Lithium ion battery state of health prediction method and system based on quantum gate coupling

Abstract

A lithium ion battery state of health prediction method and system based on quantum gate coupling. The cycle operation data of the lithium ion battery is acquired, the health characteristics are constructed, and the time series samples are constructed. A quantum enhanced gate recurrent neural network is constructed, a plurality of parameterized quantum circuits are configured in at least one gate module, and a quantum gate coupling mechanism is established between different gate modules. The quantum feature representation output by the plurality of parameterized quantum circuits is fused with the inter-gate coupling information, and the gate function value is calculated. The gate function value is used to update the recurrent state, and finally the SOH prediction value is output. The application enhances the single gate expression capability through the structured combination of multiple quantum circuits, realizes information interaction and collaborative calculation through inter-gate coupling, and improves the nonlinear modeling capability of the model for the battery degradation process and the cross-cell generalization capability.

Description

Technical Field

[0001] This invention relates to the fields of lithium-ion battery state monitoring, battery management system, artificial intelligence prediction, and quantum machine learning, and more specifically to a method and system for predicting the health status of lithium-ion batteries based on quantum gating coupling. Background Technology

[0002] Lithium-ion batteries are widely used in new energy vehicles, energy storage power stations, and other fields. As the number of charge-discharge cycles increases, batteries gradually age, making accurate prediction of state of health (SOH) crucial for battery management systems.

[0003] Among existing data-driven methods, gated recurrent neural networks such as Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs) can process time-series data and have been used for SOH prediction. In recent years, researchers have attempted to introduce parameterized quantum circuits into gated recurrent neural networks to form quantum-enhanced LSTM (QLSTM) or quantum-enhanced GRU (QGRU) to enhance nonlinear expressive power.

[0004] However, existing quantum augmentation recurrent neural networks still have the following shortcomings:

[0005] 1. A single gated module typically has only one parameterized quantum circuit, resulting in a limited number of quantum state mapping paths, making it difficult to fully characterize the higher-order nonlinear relationships in the battery degradation process; 2. Different gating modules (such as forget gate, input gate, and output gate) are independent of each other and lack an effective information coupling mechanism. The model cannot make full use of the correlation information generated during the calculation of different gating modules. 3. There is a lack of structured combination methods among multiple quantum circuits. Even when multiple quantum circuits are introduced, they are mostly used in simple parallel configurations, lacking collaborative computing. 4. The integration method between quantum circuit output and gated computation is not flexible enough, which limits the efficiency of quantum feature utilization; 5. The model structure lacks versatility and scalability, and is mostly designed for specific network structures, making it difficult to extend to other gated sequence models such as LSTM and GRU; 6. It is difficult to balance the model's expressive power, information interaction ability, and trainability; The aforementioned problems limit the ability of existing quantum augmentation recurrent neural networks to model complex time-series data, resulting in insufficient accuracy, stability, and generalization ability in SOH prediction. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a quantum-gated coupling-based method for predicting the health status of lithium-ion batteries that overcomes or at least partially solves the above problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: S1. Obtain lithium-ion battery cycle operation data, construct health features characterizing the battery degradation state based on the cycle operation data, and construct the health features as time series input samples according to a preset time window; the time series input samples contain multiple health feature vectors arranged in time steps; The lithium-ion battery cycle operation data includes one or more of the following: voltage, current, capacity, temperature, charging time, discharging time, and number of cycles. S2. Construct a quantum-enhanced gated recurrent neural network, wherein the quantum-enhanced gated recurrent neural network includes multiple gated modules; For each time step in the time series input sample, perform the following sub-steps: The health feature vector of this time step is combined with the recursive state of the previous time step to form the combined input vector of the current time step; Multiple parameterized quantum circuits are configured within at least one of the gated modules, and the combined input vector is input to the multiple parameterized quantum circuits respectively. Each parameterized quantum circuit independently performs quantum computation and outputs a quantum feature representation. The quantum feature representations output by multiple parameterized quantum circuits within the same gating module are combined and fused to form the comprehensive quantum feature of the gating module at the current time step; A quantum gating coupling mechanism is established between different gating modules. The comprehensive quantum features of the latter gating module at the current time step are fused with the comprehensive quantum features obtained from the former gating module at the current time step to obtain fused gating features. The gating function value of the gating module at the current time step is calculated based on the fused gating features, and the recursive state of the current time step is updated according to the gating function value. S3. Repeat the above processing for each time step until all time steps in the time series input sample have been processed, and output the predicted value of the lithium-ion battery health status.

[0008] Preferably, the parameterized quantum circuit includes a data encoding layer, a parameterized rotation layer, a quantum entanglement layer, and a measurement output layer, which are executed sequentially. The data encoding layer is used to map the input combined input vector into a quantum state; The parameterized rotation layer is used to apply a rotation gate with trainable parameters to the quantum state; The quantum entanglement layer is used to apply a controlled NOT gate or a controlled phase gate to the rotated quantum state to establish entanglement between qubits; The measurement output layer is used to perform Pauli-Z measurements on the entangled quantum state and output classical values ​​as the representation of the quantum features.

[0009] Preferably, the topology used when combining and fusing the quantum feature representations output by the multiple parameterized quantum circuits within the same gating module is one of the following: Parallel structure: The input ends of the multiple parameterized quantum circuits receive the same combined input vector, and the output ends are calculated independently and then fused. Serial connection: The output of the previous parameterized quantum circuit is connected to the input of the next parameterized quantum circuit; Hybrid structure: At least two of the plurality of parameterized quantum circuits are connected in parallel, and their output terminals are then connected in series with at least one parameterized quantum circuit.

[0010] Preferably, the specific method for combining and fusing the quantum feature representations output by the multiple parameterized quantum circuits within the same gating module is as follows: the input ends of the multiple parameterized quantum circuits receive the same combined input vector, each calculates independently, and then the output results are directly spliced ​​together to form the comprehensive quantum feature of the gating module.

[0011] Preferably, the quantum-gated coupling mechanism includes at least one of the following: Quantum feature transfer: The comprehensive quantum features of the previous gating module at the current time step are used as the coupling information of the next gating module; Parameter sharing: Sharing trainable parameters of parameterized quantum circuits among different gating modules; Measurement feedback: The integrated quantum characteristics of one gating module are used to generate adjustment factors to adjust the inputs, fusion weights, or parameter settings of other gating modules; Gated modulation: Utilizing the comprehensive quantum characteristics of one gating module to modulate the gating function value or candidate state calculation of another gating module.

