁵⁵ times, and the micropillar resonant frequency shift was less than 0.2%.

[0217] Basic Implementation: Using a 32×32 threshold matrix with a 60μs update, the SOC variance was reduced to 1.4%, and the peak current was 191A.

[0218] Extended Implementation: The array resolution was reduced to 16×16, and the number of pixel drive I / Os was reduced simultaneously. The SOC variance reduction effect was maintained, but the peak current only dropped to 200A, indicating that 32×32 resolution is more advantageous for high-power pulsating scenarios.

[0219] The inverter is driven according to the execution control command data, and the magnetic-acoustic metastructure is tuned according to the magnetic-acoustic metastructure tuning data to achieve wireless power coupling. Power flow data and temperature data are measured in real time. The optimal transmission algorithm is used to calculate the coupling correction gradient to update the magnetic-acoustic metastructure. The power flow data, temperature data and execution control command data are formed into feedback data and returned to the power coupling control target data generation step, thereby realizing wireless power dynamic balance between battery modules within the closed loop.

[0220] The magneto-acoustic metasurface (MAM) consists of 32×32 independent micropillar units, each containing a piezoelectric thin-film transducer layer with a driving voltage denoted as V. pz Magnetostrictive coatings, whose volumetric magnetic susceptibility is denoted as Flexible support substrate.

[0221] When the thin film is excited by a 45kHz surface acoustic wave, the magnetostrictive layer generates micron-level strain, effectively changing the resonant frequency ω0 of the unit cell. In the wireless power band (50kHz-150kHz) far below the electro-ion oscillation frequency, the device can be described as an anisotropic permeability tensor:

[0222]

[0223] μ0 represents the free permeability; F represents the resonance intensity factor, and... Proportional; ω0 represents the mechanical-magnetic coupling resonant frequency of the unit, which can be obtained from V pz Tuning; γ represents the damping coefficient; ω represents the operating angular frequency of the coupling coil.

[0224] The above expression shows that changing V pz It can quickly offset ω0, thereby modifying μ at the operating frequency. eff When μ eff Decreasing μ causes the unit to scatter through magnetic field lines, effectively "closing" the pixel channel; when μ eff Increase the magnetic flux, focus the magnetic flux, and "open" the channel.

[0225] The wireless electromagnetic propagation is transformed into dynamic equalization between modules. The external induction coil of the energy storage system and the MAM together form a "source-lens-receiver" magnetic circuit. 32×32 patterned on-off pixels form a spatial magnetic lens, enabling the redistribution of coil mutual inductance M within milliseconds. ij : Open channel → Increase M ij →Increase coupling power; close the channel for the opposite effect. Using the mutual inductance gain matrix ΔM(P) corresponding to the pixel matrix P, we can simultaneously establish the total mutual inductance:

[0226] M new =Mbase +ΔM(P)

[0227] It can change the direction and amplitude of power flow in real time to achieve wireless equalization without the need for a physical switch.

[0228] M base ΔM represents the mutual inductance matrix without tuning; ΔM represents the mutual inductance compensation matrix determined by the pixel state.

[0229] Control link and performance, the magneto-acoustic pixel map is obtained by threshold mapping from the power coupling target matrix: when the target element W T When (m,n)>0.35, the pixel setting is enabled, corresponding to μ. eff Increase; otherwise, turn off. The FPGA drives the 32×32 piezoelectric pin via LVDS within 60μs, shifting ω0 by 400Hz, which is sufficient to make μ... eff After the operating frequency band is changed by 35% and the expression of the coil mutual inductance is updated, the Sinkhorn optimal transmission algorithm recalculates the gradient to determine the next frame pixel, forming a 100Hz closed loop.

[0230] Unlike existing wireless electromagnetic propagation technologies, which use MAMs for antenna beamforming or near-field communication and only modulate the direction of electromagnetic radiation, this invention is innovative in the following ways: 1. Operating scale: frequency band of 50-150kHz, targeting inductive power coupling rather than GHz antennas; 2. Coupling target: tuning mutual inductance M... ij Instead of radiation gain, it directly affects the power flow matrix. 3. Control link: Combined with digital twin - Sinkhorn closed loop, it realizes gradient update of each frame pixel, making the wireless equalization accuracy better than the traditional slice switch scheme.

