Quantum memory optimization control method and system based on physical information neural network

Abstract

The application provides a quantum memory optimization control method and system based on a physical information neural network, including: in a training phase, a physical information neural network training module takes a quantum memory dynamics equation, boundary conditions, initial conditions, numerical simulation data, measurable experimental data and a target performance function as constraints, and trains to obtain a PINN proxy model. In the deployment phase, the trained PINN proxy model runs with fixed network weights, is used to back-propagate or search for optimized control parameters according to the target performance function, and applies the control parameters to a real quantum memory. The control parameters include write control light, read control light, control light phase, control light power, Raman detuning, pulse shape, storage time, read timing and read-write light propagation configuration. For the spin wave spatial distribution that cannot be directly measured, the constraints are obtained through physical equation simulation, Hankel mapping residual or network internal state.

Description

Technical Field

[0001] This invention relates to the fields of quantum information processing, quantum storage, continuous variable quantum optics, Raman quantum storage, intelligent control, physical information neural networks, machine learning-assisted quantum experiments, and real-time feedback control. Specifically, it relates to a method and system for optimizing the control of quantum memories based on physical information neural networks. Background Technology

[0002] Quantum memories are key devices in quantum communication, quantum computing, quantum networks, and quantum-enhanced sensing. Their performance is typically determined by the following metrics: 1. Storage efficiency; 2. Storage lifetime; 3. Storage bandwidth; 4. Output state fidelity. Taking Raman quantum memories as an example, optical quantum states are mapped to atomic spin wave modes by writing control light, and then converted back to optical modes by reading control light. This process is affected by various parameters, including control light power, detuning, pulse shape, optical depth, ACStark shift, phase noise, four-wave mixing noise, mode matching, and probe system stability. Traditional optimization methods typically rely on manual parameter scanning, single-variable empirical optimization, offline numerical simulation, pre-calibrated lookup tables, or low-dimensional feedback control. These methods have the following drawbacks: 1. Low efficiency in high-dimensional parameter space search; 2. Difficulty in simultaneously optimizing efficiency, noise, bandwidth, and fidelity; 3. Difficulty in deriving the optimal control input from the target performance; 4. Difficulty in adapting to different input quantum states and different read / write configurations; 5. Difficulty in unifying offline physical simulation results, historical experimental data, and real-world automatic system control.

[0003] Existing research on quantum memory optimization has shown that theoretical models, pulse optimization, phase compensation, spin wave spatial distribution design, reverse readout configurations, parameter scanning, or experimental calibration can improve storage efficiency and reduce noise. However, the spin wave spatial distribution typically cannot be directly measured experimentally and must be inferred and constrained through quantum memory dynamics equations, Hankel mappings, or numerical simulations. Furthermore, a long-standing criticism of quantum memories is that their optimal control requires prior information about the temporal patterns of the input signal, a requirement often unmet in practical applications. Relying solely on real-time feedback from a real system for high-dimensional searches results in numerous experiments, slow convergence, and susceptibility to system drift.

[0004] Therefore, a new method for optimizing the control of quantum memories is urgently needed: first, train the PINN model using physical equations and measurable data so that it can learn to deduce the input signal and optimal control under limited control input and output of the quantum memory; then, in the deployment phase, use the trained PINN model to achieve the optimal control output and apply this output to the real quantum memory system to achieve automatic control and experimental feedback correction. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide an optimized control system and method for quantum memories based on physical information neural networks.

[0006] A quantum memory optimization control system based on a physical information neural network according to the present invention includes: The training data construction module is used to acquire data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, and to construct training data based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity. A physical information neural network training module is used to train a PINN model based on the training data. The model deployment module is used to deploy the trained PINN model into a real quantum storage and control system, and to use the trained PINN model deployed in the real quantum storage and control system to back-calculate the optimized control parameters based on the actual optical quantum state. A real quantum memory module for receiving, storing, and reading optical quantum states based on quantum memory; A control light generation module is used to generate at least one control light field acting on the real quantum memory module according to the optimized control parameters. The experimental measurement module is used to collect data related to the input state, output state, control light, noise, phase, efficiency, readout efficiency, or fidelity of the real quantum memory module. An automatic feedback execution module is used to evaluate the control effect based on the measurement results of the experimental measurement module and trigger the control input inversion module to regenerate the control parameters.