[0012] Preferably, the quantum gating coupling mechanism is quantum feature transfer: according to a preset order, the comprehensive quantum feature output by the previous gating module at the current time step is spliced ​​and fused with the comprehensive quantum feature of the subsequent gating module itself; wherein, the initial gating module does not receive the preceding coupling information, and its fused gating feature is calculated only based on its own comprehensive quantum feature.

[0013] Preferably, when the quantum-enhanced gated recurrent neural network is a quantum-enhanced long short-term memory network, the quantum feature transfer is performed in the order of forget gate, input gate, candidate state gate, and output gate: The comprehensive quantum features output by the forget gate at the current time step are used as the coupling information for the input gate computation. The comprehensive quantum features output by the input gate are used as the coupling information for the candidate state generation module. The comprehensive quantum features output by the candidate state generation module are used as the coupling information for the output gate computation. Among them, the forget gate, as the starting gating module, does not receive preceding coupling information.

[0014] Preferably, the health characteristics include one or more of the following: capacity characteristics, voltage characteristics, current characteristics, temperature characteristics, cycle count characteristics, charge / discharge time characteristics, energy characteristics, curve morphology statistical characteristics, and incremental capacity analysis characteristics; The capacity characteristic reflects the decay of the battery's usable capacity, the voltage characteristic reflects the change in voltage plateau, the current characteristic reflects the decay of current during the constant voltage stage, the charge / discharge time characteristic reflects the change in charge / discharge time, the curve morphology statistical characteristics include the slope, skewness, and entropy changes of the voltage curve, current curve, or capacity curve, and the incremental capacity analysis characteristics include the peak position, peak amplitude, or peak area changes in the dQ / dV curve.

[0015] Preferably, the method further includes the step of training the quantum-enhanced gated recurrent neural network: Step A: Input the time series input samples into the quantum-enhanced gated recurrent neural network to obtain the health status prediction value; Step B: Calculate the mean squared error or mean absolute error between the predicted health status value and the preset true label, and use it as the loss function value; Step C: Using automatic differentiation, calculate the gradient of the loss function value with respect to the following trainable parameters: The angular parameters of all rotating doors within each parameterized quantum circuit; The weights and biases of the fully connected layer in the quantum-enhanced gated recurrent neural network used to map fused gated features to gated function values; The weights and biases of the output layer in the quantum-enhanced gated recurrent neural network that map the final recurrent state to a health state prediction value; Step D: Update all trainable parameters mentioned above based on the calculated gradients. Step E: Repeat steps A through D until the loss function value converges or the preset number of training rounds is reached.

[0016] Based on the same inventive concept, this invention also discloses a lithium-ion battery health state prediction system based on quantum-gated coupling, used to execute any of the above-mentioned lithium-ion battery health state prediction methods based on quantum-gated coupling, including: The data input module is used to acquire lithium-ion battery cycle operation data; A health feature construction module is used to construct health features characterizing the battery degradation state based on the cyclic operation data. A time series sample construction module is used to construct the health features as time series input samples according to a preset time window; A quantum-enhanced gated recurrent neural network module is configured to include multiple gate modules. For each time step in the time series input sample, the health feature vector of that time step is combined with the recurrent state of the previous time step to form a combined input vector. Multiple parameterized quantum circuits are configured within at least one of the gate modules, and the combined input vector is input to each of the multiple parameterized quantum circuits to obtain a quantum feature representation output by each parameterized quantum circuit. Multiple quantum feature representations within the same gate module are combined and fused to form the comprehensive quantum feature of that gate module at the current time step. A quantum gating coupling mechanism is established between different gate modules, fusing the comprehensive quantum feature of the subsequent gate module with the comprehensive quantum feature obtained from the previous gate module to obtain a fused gating feature. A gating function value is calculated based on the fused gating feature, and the recurrent state of the current time step is updated. The output module is used to output the predicted health status of the lithium-ion battery after processing all time steps.

[0017] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following: First, it offers stronger gating capabilities. Multiple parameterized quantum circuits form a multi-path quantum computing structure within the same gating module, allowing input features to be processed through different quantum state evolution paths, thereby obtaining richer quantum feature representations.

[0018] Second, the information exchange between gating modules is more comprehensive. This invention establishes coupling relationships between different gating modules through quantum feature transfer, parameter sharing, entanglement coupling, or measurement feedback. Each subsequent gating module not only utilizes its own quantum features but also integrates information transmitted by the preceding gating module.

[0019] Third, it has a stronger ability to model complex time series. This invention enables the output of multiple quantum circuits and inter-gate coupling information to participate in recursive state updates, which is beneficial for the model to capture nonlinear changes, long-term dependencies and multidimensional feature associations in time series.

[0020] Fourth, the structural configuration is more flexible. This invention does not limit the number of quantum circuits, the number of qubits, the depth of the quantum circuits, the fusion method, or the inter-gate coupling method. In practical applications, the model structure can be flexibly configured according to the task complexity and computational resources.

[0021] Fifth, it has a wider range of applications. This invention is not only applicable to quantum-enhanced long short-term memory networks, but also to gated recurrent units and other gated sequence models. Its applications include battery health prediction, equipment condition monitoring, fault diagnosis, energy prediction, traffic flow prediction, and other time series modeling tasks.

[0022] Sixth, it facilitates quantum-classical hybrid training. This invention combines classical neural networks with parameterized quantum circuits. The classical part is responsible for feature mapping, fusion, and output prediction, while the quantum part is responsible for quantum feature extraction. The two can be jointly optimized under a unified loss function. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 This is a flowchart of the overall process for predicting the health status of lithium-ion batteries provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the multi-quantum gated coupling mechanism provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a single-parameterized quantum circuit structure provided in an embodiment of the present invention; Figure 4 This is a system framework diagram of the lithium-ion battery health status prediction method provided in the embodiments of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention discloses a method for predicting the health status of lithium-ion batteries based on quantum-gated coupling, comprising the following steps: S1. Obtain lithium-ion battery cycle operation data, construct health features characterizing the battery degradation state based on the cycle operation data, and construct time series input samples of the health features according to a preset time window; the time series input samples contain multiple health feature vectors arranged in time steps; The lithium-ion battery cycle operation data includes one or more of the following: voltage, current, capacity, temperature, charging time, discharging time, and number of cycles. S2. Construct a quantum-enhanced gated recurrent neural network, which includes multiple gated modules; For each time step in the time series input sample, perform the following sub-steps: The health feature vector of this time step is combined with the recursive state of the previous time step to form the combined input vector of the current time step; Multiple parameterized quantum circuits are configured within at least one gated module. The combined input vector is input to the multiple parameterized quantum circuits respectively. Each parameterized quantum circuit independently performs quantum computation and outputs a quantum feature representation. The quantum feature representations output by multiple parameterized quantum circuits within the same gating module are combined and fused to form the comprehensive quantum feature of the gating module at the current time step; A quantum gating coupling mechanism is established between different gating modules. The comprehensive quantum features of the latter gating module at the current time step are fused with the comprehensive quantum features obtained from the former gating module at the current time step to obtain fused gating features. The gating function value of the gating module at the current time step is calculated based on the fused gating features, and the recursive state of the current time step is updated according to the gating function value. S3. Repeat the above processing for each time step until all time steps in the time series input sample have been processed, and output the predicted value of the lithium-ion battery health status.

