A method for predicting the state of health of a power battery of a drone

Abstract

This invention relates to the field of battery state prediction technology and discloses a method for predicting the health status of UAV power batteries. The key technical points of this method are: data acquisition and quantum encoding; quantum feature extraction; construction of a quantum-heuristic deep learning model; model training and optimization; and health status prediction and assessment. Through quantum encoding, a quantum-heuristic model architecture, and a dynamic optimization mechanism, this method overcomes the limitations of traditional methods in modeling complex nonlinear relationships and local optimization, providing a high-precision and robust solution for the health management of UAV power batteries, and has significant engineering application value.

Description

Technical Field

[0001] This invention relates to the field of battery state prediction technology, and more specifically, to a method for predicting the health status of a drone's power battery. Background Technology

[0002] Battery health prediction methods are typically based on traditional machine learning algorithms (such as support vector machines and random forests) or statistical models (such as Kalman filtering). However, these methods have significant limitations when dealing with the complex nonlinear relationships between multiple battery parameters (voltage, current, temperature, number of charging cycles, etc.): insufficient data representation capabilities, low model optimization efficiency, and feature engineering relying on experience.

[0003] To address the aforementioned problems, most complex nonlinear issues are solved using quantum computing techniques. The superposition and entanglement of quantum states can efficiently characterize the correlations of multidimensional data, while quantum heuristic algorithms (such as quantum genetic algorithms) exhibit significant advantages in global optimization and escaping local optima. However, current technologies have not yet effectively combined quantum computing with deep learning frameworks to solve the battery health prediction problem.

[0004] Therefore, the present invention provides a method for predicting the health status of a drone's power battery, thereby improving the aforementioned technical problems. Summary of the Invention

[0005] This disclosure aims to address the shortcomings of existing technologies by providing a method for predicting the health status of drone power batteries. The present invention is based on a quantum-inspired deep learning prediction algorithm to predict the health status of drone power batteries.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: a method for predicting the health status of a drone's power battery, comprising the following steps:

[0007] S1, Data Acquisition and Quantum Encoding;

[0008] S2, Quantum Feature Extraction;

[0009] S3. Constructing a quantum-inspired deep learning model;

[0010] S4. Model training and optimization;

[0011] S5. Health Status Prediction and Assessment.

[0012] As a preferred embodiment of the present invention, the data acquisition and quantum encoding process is as follows:

[0013] First, the battery voltage is collected in real time by the sensors of the drone's battery management system. Current ,temperature Number of charging cycles Then, each parameter is quantized and encoded, converting it into its corresponding quantum state. , , , ; and combine the four quantum states into a global quantum state. ;in, This represents the tensor product operation.

[0014] As a preferred embodiment of the present invention, the calculation formula for quantized encoding is as follows:

[0015]

[0016] in, , Indicates voltage, Indicates current, Indicates temperature, Indicates the number of charging cycles; and This indicates the lower and upper limits of the value range for each parameter.

[0017] As a preferred embodiment of the present invention, the quantum feature extraction process is as follows:

[0018] First, the relative rate of change of quantum states of each parameter is calculated to characterize the quantum state difference between adjacent time steps; then, the quantum entanglement degree of each pair of parameters is calculated to extract the multi-parameter quantum entanglement correlation characteristics; finally, the quantum state probability distribution entropy is calculated to quantify the uncertainty of the parameters.

[0019] As a preferred embodiment of the present invention, the process of constructing a quantum-inspired deep learning model is as follows:

[0020] First, a deep neural network containing qubit neurons is constructed, with the neuron state represented as a superposition of multiple ground states. Then, the quantum tunneling effect is introduced to improve the weight update rule. The neuron collaboration is enhanced through the quantum entanglement mechanism. Finally, a quantum measurement-inspired attention mechanism is used to dynamically allocate feature weights.

[0021] As a preferred embodiment of the present invention, the process of constructing a deep neural network containing qubit neurons is as follows: assuming the neuron receives voltage... Current ,temperature Number of charging cycles The encoded quantum state inputs are respectively , , , For a person with The state of a neuron with input channels and its qubit neuron. for:

[0022]

[0023] in, It is the probability amplitude, satisfying , yes The ground state of 1 qubit;

[0024] The process of introducing the quantum tunneling effect to improve the weight update rule is as follows: Let the neuron With neurons The connection weights between them are The loss function is In the In the next iteration, the weight update calculation formula is:

[0025]

[0026] in, It's the learning rate. It is a reference value for the weight. It is a parameter that controls the probability of tunneling;

[0027] The process of enhancing neuronal synergy through quantum entanglement is as follows: Let the neuron... and neurons Through weight Connected, neurons The output is Neuron The input is After considering quantum entanglement, neurons The update calculation formula is:

[0028]

[0029] in, It is an activation function. Neuron and The entanglement coefficient between them;

[0030] The process of dynamically allocating feature weights using a quantum measurement-inspired attention mechanism is as follows: Let the feature vector input to the model be... ,in Corresponding voltages Current ,temperature Number of charging cycles Other features; feature vectors processed by the attention mechanism for:

[0031]

[0032] in, It is the attention weight, inspired by the quantum measurement process, and its calculation formula is:

[0033]

[0034] in, and Is with characteristics and The corresponding quantum state, It is a temperature parameter used to control the smoothness of weight distribution.