[0007] Preferably, the training data construction module includes: acquiring data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity; and simultaneously, calculating data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the quantum memory dynamics equations. Training data is constructed based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, as well as the data calculated through the quantum memory dynamics equations.

[0008] Preferably, the physical information neural network training module includes: the training loss function of the PINN model is:

[0009] in: The residual of the quantum storage dynamics equation; The fitting loss for measurable experimental data; For boundary condition loss; Loss due to initial conditions; Regularization terms are used to control optical smoothness, energy limitations, bandwidth limitations, or physical realizability. , , , , , The weights for each loss item.

[0010] Preferably, the physical information neural network training module further includes: during the PINN model training process, adaptively switching the target performance function based on different input quantum states or different task objectives; The target performance function includes any one or more of the following: storage efficiency, read efficiency, additional noise, four-wave mixing noise, output state fidelity, output compression, storage bandwidth, and phase stability.

[0011] Preferably, the optimized control parameters in the model deployment module include at least one of the following: optimized write control light, read control light, detuning, control light phase, control light power, control light pulse shape, storage time, read timing, or read / write light propagation configuration.

[0012] Preferably, the quantum memory in the real quantum memory module includes any one of the following: Raman quantum memory, EIT quantum memory, AFC quantum memory, atomic ensemble quantum memory, rare earth ion quantum memory, and solid-state spin quantum memory.

[0013] Preferably, the experimental measurement module includes: a balanced zero-beat detector and a photon counter; The optical quantum state is measured using a balanced zero-beat detector and a photon counter. The quantum storage control parameters are then deduced from the measured photon quantum state using a trained PINN model. The optical quantum states include coherent states, squeezed states, displacement-squeezed states, entangled states, single-photon states, time-bin encoded states, frequency-bin encoded states, or combinations thereof.

[0014] Preferably, the system further includes: monitoring parameters of the quantum memory, including effective optical depth, Raman detuning, decoherence rate, AC Stark shift, optical loss, additional noise, four-wave mixing noise, phase noise, control optical amplitude drift, and mode mismatch; when any one or more parameters do not meet the preset requirements, the control input inversion module is triggered to regenerate the control parameters.

[0015] A quantum memory optimization control method based on a physical information neural network, provided by the present invention, includes: Step S1: Obtain data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the training data construction module, and construct training data based on the obtained data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity. Step S2: Train the PINN model based on the training data using the physical information neural network training module; Step S3: Deploy the trained PINN model into the real quantum storage and control system through the model deployment module, and use the trained PINN model deployed in the real quantum storage and control system to back-calculate the optimized control parameters based on the actual optical quantum state; Step S4: Receive, store, and read optical quantum states using a real quantum memory module based on a quantum memory; Step S5: The light generation module generates at least one control light field that acts on the real quantum memory module according to the optimized control parameters. Step S6: Collect relevant data on the input state, output state, control light, noise, phase, efficiency, readout efficiency, or fidelity of the real quantum memory module through the experimental measurement module; Step S7: The automatic feedback execution module evaluates the control effect based on the measurement results of the experimental measurement module and triggers the control input inversion module to regenerate the control parameters.

[0016] Preferably, the training data construction module includes: acquiring data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity; and simultaneously, calculating data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the quantum memory dynamics equations. Training data is constructed based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, as well as the data calculated through the quantum memory dynamics equations. The physical information neural network training module includes: the training loss function of the PINN model is:

[0017] in: The residual of the quantum storage dynamics equation; The fitting loss for measurable experimental data; For boundary condition loss; Loss due to initial conditions; Regularization terms are used to control optical smoothness, energy limitations, bandwidth limitations, or physical realizability. , , , , , Weights for each loss item; During the training of the PINN model, the target performance function is adaptively switched based on different input quantum states or different task objectives. The target performance function includes any one or more of the following: storage efficiency, read efficiency, additional noise, four-wave mixing noise, output state fidelity, output compression, storage bandwidth, and phase stability.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention enables quantum memories to first obtain a PINN model through physical equations and data training, and then use the trained PINN to automatically generate control parameters and improve storage performance in a real system. 2. This invention can embed dynamic phase compensation, arbitrary input waveform control, hardware distortion correction, Hankel transform constraints, and reverse readout configuration into the control input inversion framework of the trained PINN surrogate model to achieve automatic control and low-noise, high-efficiency readout of quantum memory; 3. PINN is mainly trained using physical equations, numerical simulation data, and measurable experimental data during the training phase, without requiring a large amount of actual data; during the deployment phase, the trained PINN is used as a fixed proxy model to improve the operating efficiency of the real system. 4. This invention unifies the quantum storage dynamics equations and data into PINN training, avoiding the problem of a lack of physical constraints in a purely data-driven model; 5. In this invention, the target performance function is related to the write control light, read control light, phase, detuning, read timing, and read / write light propagation configuration. The PINN network learns the relationship between the write control light, read control light, phase, detuning, read timing, and read / write light propagation configuration and the target performance function. Given the target performance function, the write control light, read control light, phase, detuning, read timing, and read / write light propagation configuration can be obtained through PINN. 6. The online component of the real system of this invention is used to perform control, measure performance, and trigger the re-inverse control input; the new data can be used for subsequent batch retraining or calibration. Attached Figure Description

[0019] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1This is a schematic diagram of an optimized control system for a quantum memory based on a physical information neural network.

[0020] Figure 2 This is a schematic diagram of PINN network training.

[0021] Figure 3 This is a diagram showing the optical path and electronic signal connections in a practical quantum storage system. Detailed Implementation

[0022] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0023] Example 1 A quantum memory optimization control system based on a physical information neural network according to the present invention includes: The training data construction module is used to acquire data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, and to construct training data based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity. Specifically, the training data construction module includes: acquiring data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity; and simultaneously, calculating data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the quantum memory dynamics equations; wherein, the quantum memory dynamics equations include the light field propagation equation, spin wave evolution equation, light-matter coupling equation, noise evolution equation, phase evolution equation, Hankel mapping constraint, or a combination thereof.

[0024] Training data is constructed based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, as well as the data calculated through the quantum memory dynamics equations.

[0025] A physical information neural network training module is used to train a PINN model based on the training data. Specifically, the physical information neural network training module includes: the training loss function of the PINN model is:

[0026] in: The residual of the quantum storage dynamics equation; The fitting loss for measurable experimental data; For boundary condition loss; Loss due to initial conditions; Regularization terms are used to control optical smoothness, energy limitations, bandwidth limitations, or physical realizability. , , , , , The weights for each loss item.

[0027] The physical information neural network training module also includes: during the PINN model training process, the target performance function is adaptively switched based on different input quantum states or different task objectives; The target performance function includes any one or more of the following: storage efficiency, read efficiency, additional noise, four-wave mixing noise, output state fidelity, output compression, storage bandwidth, and phase stability.

[0028] The model deployment module is used to deploy the trained PINN model into a real quantum storage and control system, and to use the trained PINN model deployed in the real quantum storage and control system to back-calculate the optimized control parameters based on the actual optical quantum state. Specifically, the optimized control parameters in the model deployment module include at least one of the following: optimized write control light, read control light, detuning, control light phase, control light power, control light pulse shape, storage time, read timing, or read / write light propagation configuration.

[0029] A real quantum memory module for receiving, storing, and reading optical quantum states based on quantum memory; Specifically, the quantum memory in the real quantum memory module includes any one of the following: Raman quantum memory, EIT quantum memory, AFC quantum memory, atomic ensemble quantum memory, rare earth ion quantum memory, and solid-state spin quantum memory.

[0030] A control light generation module is used to generate at least one control light field acting on the real quantum memory module according to the optimized control parameters. Specifically, the control light generation module includes: AWG, AOM, EOM, laser frequency controller or phase-locked module.

[0031] The experimental measurement module is used to collect data related to the input state, output state, control light, noise, phase, efficiency, readout efficiency, or fidelity of the real quantum memory module. Specifically, the experimental measurement module includes: a balanced zero-beat detector and a photon counter; The optical quantum state is measured using a balanced zero-beat detector and a photon counter. The quantum storage control parameters are then deduced from the measured photon quantum state using a trained PINN model. The optical quantum states include coherent states, squeezed states, displacement-squeezed states, entangled states, single-photon states, time-bin encoded states, frequency-bin encoded states, or combinations thereof.