[0027] In one embodiment, this invention addresses the scenario of predicting the state of health (SOH) of lithium-ion batteries. By acquiring cyclic operation data of lithium-ion batteries, it constructs health features to characterize the battery's degradation state. These health features are then constructed as time-series samples and input into a quantum-enhanced gated recurrent neural network based on a multi-quantum-gated coupling mechanism. The resulting output is a predicted SOH result for the lithium-ion battery. This is then combined with... Figure 1 The overall processing flow shown above provides a detailed explanation of the steps described.

[0028] Specifically, the steps include the following: Step 1: Obtain lithium-ion battery cycle operation data This invention obtains operational data of lithium-ion batteries during cyclic charging and discharging. The operational data includes one or more of the following: voltage, current, capacity, temperature, charging time, discharging time, and number of cycles.

[0029] In practice, data is sourced from battery testing platforms, vehicle battery management systems, energy storage battery monitoring systems, or publicly available battery aging datasets. Each battery cell generates corresponding charge-discharge curves, capacity decay data, and cycle count information over multiple cycle periods. This data is used to describe the aging process of lithium-ion batteries as the number of cycles increases.

[0030] In one embodiment, experimental verification was conducted using data from four batteries in the publicly available lithium-ion battery aging dataset CALCE, designated CS2_35, CS2_36, CS2_37, and CS2_38. The data for each battery included operational data such as voltage, current, capacity, and number of cycles during the charge-discharge cycle.

[0031] The State of Health (SOH) of a battery can be determined based on the relationship between the current discharge capacity and the initial or rated capacity. Alternatively, it can be represented by capacity retention rate, remaining usable capacity, or other indicators that characterize the degree of battery aging, depending on the specific application requirements. In this embodiment, SOH serves as the supervised learning label and output result of the model. The model outputs the predicted SOH value corresponding to the target cycle period based on the battery's historical cycle characteristic sequence.

[0032] Specifically, the battery health status can be expressed as:

[0033] in, This represents the discharge capacity of the battery in the t-th cycle. This indicates the battery's initial or rated capacity. The formula above is used to define the battery health status label.

[0034] Step 2: Preprocess the battery cycle operation data The acquired battery cycle operation data is preprocessed. Preprocessing includes sorting the data according to the number of cycles or sampling time, processing missing values, outliers and noise data, and normalizing or standardizing data such as voltage, current, capacity, temperature and time.

[0035] This step can reduce the impact of abnormal sampling, different dimensions, and data fluctuations on subsequent health feature construction and SOH prediction results, making data from different battery cells or different cycle periods comparable.

[0036] Specifically, the acquired battery cycle data is sorted according to the cycle period, abnormal sampling points that are significantly outside the normal range are deleted, a small number of missing values ​​are filled by linear interpolation, and then all numerical data are normalized to map the data range to the [0,1] interval.

[0037] Step 3: Constructing the health characteristics of lithium-ion batteries Based on preprocessed battery cycle data, health characteristics are constructed to characterize the degradation state of lithium-ion batteries. These health characteristics include one or more of the following: capacity characteristics, voltage characteristics, current characteristics, temperature characteristics, cycle count characteristics, charge / discharge time characteristics, energy characteristics, curve morphology statistical characteristics, and incremental capacity analysis characteristics.

[0038] Among them, capacity characteristics are used to reflect the decay of the battery's usable capacity; voltage characteristics are used to reflect changes in voltage plateau and polarization state; current characteristics are used to reflect the decay of current during the constant voltage stage; time characteristics are used to reflect changes in charge and discharge time; curve morphology statistical characteristics are used to reflect changes in the shape of voltage curves, current curves, or capacity curves; and incremental capacity analysis characteristics are used to reflect changes related to electrochemical degradation inside the battery.

[0039] It should be noted that this invention does not limit the specific number of health features. In one specific embodiment, the health features can be 13-dimensional; in other embodiments, the health features can be added, reduced, or replaced according to the battery data type, sensor acquisition capabilities, and prediction task requirements. The number and type of health features can be flexibly adjusted according to the battery data type and prediction task.

[0040] The 13-dimensional health characteristics constructed in this embodiment specifically include: 1. Initial charging voltage; 2. Slope of the curve during the constant current and constant voltage stage; 3. Entropy value of the constant current and constant voltage stage curve; 4. Curve skewness coefficient during the constant current and constant voltage stage; 5. Energy during the constant voltage and constant current stage; 6. Energy ratio at different charging stages; 7. Constant current charging time; 8. Constant voltage charging time; 9. Current during the constant voltage phase at a specified time; 10. Specify voltage at specified time; 11. Peak voltage of incremental capacity curve; 12. Area of ​​the peak of the incremental capacity curve; 13. Maximum value of the incremental capacity curve.

[0041] The aforementioned 13 health characteristics comprehensively characterize the degradation process of lithium-ion batteries from the perspectives of voltage response, charging time, energy change, curve shape, and incremental capacity analysis. Each cycle corresponds to a 13-dimensional feature vector.

[0042] Step 4: Constructing time series samples and SOH labels Based on a preset time window length, health features from multiple consecutive cycles are constructed as time series input samples. Each time series sample includes a health feature vector representing multiple consecutive cycle times, indicating the degradation and evolution process of a lithium-ion battery within a historical cycle interval.

[0043] Further develop the SOH tag. The SOH tag can be determined based on the relationship between the current cycle available capacity and the initial capacity or rated capacity. For example: SOH = Current available capacity of the cycle / Initial capacity.

[0044] During the training phase, this invention combines time-series input samples with corresponding State of Health (SOH) labels to form training samples, enabling the model to learn the mapping relationship between historical degradation characteristics of lithium-ion batteries and SOH. During the prediction phase, this invention inputs time-series samples of the battery to be predicted into the trained model and outputs predicted SOH values ​​for the corresponding cycle time or future cycle time.