[0231] The execution of control commands first drives 46 GaN three-level inverters, ensuring each pair of coils receives a specified amplitude and phase. Simultaneously, the magneto-acoustic meta-structure switches micropillar units at a 45kHz resonant frequency based on a 32×32 pixel matrix, shaping the transient magnetic field distribution. The system acquires two types of feedback in real time: one is the bus power flow matrix P... m (The module-side current is calculated using a 1MHz ADC), and the second is the module temperature vector T. m (Thermocouple array 200kS / s) -1 (Sampling). The two sets of data are packaged with the current instruction frame into a feedback vector and sent back to the control target generation layer.

[0232] To rapidly correct the field distribution within a 0.1s period, this invention will "transfer the desired power flow P..." ref The problem of "transferring to the module" can be expressed as entropy-regularized optimal transmission:

[0233]

[0234] Here, C is the coupling loss cost matrix, H(W) is the entropy term, and ε = 0.05. The gradient can be obtained by using a Sinkhorn iteration of 15 steps.

[0235]

[0236] Then update the entries in the pixel matrix that need adjustment at a learning rate of 0.12. Each round of gradient-pixel mapping takes 38μs, which meets the 60μs switching window of the meta-device.

[0237] In practical applications, the inverter command frame size is 512B, delivered to the GaN board within 3μs via the Aurora-4lane; the FPGA side uses single-step Newton iteration to reduce the duty cycle error to ±0.3%. The pixel matrix is ​​written to the SAW-FPGA via the LVDS parallel port, triggering the micropillar piezoelectric element with a 3.3V drive level, with a response time of 44μs. The power flow matrix is ​​calculated by the Hall current probe and the 32-bit MAC unit inside the FPGA; the temperature vector is filtered by a third-order IIR low-pass filter to suppress measurement noise. All feedback data is sent back to the control server in 4kB UDP frames, with a network round trip time of 110μs.

[0238] Preferably, power flow data is acquired at a rate of 1,000,000 samples per second via an 18-bit analog-to-digital converter, and temperature data is acquired at a rate of 200,000 samples per second via a 16-bit analog-to-digital converter.

[0239] The power waveform of the wireless coupling system exhibits its strongest fundamental frequency around 20kHz, with switching harmonics extending up to 200kHz. According to the Nyquist criterion, at least 400kS / s is required. -1 Only with sufficient sampling bandwidth can the instantaneous power be fully reconstructed. This invention uses an 18-bit analog-to-digital converter (TIADS8689) to acquire the bus voltage v[n] and current i[n] in parallel, with a single channel sampling rate of 1 MS / s. -1 It provides a 25dB jitter margin at an oversampling factor of 200kHz×5; the 18-bit effective bits guarantee a 68dB signal-to-noise ratio even at a current limit of 0.2A. Instantaneous power is controlled within the FPGA as follows:

[0240] p[n] = v[n]·i[n]

[0241] The calculated frequency is then down-converted to 100Hz within a 0.1s window using a CIC-4 level filter to match the control beat.

[0242] The temperature channel is limited by the thermocouple's own 10kHz thermal response bandwidth, and a 16-bit ADC (ADIAD7689) with a speed of 200kS / s is used. -1 Sampling is sufficient to meet oversampling requirements. Resolution conversion: Input noise 5μV rmsWith a temperature quantization step of ±0.012K, it is sufficient to detect a 2K temperature rise 0.8s in advance.

[0243] The hardware and software work in tandem. The voltage and current probes are differential Hall effect sensors with a 500kHz third-order Butterworth anti-aliasing filter at the front end. The thermocouples are amplified by an AD8495 and then subjected to a 4kHz RC low-pass filter. All ADC synchronization pins are connected to a 1MHz clock generated by the FPGA, with the falling edge triggering the hardware timestamp unit. The sampled word stream enters a 4kB circular buffer and is then written to the AXI bus in 64-bit frames by DMA, with an overall transmission delay of 2.1μs.