[0035] As a preferred embodiment of the present invention, the model training and optimization process comprises: using quantum rotation gates and mutation operations to adjust the qubit states to optimize the model parameters; and dynamically adjusting the regularization parameters. Use root mean square error (RMSE) and coefficient of determination. and quantum state fidelity Comprehensive evaluation of model performance.

[0036] As a preferred technical solution of the present invention, the process of optimizing the model parameters by adjusting the state of the qubits using quantum rotation gates and mutation operations is as follows:

[0037] For the Let a given individual in the population have a parameter represented by one of its qubits. The corresponding quantum state is According to the update calculation formula of the quantum rotating gate:

[0038] =

[0039] in, It is the rotation angle;

[0040] To increase population diversity, a quantum mutation operation is introduced: with a certain mutation probability... Mutate a qubit; for a qubit If a mutation occurs, an angle will be randomly selected. And perform the following operations:

[0041] =

[0042] Dynamically adjust regularization parameters The formula for calculation is:

[0043] in, and It is the initial value. This represents the variance of the predicted results.

[0044] As a preferred embodiment of the present invention, the process of health status prediction and assessment is as follows: first, the newly acquired battery data is encoded into quantum states. And input it into the trained model to obtain the predicted value. ; and then introduce a comprehensive evaluation indicator. The root mean square error (RMSE) and coefficient of determination are... Quantum state fidelity Distance to the overall quantum state A comprehensive assessment will be conducted.

[0045] As a preferred technical solution of the present invention, the predicted value The calculation formula is: ;

[0046] The overall quantum state distance The calculation formula is:

[0047]

[0048] in, , , , These are weighting coefficients, satisfying... ;

[0049] The comprehensive evaluation indicators The calculation formula is:

[0050]

[0051] in,

[0052] In summary, the present invention has the following beneficial effects:

[0053] Firstly, data representation capabilities are significantly enhanced: By encoding parameters such as voltage, current, temperature, and number of charging cycles into quantum states, the superposition and entanglement of quantum states are fully utilized to explore the nonlinear relationships between parameters. For example, the coordinated changes in voltage and current can be quantified through quantum entanglement characteristics, avoiding the information loss problem of traditional numerical encoding. Transient anomalies of parameters (such as sudden voltage drops and temperature changes) can be captured in real time through the relative change rate of quantum states, improving the sensitivity to the dynamic aging process of batteries.

[0054] Secondly, improved model optimization efficiency and global search capability: The weight update formula introduces an exponential decay term, reducing the update step size when the model is far from the optimal solution and accelerating convergence when it is close to the optimal solution, thus improving training speed compared to the traditional gradient descent method. Adjusting parameter direction through quantum rotation gates and increasing population diversity through mutation operations significantly improves parameter search efficiency.

[0055] Third, prediction accuracy and robustness are significantly improved: By comprehensively evaluating the synergistic effect and uncertainty among parameters through quantum entanglement correlation characteristics and probability distribution entropy, the regularization parameters are dynamically adjusted based on prediction uncertainty, effectively balancing model complexity and generalization ability.

[0056] Fourth, comprehensive verification using multiple indicators: Quantum state fidelity and quantum state distance are introduced to quantify the difference between the predicted results and the actual state at the quantum level, combined with classical indicators RMSE and... This forms a "quantum-classical" dual-dimensional evaluation system to ensure the reliability of the prediction results. Attached Figure Description

[0057] Figure 1 A flowchart of a method for predicting the health status of a drone's power battery, provided as an embodiment of the present invention. Detailed Implementation

[0058] The present application 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 application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] It should be noted that, unless there is a conflict, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. In addition, the terms "first," "second," and "third" used herein do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0061] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0062] Furthermore, the technical features involved in the various embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0063] This disclosure aims to address the problem of the lack of an effective combination of quantum computing and deep learning frameworks in the prior art for battery health prediction. Therefore, this disclosure proposes a method for predicting the health status of UAV power batteries. Through quantum coding, a quantum heuristic model architecture, and a dynamic optimization mechanism, it overcomes the limitations of traditional methods in modeling complex nonlinear relationships and local optimization, providing a high-precision and robust solution for UAV power battery health management, and has significant engineering application value.