[0045] In one specific embodiment, the time window length is set to L=10, meaning each sample contains health features for 10 consecutive cycles. It should be noted that the window length L can be set to any integer greater than or equal to 2, depending on the actual application requirements, and is not limited to the value in this embodiment. Specifically, the 13-dimensional health features from the i-th to the (i+9-th)-th cycles are arranged in chronological order to form a two-dimensional matrix of shape (10, 13), which serves as an input sample for the model. The corresponding SOH label is taken from the SOH value of the (i+10-th)-th cycle (i.e., the cycle following the window).

[0046] Step 5: Construct a quantum-enhanced gated recurrent neural network based on a multi-quantum-gated coupling mechanism. Construct a quantum-enhanced gated recurrent neural network with a gated state update mechanism. This quantum-enhanced gated recurrent neural network can be a quantum-enhanced long short-term memory network (QLSTM), a quantum-enhanced gated recurrent unit (QGRU), or other recurrent neural network structures with a gated state update mechanism.

[0047] It should be noted that this invention is not limited to a single fixed recurrent network structure, but rather provides a multi-quantum circuit combination and inter-gate coupling mechanism applicable to various types of gated recurrent neural networks. In other embodiments, QGRU or other gated recurrent neural networks can be selected.

[0048] Unlike ordinary QLSTM, this invention introduces multiple parameterized quantum circuits into the gating module and establishes a quantum gating coupling mechanism between different gating modules, so that the outputs of multiple parameterized quantum circuits and the inter-gate coupling information jointly participate in the gating function calculation and recursive state update.

[0049] like Figure 2As shown, the quantum-enhanced gated recurrent neural network constructed in this embodiment includes at least one recurrent unit, which is recursively reused at different time steps. For any time step t, the input features of the current time step are... Hidden state compared to the previous time step The vectors are concatenated to form a combined input vector. The combined input vectors are then input into multiple gating modules within the recursive unit.

[0050] Multiple parameterized quantum circuits are configured within at least one gating module. Figure 2 The following explanation uses two parameterized quantum circuits, VQC1 and VQC2, configured within each gating module as an example. Parameterized quantum circuits VQC1 and VQC2 perform quantum computations on the combined input vector and output the corresponding quantum feature representations. The quantum feature representations output by multiple parameterized quantum circuits within the same gating module are then concatenated and fused within the gate to form the comprehensive quantum feature of that gating module.

[0051] Multiple gating modules are linked together in a chain according to a preset order. Gating module 1, as the starting gating module, does not receive information from preceding coupling modules and relies on its own comprehensive quantum characteristics. Perform fusion mapping and gating function calculation to obtain the corresponding gating function value. For any non-initial gating module k, calculate the integrated quantum characteristic of that gating module. Pre-coupling information passed from the previous gating module Inter-gate splicing and fusion are performed to form a fused gating feature; based on the fused gating feature, fusion mapping and gating function calculation are performed to obtain the gating function value corresponding to the gating module. The end-gating module K adopts the same method to integrate its own quantum features. Coupled with preceding information Perform inter-gate splicing and fusion, and calculate the corresponding gating function values.

[0052] The gating function values ​​output by each gating module participate in the recursive state update to obtain the recursive state at the current time step. Recursive state It is passed to the next time step, thereby realizing the recursive extraction and state update of battery degradation time sequence features.

[0053] In one specific embodiment, the quantum-enhanced gated recurrent neural network employs a quantum-enhanced long short-term memory (QLSTM) network. The QLSTM includes a forget gate, an input gate, a candidate state generation module, and an output gate. Inter-gate coupling information is chained in the order of forget gate, input gate, candidate state generation module, and output gate. The comprehensive quantum features output by the forget gate serve as the pre-coupling information for the input gate; the comprehensive quantum features output by the input gate serve as the pre-coupling information for the candidate state generation module; and the comprehensive quantum features output by the candidate state generation module serve as the pre-coupling information for the output gate. The gating function values ​​or candidate state values ​​output by each module are used to update the unit state and hidden state at the current time step.

[0054] Step Six: Perform multi-parameter quantum circuit calculations within the gating module. For any gated module, this invention maps the combined input vector of the current time step to quantum circuit input features suitable for parameterized quantum circuit processing. The dimension of the quantum circuit input features can be matched with the number of qubits, the quantum encoding method, or the quantum circuit structure.

[0055] (1) Multiple parameterized quantum circuits are configured within the gating module. In this embodiment, each gating module (forget gate, input gate, candidate state gate, and output gate) is configured with two parallel parameterized quantum circuits, denoted as the first parameterized quantum circuit VQC1 and the second parameterized quantum circuit VQC2, respectively. The structure of each parameterized quantum circuit is as follows: Figure 3 As shown, it specifically includes a data encoding layer, a parameterized rotation layer, a quantum entanglement layer, and a measurement output layer, which are executed sequentially: Data encoding layer: Maps the input features of the quantum circuit to quantum states. Specifically, Hadamard gates are first applied to each qubit to put it into a superposition state, and then rotation gates are applied to each qubit according to the input features to complete the encoding from classical features to quantum states.

[0056] Parameterized rotation layer: A trainable rotation gate (including one or more of RX, RY, and RZ) is applied to the encoded quantum state, and the rotation angle is a trainable parameter; this produces the first type of trainable parameters: quantum circuit parameters. Each rotation gate corresponds to an independent angle parameter.

[0057] Quantum entanglement layer: A controlled NOT gate or a controlled phase gate is applied to the rotated quantum state to establish entanglement between qubits. Entanglement can be linear, circular, fully connected, or layered.

[0058] Measurement output layer: Perform Pauli-Z measurements on the entangled quantum state and output the quantum characteristics in classical numerical form.

[0059] In this embodiment, each parameterized quantum circuit uses 4 qubits, and the circuit depth is set to 3. It should be noted that the number of qubits, circuit depth, and entanglement topology can all be changed according to actual needs.

[0060] (2) Integration within the door For any given gating module, the same combined input vector is simultaneously input to two parallel parameterized quantum circuits, VQC1 and VQC2, yielding two quantum feature representations respectively. These two outputs are then directly concatenated to form the comprehensive quantum feature of the gating module itself. In this embodiment, the measurement output dimension of each parameterized quantum circuit is equal to the number of qubits (4), therefore the dimension of the concatenated comprehensive quantum feature is 8. In this embodiment, the two VQC outputs are concatenated end-to-end without introducing any trainable parameters. This embodiment uses this method to concatenate two 4D outputs into an 8D comprehensive quantum feature.