[0244] Data integrity and error control include:

[0245] Absolute time error: Compared with the fiber optic PTP master clock, the 24h RMS deviation is 27ns, and the peak-to-peak deviation is 48ns; meeting the upper limit of <50ns.

[0246] Power channel excitation noise: 1 MS / s -1 With 18-bit input, the equivalent input noise is 41μV. rms The converted power noise is 32mW.

[0247] Temperature channel hysteresis: 200 kS / s -1 With a low-pass filter, the phase delay is -3dB and 31μs, which is negligible for a 0.1s control window.

[0248] Performance comparison experiment: reducing the power ADC rate to 250kS / s -1 Instantaneous power waveform harmonic leakage caused the 10-second rolling mean square error to increase from 0.9% to 1.4%, and the inverter peak current to increase by 6A. The temperature ADC rate was reduced to 50kS / s. -1 The temperature rise detection was delayed by 0.4s, which caused a delay in the thermal constraint triggering and the peak inrush current of the busbar rose by 7A.

[0249] The basic implementation uses 1 MS / s -1 Power sampling at 200 kS / s -1 Temperature sampling was performed on a 1MWh prototype for 24 hours, and the module SOC variance decreased from 3.5% to 1.4%, with an early warning of temperature exceeding the limit 0.8 seconds in advance.

[0250] The extended embodiment limits the power sampling rate to 500 kS / s. -1 The GPU load decreased by 7%, but the SOC variance only decreased to 1.7%, validating 1MS / s. -1 The necessity of sampling design.

[0251] This configuration enables high-resolution capture of transient power and temperature, providing reliable input for subsequent optimal transmission gradient and closed-loop equalization. Meanwhile, its error budget and experimental data demonstrate the rationality of the rate and resolution selection.

[0252] Preferably, the optimal transmission algorithm adopts the entropy-regularized Sinkhorn algorithm with an entropy regularization coefficient of 0.05 and 30 iterations. The iteration stops when the difference between the L2 norms of the power transport matrices obtained from two consecutive iterations is less than 0.

[0253] In wirelessly coupled networks, each battery module acts as both a "power source" and a "power sink." Its instantaneous many-to-many power distribution can be abstracted as an optimal transport (OT) problem: mapping the supply vector p to the demand vector q and minimizing the cost matrix C composed of fast, tunable magnetic-acoustic channels. Directly solving the classic OT problem requires an iterative complexity of O(n^2). 3 The process cannot be completed within a 0.1s period. This invention employs the entropy-regularized Sinkhorn algorithm, adding an entropy term εH(W) (ε=0.05) to the objective function, transforming the problem into a matrix calibration in a log-linear space, which significantly accelerates convergence and naturally generates smooth gradients.

[0254] The core iteration is defined as K = exp(-C / ε). The algorithm uses row vector u... (0) and column vector v (0) Initialize all values ​​to 1, then scale alternately:

[0255]

[0256] The estimated value of the power transport matrix in the t-th iteration is:

[0257] W (t) =diag(u (t) )K diag(v (t) )

[0258] When the L2 difference of two consecutive steps is ||W (t) -W (t-1) ||2 is less than 1×10 -3 Convergence is considered achieved after 30 iterations. In GPU parallel implementation, each iteration takes 0.9 μs, with a total computation latency of 27 μs.

[0259] In practical applications, the cost matrix C on the FPGA side is a linear superposition of three parts: real-time coupling loss, switching loss, and inter-coil routing weights; the supply and demand vectors p and q are directly taken from the power margin output by the prediction layer. To prevent numerical underflow, the log-sum-exp technique is used in the CUDA kernel to maintain stability. The entropy coefficient of 0.05 is obtained through offline grid search: if it is reduced to 0.02, although the peak current decreases by 2A, the number of pixel switching increases by 30%; if it is increased to 0.10, the convergence step count decreases to 15, but the power allocation error increases to 1.4%.