[0064] Please refer to Figure 1 , Figure 1 A flowchart of a method for predicting the health status of a drone's power battery according to an embodiment of this disclosure is shown. The overall process mainly includes the following five steps:

[0065] S1: Battery data collection and quantum coding.

[0066] S1.1 Battery Data Collection.

[0067] The battery voltage is collected in real time through the hardware layer sensors of the drone's battery management system (BMS). Current ,temperature Number of charging cycles .

[0068] The battery data mentioned above are all traditional numerical data. In order to fully utilize the superposition and entanglement properties of quantum states in quantum computing and to better handle the complex relationships between these data, it is necessary to quantize and encode the battery data.

[0069] Quantum encoding of S1.2 battery data.

[0070] (1) Voltage Quantization encoding.

[0071] Let voltage The range of values ​​for is , The value range will vary depending on the type, specifications, and usage conditions of the battery; for example, common consumer-grade drone lithium batteries, It's generally around 2.5 - 3.0V. Typically, the voltage is between 4.2 and 4.35V; however, in practical applications, the exact range needs to be determined based on the specific battery's technical parameters and the operating environment. This indicates the minimum voltage that can be reached during the operation of a drone battery. It reflects the lowest voltage level of the battery under certain extreme or specific operating conditions, such as the voltage when the battery is nearly depleted. This indicates the maximum voltage that can be achieved, typically corresponding to the highest voltage value when the battery is fully charged or under specific operating conditions. =2.5V, =4.35V. (The voltage is...) A quantum state encoded as a qubit The calculation formula is as follows:

[0072]

[0073] when hour, ;when hour, .exist and When the quantum state is in between, and The superposition state, the superposition ratio is determined by the voltage The position is determined by the range of values.

[0074] (2) Current quantization encoding

[0075] Assume current The range of values ​​for is Encode it as a quantum state of a qubit. The calculation formula is as follows:

[0076]

[0077] when hour, ;when hour, .exist and When the quantum state is in between, and The superposition state, the superposition ratio is determined by the current. The position is determined by the range of values.

[0078] (3) Temperature quantization encoding

[0079] Set temperature The range of values ​​for is Encode it as a quantum state of a qubit. The calculation formula is as follows:

[0080]

[0081] when hour, ;when hour, .exist and When the quantum state is in between, and The superposition of states, the superposition ratio is determined by temperature. The position is determined by the range of values.

[0082] (4) Number of charging cycles quantization encoding

[0083] Set the number of charging cycles The range of values ​​for is Encode it as a quantum state of a qubit. The calculation formula is as follows:

[0084]

[0085] when hour, ;when hour, .exist and When the quantum state is in between, and The superposition state, the superposition ratio is determined by the number of charging cycles. The position is determined by the range of values.

[0086] (5) Multivariate data combination

[0087] The four qubits mentioned above are combined into a four-qubit system to represent the overall quantum state of this battery data. :

[0088]

[0089] in This represents the tensor product operation. This four-qubit system can fully reflect information from four data points: voltage, current, temperature, and number of charging cycles.

[0090] If the states of each qubit in this four-qubit system are as follows:

[0091]

[0092]

[0093]

[0094]

[0095] The overall quantum state of the four-qubit system The matrix is:

[0096]

[0097] The tensor product operation expands and combines the state matrices of individual qubits according to certain rules. Performing a tensor product operation on the state matrices of four qubits ultimately yields a 16-dimensional vector (because...). This fully represents the comprehensive information of the four parameters.

[0098] Because of the superposition and entanglement of quantum states in quantum computing, quantizing data and converting traditional numerical data into quantum state representation can better handle the complex relationships in the data.

[0099] S2: Quantum feature extraction.

[0100] Feature engineering is a crucial part of quantum-inspired deep learning prediction algorithms; it can analyze the acquired voltage... Current ,temperature Number of charging cycles Representative features are extracted from the data to improve the accuracy of predicting the health status of drone power batteries. The following are the feature extraction and calculation methods based on this data.

[0101] S2.1 Characteristics of the relative rate of change of quantum states.

[0102] Assuming at time step At that time, voltage Current ,temperature Number of charging cycles The encoded quantum states are respectively , , , ; at time step At that time, the corresponding quantum states are respectively , , , .

[0103] For voltage Its relative rate of change of quantum state The formula for calculation is:

[0104]

[0105] Similarly, for current ,temperature Number of charging cycles Relative rate of change of quantum state , , They are respectively:

[0106]

[0107]

[0108]

[0109] The relative rate of change of a quantum state reflects the magnitude of the change in the quantum state of each parameter within adjacent time steps. Taking voltage as an example... For example, if A large value indicates a significant voltage change between these two time steps, suggesting abnormal charging and discharging behavior of the battery. Regarding current... ,temperature and number of charging times Similarly, by monitoring these relative rates of change, dynamic changes in battery status can be captured in a timely manner, providing important clues for predicting battery health status.