[0061] It should be noted that this embodiment uses direct splicing as the fusion method for intra-gate information, which does not introduce additional trainable parameters. If a trainable linear transformation is performed on the coupled information during transmission or a parameter sharing mechanism is used, intra-gate fusion parameters will be introduced (as a variation). This experiment uses parameterless direct splicing to verify the effectiveness of the inter-gate information interaction mechanism itself.

[0062] By setting multiple parallel parameterized quantum circuits in each gating module, the same input feature can be mapped to quantum states through multiple paths in different quantum circuits, thereby enhancing the gating module's ability to express nonlinear relationships and complex temporal features during battery capacity degradation. Through this step, a multi-path quantum feature extraction structure is formed within a single gating module, which can provide richer quantum state mapping information compared to a single quantum circuit, thereby enhancing the model's ability to express the complex degradation features of lithium-ion batteries.

[0063] It should be noted that each gating module can be configured with three or more parameterized quantum circuits, and can adopt a series, series-parallel hybrid, or hierarchical structure, not limited to the two parallel structures in this embodiment. Specifically, in the parallel structure, multiple parameterized quantum circuits perform parallel calculations on the same input, and their respective outputs are spliced, weighted summed, or fused in other ways; in the series structure, the output of the previous parameterized quantum circuit is used as the input of the next parameterized quantum circuit, constituting a step-by-step quantum feature transformation; in the series-parallel hybrid structure, some of the multiple quantum circuits are calculated in parallel, while others are calculated step-by-step; in the hierarchical structure, lower-level quantum circuits are used to extract basic quantum features, and higher-level quantum circuits are used to further extract the fused higher-order quantum features.

[0064] Furthermore, this invention does not require all gating modules to be configured with multiple parameterized quantum circuits. In one embodiment, multiple parameterized quantum circuits are configured only in the forget gate, while other gating modules use classical computation methods, which can still improve the SOH prediction accuracy. In another embodiment, multiple parameterized quantum circuits are configured only in the input and output gates, while the forget gate and candidate state gate use classical computation. This invention does not require all gating modules to be configured with quantum circuits; as long as at least one gating module is configured with multiple quantum circuits, the model's expressive power can be enhanced to a certain extent.

[0065] Step 7: Establish quantum gating coupling relationships between different gating modules This invention establishes a quantum gating coupling mechanism between different gating modules, enabling quantum features generated by one gating module to participate in the computation of other gating modules. In this embodiment, chain-like quantum feature transfer is used as the reference coupling method.

[0066] In one specific embodiment, a (chain-like) quantum feature transfer method is used between different gating modules. That is, the comprehensive quantum feature output by the previous gating module is used as an additional input to the next gating module and fused with the comprehensive quantum feature generated by the next gating module itself. Specifically, in the QLSTM structure, the comprehensive quantum feature generated by the forget gate can be used as coupling information for the input gate computation, the comprehensive quantum feature generated by the input gate can be used as coupling information for the candidate state generation module, and the comprehensive quantum feature generated by the candidate state generation module can be used as coupling information for the output gate computation; that is, the quantum circuit measurement output or comprehensive quantum feature in one gating module is used as an additional input to another gating module. Let the first... a The combined quantum characteristics output by each gated module are The integrated quantum characteristic generated by the b-th gating module itself is Then the fused gating feature of the b-th gating module can be expressed as:

[0067] in, The fusion function can be a concatenation, weighted summation, linear mapping, nonlinear mapping, or attention fusion. The b-th gating module is based on the fused gating features. Calculate the corresponding gating function value.

[0068] Specifically: The forget gate, as the initial gating module in the chain-coupled structure, does not receive coupling information from preceding gating modules. Based on the outputs of its two parallel parameterized quantum circuits, the forget gate generates a comprehensive quantum feature, which is then transmitted as coupling information to the input gate. When calculating its own gating value, the input gate utilizes both the comprehensive quantum feature generated by its two parallel parameterized quantum circuits and the comprehensive quantum feature transmitted from the forget gate, fusing the two together. Thus, the input gate can incorporate quantum information from the forget gate during its computation. When calculating its candidate state value, the candidate state gate receives the comprehensive quantum feature transmitted from the input gate and fuses it with its own comprehensive quantum feature. The output gate receives the comprehensive quantum feature transmitted from the candidate state gate and fuses it with its own comprehensive quantum feature. Therefore, a hierarchically transmitted quantum gating coupling structure is formed between the forget gate, input gate, candidate state gate, and output gate. The subsequent gating module can not only use the comprehensive quantum features extracted by its own quantum circuit, but also integrate the comprehensive quantum features transmitted from the previous gating module, thereby realizing information interaction and collaborative computing between different gating modules.

[0069] Through the above method, this invention can transfer and fuse battery degradation-related quantum features extracted from different gating modules across gating modules, enabling the model to more fully express the coupling relationship between capacity decay, voltage change, current response, temperature change, cycle count, and historical state. Specifically: It should be noted that this step employs a direct concatenation of chained quantum feature transfers. That is, the combined quantum features output by the preceding gating module are directly concatenated with the combined quantum features of the subsequent gating module itself, without introducing any trainable parameters. This embodiment uses this method.

[0070] In addition to the above, inter-gate coupling can also employ other mechanisms such as parameter sharing, measurement feedback, and gating modulation, and is not limited to the quantum feature transfer method in this embodiment. However, it should be noted that this introduces additional inter-gate coupling parameters, which can also participate in subsequent joint optimization. Specifically: the parameter sharing coupling mechanism allows the parameterized quantum circuits in different gating modules to share some trainable parameters; the measurement feedback coupling mechanism uses the quantum measurement results of one gating module to adjust the input, fusion weights, or parameter settings of other gating modules; the gating modulation coupling mechanism uses the quantum features generated by one gating module to modulate the gating function value or candidate state calculation of another gating module.

[0071] In other implementations, the quantum-gated coupling mechanism may also employ at least one of the following methods: Quantum gating coupling mechanisms can also include at least one of quantum parameter sharing, measurement feedback, and gating modulation. Through this coupling mechanism, different gating modules are no longer independent of each other, but form a collaborative computing structure with information interaction relationships.

[0072] In one embodiment, the quantum gating coupling mechanism is parameter-sharing coupling. Specifically, the parameterized quantum circuits in different gating modules share trainable parameters, enabling multiple gating modules to form parameter-level correlations during training. Let the first... a The parameterized quantum circuit parameters in each gated module are: The parameterized quantum circuit parameters in the b-th gated module are: Then the two can be expressed as:

[0073]

[0074] in, Quantum circuit parameters shared between different gating modules. and These are the private parameters for the a-th and b-th gating modules, respectively. During training, these parameters are shared. The overall loss function is updated uniformly, thereby enabling parameter coupling and collaborative optimization between different gating modules.