[0032] The new experimental data obtained by the experimental measurement module is stored as subsequent batch retraining or model calibration data, but it is not required to update the network weights of the PINN model in each experimental control loop.

[0033] An automatic feedback execution module is used to evaluate the control effect based on the measurement results of the experimental measurement module and trigger the control input inversion module to regenerate the control parameters.

[0034] The system further includes: monitoring the quantum memory for parameters such as effective optical depth, Raman detuning, decoherence rate, AC Stark shift, optical loss, additional noise, four-wave mixing noise, phase noise, control optical amplitude drift, and mode mismatch. When any one or more parameters do not meet the preset requirements, the control input inversion module is triggered to regenerate the control parameters.

[0035] Example 2 Example 2 is a preferred example of Example 1. This invention provides an optimized control system for quantum memory based on a physical information neural network, such as... Figures 1 to 3 As shown, it includes the training phase and the deployment control phase.

[0036] The training phase includes: The training data construction module is used to acquire or generate training data, which includes sampling points of the quantum storage dynamics equation, boundary conditions, initial conditions, numerical simulation data, historical experimental data, and measurable performance data. The physical information neural network training module is used to write the physical equation residuals, boundary condition residuals, initial condition residuals, data fitting errors, target performance loss, and control regularization terms into the loss function to train the PINN model. The model validation module is used to verify the training PINN's ability to predict the input signal light field, optimal control light field, storage efficiency, readout efficiency, additional noise, phase drift, and output state fidelity using experimental data.

[0037] The deployment control phase includes: The control light generation module keeps the weights of the trained PINN model fixed during the deployment control process and applies them to the real quantum memory to deduce the optimal control parameters for the control input. The module generates write control light, read control light, or other auxiliary control light based on the deduced control parameters. A real quantum memory module for receiving, storing, and reading optical quantum states; The experimental measurement module is used to acquire data related to input state, output state, reference pulse, noise, efficiency, readout efficiency, phase, or fidelity. The automatic feedback execution module is used to determine whether to continue calling the trained PINN to re-infer the control input based on the actual system measurement results, or to make minor adjustments to the control parameters.

[0038] The core function of PINN is not continuous online training during the operation of the real physical system, but rather, after training, it is used to predict the output performance corresponding to the control inputs and to infer or search for the optimal control inputs required by the real physical system. The online components during the operation of the real system mainly involve measurement, control input updates, and feedback execution.

[0039] In one implementation, the training loss function of PINN includes:

[0040] in: : Residuals of the quantum storage dynamics equations; : Measurable experimental data fitting loss; Boundary condition loss; Initial condition loss; : Regularization terms that control optical smoothness, energy limitation, bandwidth limitation, or physical realizability; , , , , , Weights of each loss item.

[0041] Control input inversion based on trained PINN After training, PINN can be represented as:

[0042] in, For PINN network; For fixed network parameters after training, This indicates the output optical quantum state, experimental platform state, or known system parameters. This indicates that the control input is not optimized. This represents the output control optical field, input signal optical field, Raman detuning, phase, control optical power, storage time, noise, and fidelity predicted by PINN.

[0043] Seeking The system then converts these signals into physical control signals executable by an AWG, AOM, EOM, RF source, laser frequency controller, or phase-locked module, and applies them to the real quantum memory. The real system measurements are used to evaluate the control effect and can serve as data for subsequent batch retraining or model calibration, but online updates of the PINN weights are not required in each experimental cycle.

[0044] In a preferred embodiment, the Raman storage process includes optical loss, spin wave decoherence, four-wave mixing noise, AC Stark shift, control optical phase drift, mode mismatch, etc. It can be described by the following coupling equations:

[0045]

[0046] in: This indicates the noise term. Slowly varying envelope of the light field; : Atomic spin wave envelope; Effective optical depth; Raman detuning; : Control the Rabi frequency of light; Spin wave decoherence rate; For the evolution operator of the light field, For atomic spin-wave evolution operators; For signal light field, It is an atomic spin wave; PINN network through input System parameters, output , Or its equivalent real part, imaginary part, or orthogonal component representation, and write the residual of the above differential equation into .