[0260] Continuous operation for 24 hours on a 1MWh prototype: The updated pixel matrix of the Sinkhorn reduced the peak inrush current of the bus to 191A; if the unregularized linear programming solution (CPU 18ms) is used instead, the peak current drops to 188A, but the cycle timeout is 80%. Reducing the iteration limit to 15 steps halves the latency, but the mean square error increases from 0.9% to 1.2%, confirming that the 30-step threshold combination satisfies the accuracy-real-time tradeoff.

[0261] Basic Implementation Example: Entropy coefficient 0.05, 30 steps, threshold 1×10⁻⁵ -3 The SOC variance was reduced to 1.4%, and the control cycle delay was 78 μs.

[0262] Extended example: with an entropy coefficient of 0.10, 15 steps, and a time delay of 43μs, the SOC variance only decreased to 1.7%, verifying the rationality of the parameter selection.

[0263] Preferably, after generating feedback data, the SHA-256 hash algorithm is used to calculate the hash value and write it into the blockchain storage. The blockchain node completes the transaction confirmation containing the hash value and the Coordinated Universal Time timestamp within 200 milliseconds.

[0264] Feedback data (power flow matrix P) m Temperature vector T m Execution instruction frame C drv Together with its UTC timestamp, it forms a 4kB JSON message. The control server calculates this locally using the OpenSSL digest library.

[0265] H = SHA256(JSON)

[0266] A 256-bit hash value H is obtained. The hash and the metadata of the original message (size, structure checksum) are encapsulated into a blockchain transaction and broadcast to the four-node consortium blockchain. The chain runs RAFT-3f consensus (allowing one failure), with a block cycle of 100ms and a block size of 1MB, and can confirm a majority of 2f+1 votes within a single cycle. Experiments show that network transmission takes 25ms, sorting takes 60ms, and block writing takes 38ms, with a total latency of 123ms, and the steady-state latency is below the 200ms upper limit.

[0267] In practical applications, the node deployment consists of two local edge servers at the energy storage station, one at the dispatch center, and one in the cloud disaster recovery area; the nodes are connected via a TLS 1.3 encrypted tunnel with a bandwidth of 1Gbps. -1 Round trip time: 8ms. Time synchronization: All nodes maintain a ±50ns deviation with the National Time Service Center's UTC source via PTP to ensure traceability of hash timestamps. Data pruning: Only hash and index information are stored; the complete original feedback is retained on a local read-only SSD, keeping blockchain storage overhead to 128B / cycle. Anomaly policy: If two consecutive transactions are not confirmed within 3 seconds, the controller switches to offline mode, persisting hashes locally and recording event codes to avoid affecting the 100Hz closed loop.

[0268] After 48 hours of continuous operation, a total of 1,728,000 transactions were recorded on the blockchain, with an average confirmation latency of 137ms and a 99.9 percentile latency of 182ms. A SHA-256 collision test was conducted on 10,000 randomly selected transactions, achieving a 100% hash matching rate. One network packet loss occurred, and RAFT automatically re-elected the master node. The longest confirmation latency was 268ms, and offline mode was not triggered.

[0269] Using a traditional centralized log server, this invention detected two tampering risk alerts in the power and temperature logs within 24 hours; no further alerts were triggered after the consortium blockchain notarization. Node CPU load increased from 12% to 19%, and storage usage increased by 190MB / day (acceptable).

[0270] Basic implementation: four-node RAFT-3f, block cycle 100ms, hash threshold 1MB; feedback on-chain latency is stable at <200ms, and SOC variance and thermal equilibrium index remain at 1.4% and 2K respectively.

[0271] Extended Implementation Example: Increasing the block cycle to 500ms increases the average acknowledgment delay to 457ms. Although it has no direct impact on closed-loop control, the operation and maintenance audit delay is too large. Therefore, a 100ms cycle is the recommended configuration.