[0075] In another embodiment, the quantum gating coupling mechanism is measurement feedback coupling. Specifically, the quantum characteristics of one gating module are used to adjust the inputs, fusion weights, or parameter settings of other gating modules. Let the first... a The quantum characteristics of each gated module are represented as follows: Then it can be based on Generate the adjustment factor that acts on the b-th gating module:

[0076] in, and For trainable parameters, Activation function. Regulation factor. It can be used to adjust the input characteristics, quantum feature fusion weights, or parameterized quantum circuit parameters of the b-th gating module. For example, the fusion characteristic of the b-th gating module can be expressed as:

[0077] Therefore, the first a The quantum measurement results of the first gated module can be fed back to influence the calculation process of the b-th gated module.

[0078] In another implementation, the quantum gating coupling mechanism is gated modulation coupling. Specifically, the quantum features generated by one gating module are used to modulate the gating function value or candidate state calculation of another gating module. Let the first... a The quantum features generated by each gated module are The fusion feature of the b-th gating module itself is Then the gating function value of the b-th gating module can be expressed as:

[0079] in, , and These are trainable parameters. In this way, the first... a The quantum characteristics of each gated module directly participate in the gating value calculation of the b-th gated module, thereby realizing information modulation and cooperative control between different gated modules. The same method can be used for candidate state calculation. A candidate state generation process is introduced.

[0080] Step 8: Calculate the gating function value and update the recursive state. The integrated quantum characteristics of each gating module are fused with the inter-gate coupling information to form a fused gating characteristic for gating function calculation. Specifically, the integrated quantum characteristics of the subsequent gating module are fused with the integrated quantum characteristics obtained from the previous gating module to obtain the fused gating characteristic. Subsequently, the gating function value corresponding to each gating module is calculated based on the fused gating characteristic.

[0081] Gating function values ​​are used to control the retention, forgetting, writing, or output of current input information, historical state information, or candidate state information during the recursive state update process. According to the state update rules of the selected gated recursive neural network, the recursive state at the current time step is updated using the gating function values. The recursive state can include hidden states, unit states, or other state variables.

[0082] In the QLSTM embodiment, the present invention updates the cell state and hidden state at the current time step based on the gating values ​​of the forget gate, input gate, candidate state generation module and output gate; in the QGRU embodiment, the software updates the hidden state based on the gating values ​​of the update gate, reset gate and candidate state generation module.

[0083] For a complete input sequence, the present invention performs the above calculations sequentially in chronological order and passes the updated recursive state of the current time step to the next time step until the input sequence is processed.

[0084] In one embodiment, based on the fused gating features of each gating module, the forget gate gating value, input gate gating value, candidate state value, and output gate gating value are calculated through the corresponding fusion mapping layer. The fusion mapping layer is a fully connected layer (which may contain bias), producing a second type of trainable parameter: the gating mapping parameter. Its function is to map the fused input vector into a gating value vector with the same dimension as the unit state. Other network structures capable of feature fusion and nonlinear mapping can also be used.

[0085] Subsequently, the unit state and hidden state are updated according to the state update rules of the Long Short-Term Memory network:

[0086]

[0087] in, Indicates the forget gate gating value. Indicates the input gate control value. Indicates the candidate state value. Indicates the output gate gating value. Indicates the cell state at the current time step. This represents the hidden state at the current time step, and ⊙ represents element-wise multiplication.

[0088] For an input sequence of length L=10, the recursive unit performs the above calculations for each time step and updates the hidden state at the current time step. Proceed to the next time step until the entire sequence has been processed.

[0089] Step 9: Output SOH prediction results and train model parameters In one embodiment, after completing the recursive computation of the entire input sequence, the hidden state of the last time step is input to the output mapping layer (where a third type of trainable parameter, the output layer parameter, is generated) to obtain the SOH prediction value corresponding to the target cycle period. The output mapping layer is used to map the battery degradation time-series features extracted by the quantum-enhanced recurrent neural network to the battery health state prediction result.

[0090] The predicted SOH value can be expressed as:

[0091] in, Indicates the first t SOH prediction values ​​for each cycle period This represents the hidden state corresponding to the last time step. and This is for outputting the mapping layer parameters.

[0092] The SOH prediction results can be the SOH prediction value for the current cycle, or the SOH prediction value for future cycles, or the SOH degradation trend curve.

[0093] The training phase of the model will be explained in detail below: This invention constructs a classical regression loss function based on the difference between the predicted SOH and the actual SOH label, and updates the model parameters using a quantum-classical hybrid optimization approach. Since the parameterized quantum circuit outputs classical numerical characteristics after measurement, and both the predicted SOH and the actual SOH label are classical values, the loss function can be the mean squared error, the mean absolute error, or a combination thereof.

[0094] In one specific implementation, the mean squared error (MSE) loss function is used during training, the Adam optimizer is used, and the initial learning rate is 0.001. Both classical network parameters and quantum circuit parameters are used as trainable parameters in gradient optimization, and the gradients are calculated and jointly updated using automatic differentiation techniques.

[0095] Let the training sample set be:

[0096] in, This represents the nth time series input sample. This represents the true SOH label corresponding to the nth sample. This indicates the number of training samples. After being input into a quantum-enhanced gated recurrent neural network, the SOH prediction value is obtained:

[0097] in, This represents the predicted SOH value for the nth sample. This represents a quantum-classical hybrid prediction model. This represents the set of parameters that need to be trained in the model.

[0098] The mean squared error loss function is expressed as:

[0099] in, Indicates based on parameter set The mean square error loss, This represents the predicted SOH value output by the model. This represents the true SOH label. By minimizing this loss function, the model's output SOH prediction value is made as close as possible to the true SOH label.

[0100] In one embodiment, the model parameter set It includes one or more of the following six types of parameters (depending on the specific implementation); Classical feature mapping parameters: Trainable parameters used to map combined input vectors to quantum circuit input features, such as the weights and biases of fully connected layers; Parameterized quantum circuit parameters: rotation angles of the rotating gates (RX, RY, RZ) in each parameterized quantum circuit; Intra-gate fusion parameters: Trainable parameters introduced when fusing the outputs of multiple parameterized quantum circuits within the same gated module. These parameters do not exist when using parameter-free fusion methods such as splicing. Inter-gate coupling parameters: Trainable parameters introduced when transmitting and fusing coupling information between different gate control modules. These parameters do not exist when using parameter-free coupling methods such as direct splicing. Gating mapping parameters: The weights and biases of the fully connected layers used to calculate the values ​​of each gating function based on the fused gating features; Output layer parameters: Output mapping layer weights and biases that map hidden states to SOH predictions.