[0047] During the read phase, the read control light is preferably configured to propagate in the opposite direction to the write control light. The trained PINN uses the reverse read configuration, read control light waveform, read phase, and read timing as optimizable control inputs; the real system evaluates the control performance using measurable output light field, read efficiency, noise, and phase data.

[0048] PINN Training and Deployment Control In this embodiment, training samples are first constructed. Each training sample includes an input state description. Control input Boundary conditions, initial conditions, and output performance data obtained from numerical simulation or experiments. Control inputs may include:

[0049] in, To control the optical pull ratio frequency, To read the control light pull ratio frequency, To write control optical phase, To read the control light phase, Disharmony, To control the optical power, To read and control optical power; PINN is then trained using a loss function, and the network parameters are fixed after training. The trained PINN network is then deployed to the experimental control computer. Given the output state and target performance, the network directly outputs the optimal control. and will Converted into executable signals for AWG, AOM, EOM, RF source, and laser frequency controller.

[0050] After the real quantum storage system performs write, store, and read operations, the experimental measurement module acquires the output light field, storage efficiency, read efficiency, added noise, phase drift, and fidelity. The measurement results are used to assess the control effect and can be stored in a dataset for subsequent offline retraining or model calibration. During the deployment phase, training the PINN weights in each round of experiments is not required.

[0051] In Raman storage, control light may cause AC Stark shift, resulting in a phase shift of the spin wave:

[0052] in, The phase shift is caused by AC Stark shift. Two-photon detuning caused by AC Stark shift; During the training phase, PINN uses the phase evolution equation, reference pulse measurement data, zero-beat measurement data, or output phase measurement data together to learn the effects of AC Stark shift, two-photon detuning, and phase noise on output performance.

[0053] During the deployment phase, the trained PINN, based on the current measurable state and the target output phase, inversely derives the phase compensation for writing to or reading from the control light:

[0054] in, For the final control phase, The phase shift is caused by AC Stark shift. This is a random phase shift; In this embodiment, dynamic phase modulation is used as one of the control variables derived from PINN after training to compensate for ACStark shift and phase drift.

[0055] Arbitrary input waveform and hardware distortion correction: In this embodiment, the write control light and the read control light are represented as follows:

[0056]

[0057] in, The basis functions can be, for example: Gaussian, Hermite-Gaussian, B-spline, Fourierbasis, top-hat-like smooth pulse, Chebyshev nodal interpolation waveform, or experimentally achievable AWG waveform basis functions. This refers to the mode coefficients for the k-th time mode of the control light. The total number of patterns To read the mode coefficients in the k-th time mode of the control light, To write the final Rabi frequency of the control light, To read the final Rabi frequency of the control light; During the training phase, calibration data from AWG, AOM, EOM, or RF links can be used to train the hardware distortion correction module. Alternatively, Chebyshev node dimensionality reduction, spline interpolation, residual neural network hardware correction, differential evolution, or Bayesian optimization can be used as control input parameterization and search strategies.

[0058] During the deployment phase, the trained PINN derives its coefficients based on the target storage efficiency, output waveform, and fidelity.

[0059] This is then converted into a control optical waveform executable by real hardware. This method ensures that the control optical waveform satisfies both the quantum storage performance objectives and the power, bandwidth, and smoothness constraints of the experimental system.

[0060] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0061] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A quantum memory optimization control system based on a physical information neural network, characterized in that, include: The training data construction module is used to acquire data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, and to construct training data based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity. A physical information neural network training module is used to train the PINN model based on the training data; The model deployment module is used to deploy the trained PINN model into a real quantum storage and control system, and to use the trained PINN model deployed in the real quantum storage and control system to back-calculate the optimized control parameters based on the actual optical quantum state. A real quantum memory module for receiving, storing, and reading optical quantum states based on quantum memory; A control light generation module is used to generate at least one control light field acting on the real quantum memory module according to the optimized control parameters. The experimental measurement module is used to collect data related to the input state, output state, control light, noise, phase, efficiency, readout efficiency, or fidelity of the real quantum memory module. An automatic feedback execution module is used to evaluate the control effect based on the measurement results of the experimental measurement module and trigger the control input inversion module to regenerate the control parameters.

2. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The training data construction module includes: acquiring data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity; and simultaneously, calculating data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the quantum memory dynamics equations. Training data is constructed based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, as well as the data calculated through the quantum memory dynamics equations.

3. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The physical information neural network training module includes: the training loss function of the PINN model is: in: The residual of the quantum storage dynamics equation; The fitting loss for measurable experimental data; For boundary condition loss; Loss due to initial conditions; Regularization terms are used to control optical smoothness, energy limitations, bandwidth limitations, or physical realizability. , , , , , The weights for each loss item.

4. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The physical information neural network training module also includes: during the PINN model training process, the target performance function is adaptively switched based on different input quantum states or different task objectives; The target performance function includes any one or more of the following: storage efficiency, read efficiency, additional noise, four-wave mixing noise, output state fidelity, output compression, storage bandwidth, and phase stability.

5. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The optimized control parameters in the model deployment module include at least one of the following: optimized write control light, read control light, detuning, control light phase, control light power, control light pulse shape, storage time, read timing, or read / write light propagation configuration.

6. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The quantum memory in the real quantum memory module includes any one of the following: Raman quantum memory, EIT quantum memory, AFC quantum memory, atomic ensemble quantum memory, rare earth ion quantum memory, and solid-state spin quantum memory.

7. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The experimental measurement module includes: a balanced zero-beat detector and a photon counter; The optical quantum state is measured using a balanced zero-beat detector and a photon counter. The quantum storage control parameters are then deduced from the measured photon quantum state using a trained PINN model. The optical quantum states include coherent states, squeezed states, displacement-squeezed states, entangled states, single-photon states, time-bin encoded states, frequency-bin encoded states, or combinations thereof.

8. The quantum memory optimization control system based on physical information neural network according to claim 1, characterized in that, The system further includes: monitoring the quantum memory for parameters such as effective optical depth, Raman detuning, decoherence rate, ACStark shift, optical loss, additional noise, four-wave mixing noise, phase noise, control optical amplitude drift, and mode mismatch. When any one or more parameters do not meet the preset requirements, the control input inversion module is triggered to regenerate the control parameters.

9. A quantum memory optimization control method based on a physical information neural network, characterized in that, include: Step S1: Obtain data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the training data construction module, and construct training data based on the obtained data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity. Step S2: Train the PINN model based on the training data using the physical information neural network training module; Step S3: Deploy the trained PINN model into the real quantum storage and control system through the model deployment module, and use the trained PINN model deployed in the real quantum storage and control system to back-calculate the optimized control parameters based on the actual optical quantum state; Step S4: Receive, store, and read optical quantum states using a real quantum memory module based on a quantum memory; Step S5: The light generation module generates at least one control light field that acts on the real quantum memory module according to the optimized control parameters. Step S6: Collect relevant data on the input state, output state, control light, noise, phase, efficiency, readout efficiency, or fidelity of the real quantum memory module through the experimental measurement module; Step S7: The automatic feedback execution module evaluates the control effect based on the measurement results of the experimental measurement module and triggers the control input inversion module to regenerate the control parameters.

10. The quantum memory optimization control method based on physical information neural network according to claim 8, characterized in that, The training data construction module includes: acquiring data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity; and simultaneously, calculating data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity through the quantum memory dynamics equations. Training data is constructed based on the acquired data including input state, output state, control light, noise, phase, efficiency, readout efficiency, and fidelity, as well as the data calculated through the quantum memory dynamics equations. The physical information neural network training module includes: the training loss function of the PINN model is: in: The residual of the quantum storage dynamics equation; The fitting loss for measurable experimental data; For boundary condition loss; Loss due to initial conditions; Regularization terms are used to control optical smoothness, energy limitations, bandwidth limitations, or physical realizability. , , , , , Weights for each loss item; During the training of the PINN model, the target performance function is adaptively switched based on different input quantum states or different task objectives. The target performance function includes any one or more of the following: storage efficiency, read efficiency, additional noise, four-wave mixing noise, output state fidelity, output compression, storage bandwidth, and phase stability.

Images