[0272] By combining SHA-256 hashing with a lightweight consortium blockchain, this invention achieves tamper-proof evidence storage and millisecond-level traceability of feedback data without interfering with 100Hz real-time control, providing a reliable basis for subsequent compliance audits and fault tracing.

[0273] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0274] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A WIPI wireless power dynamic equalization method for energy storage system grid connection, characterized in that, The method comprises: Establishing a digital twin model, integrating system parameters representing coil geometry, battery module electrochemical impedance, and coil coupling into a unified model to describe the electromagnetic topology and electrochemical state of the energy storage system; The system parameters consist of the following data: coil center coordinates, coil normal vector, coil effective area, battery module zero-frequency internal resistance value, battery module polarization resistance value, and coupling coefficient between each pair of coils; The digital twin model is established by the following steps: the coil coupling impedance matrix is calculated by using a finite element grid with a unit size of not greater than 0.5 millimeters in the electromagnetic part; the battery module impedance curve is fitted by using a fractional order equivalent circuit containing a constant phase element in the electrochemical part; the fractional order order and the circuit parameters are determined by using a sparse identification dynamics algorithm with a penalty coefficient of for the purpose of minimizing the mean square error of the model output and the measured impedance spectrum. Collect and synchronize electrical and magnetic flux data, perform fractional order dynamics prediction based on the digital twin model, obtain state of charge data and temperature constraint data, and generate control optimization initialization data; According to the prediction results and initialization data, generate power coupling control target data, and generate magnetic-acoustic metamaterial tuning data and execute control instruction data accordingly; According to the execution control instruction data, drive the inverter, and according to the magnetic-acoustic metamaterial tuning data, tune the magnetic-acoustic metamaterial device to realize wireless power coupling, measure power flow data and temperature data in real time, calculate the coupling correction gradient using the optimal transmission algorithm to update the magnetic-acoustic metamaterial device, and form feedback data from the power flow data, temperature data, and execution control instruction data to return to the power coupling control target data generation step, thereby realizing dynamic balancing of wireless power between battery modules in a closed loop.

2. The method of claim 1, wherein, The synchronization of electrical and magnetic flux data is completed using a time-sensitive optical fiber network, with a synchronization jitter of no more than 50 nanoseconds.

3. The method of claim 1, wherein, The order of fractional order dynamics prediction is a value between 0.80 and 0.90, which is determined by minimizing the mean square error of state of charge within a 10-second prediction window.

4. The method of claim 3, wherein, The power coupling control target data is generated by linearly fusing the coupling weight matrix output by the graph neural network and the optimal coupling matrix output by the variational quantum optimizer with weight coefficients of 0.60 and 0.40, respectively. The graph neural network uses a 3-layer GraphSAGE structure with 256 hidden nodes per layer, and the variational quantum optimizer uses a 20-qubit and 6-layer parameterized quantum circuit.

5. The method of claim 4, wherein, The magnetic-acoustic metamaterial tuning data is a 32x32 pixel matrix, where the corresponding pixel is set to an open state when the element value in the coupling matrix target data is greater than 0.35, and the remaining pixels are set to a closed state. The pixel switching completion time is no more than 60 microseconds.

6. The method of claim 1, wherein, The power flow data is collected at a rate of 1,000,000 samples per second using an 18-bit analog-to-digital converter, and the temperature data is collected at a rate of 200,000 samples per second using a 16-bit analog-to-digital converter.

7. The method of claim 1, wherein, The optimal transport algorithm adopts the entropy-regularized Sinkhorn algorithm, the entropy regularization coefficient is 0.05, the iteration number is 30 times, and the iteration is stopped when the difference between the two-norms of the power transport matrices obtained by two consecutive iterations is less than 0.0001.

8. The method of claim 1, wherein, After generating the feedback data, a SHA-256 hash algorithm is used to calculate the hash value and write it to the blockchain storage. The blockchain node completes transaction confirmation including the hash value and coordinated universal time timestamp within 200 milliseconds.

Images