[0101] It should be noted that when intra-gate fusion uses a simple splicing method, or when inter-gate coupling uses the quantum features of the preceding gating module directly as additional input to the following gating module, the splicing or transfer operation itself does not introduce additional trainable parameters. When using weighted fusion, linear mapping, nonlinear mapping, measurement feedback, gating modulation, or parameter sharing, the corresponding fusion weights, mapping parameters, feedback adjustment parameters, modulation parameters, or shared parameters can be used as model parameters to participate in joint optimization under a unified loss function. In this embodiment, splicing is used for intra-gate fusion, and direct splicing is used for inter-gate coupling; the parameter set... This includes: classical feature mapping parameters, all parameterized quantum circuit parameters, gating mapping parameters, and output layer parameters. There are no independently trainable intra-gate fusion parameters or inter-gate coupling parameters at this point.

[0102] The training process can be represented as:

[0103] in, This represents the set of parameters to be optimized in the model. Indicates in the parameter set The mean squared error loss function is as follows: This represents the search for the set of parameters that minimizes the loss function. This represents the set of optimal model parameters obtained after training.

[0104] The loss function can be mean squared error (MSE), mean absolute error (MAE), or a combination thereof; evaluation metrics can include MAE, RMSE, and R². Through training, this invention achieves joint optimization of classical network parameters and quantum circuit parameters, thereby improving the SOH prediction accuracy and model generalization ability of lithium-ion batteries.

[0105] In this embodiment, four cells (CS2_35, CS2_36, CS2_37, and CS2_38) from the CALCE dataset were used to conduct a cross-cell SOH prediction experiment. The experiment used a leave-one-out method for evaluation, that is, three cells were selected as the training set each time, and the remaining cell was used as the test set.

[0106] The specific division method is as follows: Group 1: CS2_36, CS2_37, and CS2_38 are used as the training set, and CS2_35 is used as the test set; The second set: CS2_35, CS2_37, and CS2_38 are used as the training set, and CS2_36 is used as the test set; The third group: CS2_35, CS2_36, and CS2_38 are used as the training set, and CS2_37 is used as the test set; The fourth set consists of CS2_35, CS2_36, and CS2_37 as the training set and CS2_38 as the test set.

[0107] Using the ordinary QLSTM model as a comparison model, the quantum-enhanced long short-term memory network model based on the multi-quantum gating coupling mechanism of this invention is denoted as MQ_LSTM. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are used as evaluation metrics. Smaller MAE and RMSE indicate lower prediction errors, and R² closer to 1 indicates a better fit to the SOH degradation trajectory.

[0108] The experimental results are shown in the table below:

[0109] Experimental results show that, in four leave-one-out validation sets, the MQ_LSTM model of this invention achieves lower MAE, RMSE and higher R² in most cases, proving the effectiveness of the multi-quantum circuit structure and inter-gate coupling mechanism.

[0110] Combination Figure 4 As shown, based on the same inventive concept, this invention also discloses a lithium-ion battery health state prediction system based on quantum gating coupling, used to execute any lithium-ion battery health state prediction method based on quantum gating coupling. To suit specific application scenarios, this embodiment includes both a data preprocessing module and a model optimization module, aiming to more intuitively demonstrate the technical solution of this application. Specifically, it includes: The data input module is used to acquire lithium-ion battery cycle operation data; The data preprocessing module is used to clean, normalize, and align the data from the loop. The health feature construction module is used to construct health features characterizing the battery degradation state based on preprocessed cyclic operation data. The time series sample construction module is used to construct health features into time series input samples according to a preset time window. A quantum-enhanced gated recurrent neural network module is configured to include multiple gate modules. For each time step in the time series input sample, the health feature vector of that time step is combined with the recurrent state of the previous time step to form a combined input vector. Multiple parameterized quantum circuits are configured within at least one gate module, and the combined input vector is input to each parameterized quantum circuit to obtain the quantum feature representation output by each parameterized quantum circuit. Multiple quantum feature representations within the same gate module are combined and fused to form the comprehensive quantum feature of that gate module at the current time step. A quantum gating coupling mechanism is established between different gate modules, fusing the comprehensive quantum feature of the subsequent gate module with the comprehensive quantum feature obtained from the previous gate module to obtain a fused gating feature. The gating function value is calculated based on the fused gating feature, and the recurrent state of the current time step is updated. The training optimization module is used to update model parameters based on the difference between the predicted health status values ​​and the true labels.

[0111] The output module is used to output the predicted health status of the lithium-ion battery after processing all time steps.

[0112] It should be noted that this invention is not limited to lithium-ion battery health status prediction. The quantum-enhanced gated recurrent neural network based on a multi-quantum gating coupling mechanism is also applicable to other scenarios requiring modeling and prediction of complex nonlinear time-series data, such as: industrial equipment condition monitoring and fault diagnosis (motors, fans, machine tools, etc.), energy system prediction (electricity load, photovoltaic / wind power generation, energy storage scheduling), financial time series analysis (stock prices, exchange rates, volatility), traffic flow prediction and intelligent transportation, medical and health monitoring (ECG, EEG, blood pressure, and other physiological signals), intelligent manufacturing process control, as well as weather forecasting and network traffic prediction. Any application that introduces multi-parameterized quantum circuits and inter-gate coupling mechanisms into the gated recurrent neural network to enhance time-series modeling capabilities falls within the scope of this invention.

[0113] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0114] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the state of health of lithium-ion batteries based on quantum-gated coupling, characterized in that, Includes the following steps: S1. Obtain lithium-ion battery cycle operation data, construct health features characterizing the battery degradation state based on the cycle operation data, and construct the health features as time series input samples according to a preset time window; the time series input samples contain multiple health feature vectors arranged in time steps; The lithium-ion battery cycle operation data includes one or more of the following: voltage, current, capacity, temperature, charging time, discharging time, and number of cycles. S2. Construct a quantum-enhanced gated recurrent neural network, wherein the quantum-enhanced gated recurrent neural network includes multiple gated modules; For each time step in the time series input sample, perform the following sub-steps: The health feature vector of this time step is combined with the recursive state of the previous time step to form the combined input vector of the current time step; Multiple parameterized quantum circuits are configured within at least one of the gated modules, and the combined input vector is input to the multiple parameterized quantum circuits respectively. Each parameterized quantum circuit independently performs quantum computation and outputs a quantum feature representation. The quantum feature representations output by multiple parameterized quantum circuits within the same gating module are combined and fused to form the comprehensive quantum feature of the gating module at the current time step; A quantum gating coupling mechanism is established between different gating modules. The comprehensive quantum features of the latter gating module at the current time step are fused with the comprehensive quantum features obtained from the former gating module at the current time step to obtain fused gating features. The gating function value of the gating module at the current time step is calculated based on the fused gating features, and the recursive state of the current time step is updated according to the gating function value. S3. Repeat the above processing for each time step until all time steps in the time series input sample have been processed, and output the predicted value of the lithium-ion battery health status.

2. The method according to claim 1, characterized in that, The parameterized quantum circuit includes a data encoding layer, a parameterized rotation layer, a quantum entanglement layer, and a measurement output layer, which are executed sequentially. The data encoding layer is used to map the input combined input vector into a quantum state; The parameterized rotation layer is used to apply a rotation gate with trainable parameters to the quantum state; The quantum entanglement layer is used to apply a controlled NOT gate or a controlled phase gate to the rotated quantum state to establish entanglement between qubits; The measurement output layer is used to perform Pauli-Z measurements on the entangled quantum state and output classical values ​​as the representation of the quantum features.

3. The method according to claim 1, characterized in that, The topology used when combining and fusing the quantum feature representations output by multiple parameterized quantum circuits within the same gating module is one of the following: Parallel structure: The input ends of the multiple parameterized quantum circuits receive the same combined input vector, and the output ends are calculated independently and then fused. Serial connection: The output of the previous parameterized quantum circuit is connected to the input of the next parameterized quantum circuit; Hybrid structure: At least two of the plurality of parameterized quantum circuits are connected in parallel, and their output terminals are then connected in series with at least one parameterized quantum circuit.

4. The method according to claim 1, characterized in that, The specific method for combining and fusing the quantum feature representations output by the multiple parameterized quantum circuits within the same gating module is as follows: the input ends of the multiple parameterized quantum circuits receive the same combined input vector, each calculates independently, and then the output results are directly spliced ​​together to form the comprehensive quantum feature of the gating module.

5. The method according to claim 1, characterized in that, The quantum-gated coupling mechanism includes at least one of the following: Quantum feature transfer: The comprehensive quantum features of the previous gating module at the current time step are used as the coupling information of the next gating module; Parameter sharing: Sharing trainable parameters of parameterized quantum circuits among different gating modules; Measurement feedback: The integrated quantum characteristics of one gating module are used to generate adjustment factors to adjust the inputs, fusion weights, or parameter settings of other gating modules; Gated modulation: Utilizing the comprehensive quantum characteristics of one gating module to modulate the gating function value or candidate state calculation of another gating module.

6. The method according to claim 1, characterized in that, The quantum gating coupling mechanism is a quantum feature transfer: according to a preset order, the comprehensive quantum feature output by the previous gating module at the current time step is spliced ​​and fused with the comprehensive quantum feature of the subsequent gating module itself; wherein, the initial gating module does not receive the preceding coupling information, and its fused gating feature is calculated only based on its own comprehensive quantum feature.

7. The method according to claim 6, characterized in that, When the quantum-enhanced gated recurrent neural network is a quantum-enhanced long short-term memory network, the quantum feature transfer is performed in the order of forget gate, input gate, candidate state gate, and output gate: The comprehensive quantum features output by the forget gate at the current time step are used as the coupling information for the input gate computation. The comprehensive quantum features output by the input gate are used as the coupling information for the candidate state generation module. The comprehensive quantum features output by the candidate state generation module are used as the coupling information for the output gate computation. Among them, the forget gate, as the starting gating module, does not receive preceding coupling information.

8. The method according to claim 1, characterized in that, The health characteristics include one or more of the following: capacity characteristics, voltage characteristics, current characteristics, temperature characteristics, cycle count characteristics, charge / discharge time characteristics, energy characteristics, curve morphology statistical characteristics, and incremental capacity analysis characteristics. The capacity characteristic reflects the decay of the battery's usable capacity, the voltage characteristic reflects the change in voltage plateau, the current characteristic reflects the decay of current during the constant voltage stage, the charge / discharge time characteristic reflects the change in charge / discharge time, the curve morphology statistical characteristics include the slope, skewness, and entropy changes of the voltage curve, current curve, or capacity curve, and the incremental capacity analysis characteristics include the peak position, peak amplitude, or peak area changes in the dQ / dV curve.

9. The method according to claim 1, characterized in that, It also includes the step of training the quantum-enhanced gated recurrent neural network: Step A: Input the time series input samples into the quantum-enhanced gated recurrent neural network to obtain the health status prediction value; Step B: Calculate the mean squared error or mean absolute error between the predicted health status value and the preset true label, and use it as the loss function value; Step C: Using automatic differentiation, calculate the gradient of the loss function value with respect to the following trainable parameters: The angular parameters of all rotating doors within each parameterized quantum circuit; The weights and biases of the fully connected layer in the quantum-enhanced gated recurrent neural network used to map fused gated features to gated function values; The weights and biases of the output layer in the quantum-enhanced gated recurrent neural network that map the final recurrent state to a health state prediction value; Step D: Update all trainable parameters mentioned above based on the calculated gradients. Step E: Repeat steps A through D until the loss function value converges or the preset number of training rounds is reached.

10. A lithium-ion battery health state prediction system based on quantum-gated coupling, used to execute the lithium-ion battery health state prediction method based on quantum-gated coupling as described in any one of claims 1-9, characterized in that, include: The data input module is used to acquire lithium-ion battery cycle operation data; A health feature construction module is used to construct health features characterizing the battery degradation state based on the cyclic operation data. A time series sample construction module is used to construct the health features as time series input samples according to a preset time window; A quantum-enhanced gated recurrent neural network module is configured to include multiple gate modules, and for each time step in the time series input sample, the health feature vector of the time step is combined with the recurrent state of the previous time step to form a combined input vector; multiple parameterized quantum circuits are configured in at least one of the gate modules, and the combined input vector is respectively input to the multiple parameterized quantum circuits to obtain the quantum feature representation output by each parameterized quantum circuit; Multiple quantum feature representations within the same gating module are combined and fused to form the comprehensive quantum feature of the gating module at the current time step; a quantum gating coupling mechanism is established between different gating modules to fuse the comprehensive quantum feature of the latter gating module itself with the comprehensive quantum feature obtained from the former gating module to obtain the fused gating feature; the gating function value is calculated based on the fused gating feature and the recursive state of the current time step is updated. The output module is used to output the predicted health status of the lithium-ion battery after processing all time steps.

Images