[0110] S2.2 Multi-parameter quantum entanglement correlation characteristics.

[0111] Considering voltage and current Two-parameter quantum entanglement correlation characteristics First, define the joint quantum state. Its density matrix .right Partial trace finding yields the reduced density matrix. and Negativity is used to measure the degree of entanglement in quantum states. The following can be calculated using the negative logarithmic entanglement metric of quantum entanglement:

[0112]

[0113] in, yes Regarding subsystems (e.g., a partial transpose of a voltage-corresponding subsystem) The trace norm of a matrix is ​​denoted by .

[0114] Similarly, the voltage can be calculated. With temperature Association features ,Voltage With the number of charging times Association features Current With temperature Association features Current With the number of charging times Association features ,temperature With the number of charging times Association features .

[0115] Multi-parameter quantum entanglement correlation characteristics are used to measure the degree of intrinsic correlation between different parameters. For example, it reflects the voltage. and current The entangled relationship between them. During battery operation, voltage and current are usually interrelated; if A larger value indicates a close correlation between the two, suggesting a pattern of coordinated change; conversely, if... Smaller values ​​indicate a weaker correlation between them.

[0116] In the above embodiments, for two-body quantum states (here , (corresponding to voltage and current subsystems respectively), its partial transpose The negative logarithmic degree of entanglement is defined as .

[0117] From a mathematical derivation perspective:

[0118] First, partial transpose is a transpose operation performed on a specific subsystem of the density matrix. For Its elements ( After partial transpose, if regarding the subsystem If the (voltage subsystem) is transposed, the new element becomes (Only for subsystems) (Related indicators transposed).

[0119] Then, trace norm Defined as the sum of the singular values ​​of a matrix. For a given matrix... Its singular value is ( yes The square root of the eigenvalues ​​of the conjugate transpose of (the eigenvalues ​​of the ...

[0120] For separable states (non-entangled states), the trace norm after partial transpose As for entangled states, .pass This formula can quantify the degree of entanglement. The larger the value, the stronger the entanglement. In battery parameter correlation, it means that the correlation between voltage and current is closer.

[0121] S2.3 Features based on the entropy of quantum state probability distribution

[0122] For voltage Encoded quantum state Its probability distribution is , .Voltage quantum state probability distribution entropy The calculation formula is:

[0123]

[0124] Similarly, for current ,temperature Number of charging cycles Its quantum state probability distribution entropy , , They are respectively:

[0125]

[0126]

[0127]

[0128] The entropy of the quantum state probability distribution describes the degree of uncertainty of each parameter in the quantum state. Taking voltage as an example... For example, if A larger value indicates a more uniform probability distribution of the quantum states corresponding to the voltage, meaning the voltage state has greater uncertainty and is in an unstable operating state; if... A smaller value indicates a relatively certain voltage state and a more stable operating mode. By analyzing the quantum state probability distribution entropy of each parameter, the stability of each battery parameter can be assessed, thus providing a basis for predicting the battery's health status.

[0129] S3: Constructing quantum-inspired deep learning models.

[0130] The specific steps for constructing a quantum-inspired deep learning model are as follows:

[0131] S3.1 Quantum bit neuron state.

[0132] In quantum-inspired deep neural networks, neurons are introduced with qubits, and their states are represented by a superposition of multiple ground states. A quantum-inspired classical neuron is chosen as the implementation method. In this approach, when a quantum state is input into the neural network, it is transmitted... As two real numbers as input. This input method preserves the superposition information of the quantum states. and Corresponding to quantum states in and The probability amplitude in the ground state can more comprehensively reflect the characteristics of a quantum state. If it is only obtained through measurement... As a scalar input, phase information is lost, the superposition property of quantum states cannot be fully utilized, and it is not conducive to the model learning complex nonlinear relationships.

[0133] Assuming the neuron receives voltage Current ,temperature Number of charging cycles The encoded quantum state inputs are respectively , , , For a having The state of a neuron with input channels and its qubit neuron. for:

[0134]

[0135] in, It is the probability amplitude, satisfying , yes The ground state of a quantum bit.

[0136] With voltage For example, if only voltage input is considered (i.e. ),but , here and Based on the quantization encoding of voltage Determined, that is , Similarly, for current... ,temperature Number of charging cycles Similarly, the corresponding quantum bit neuron state components can be determined.

[0137] This quantum bit neuron state representation method enables neurons to process different combinations of multiple inputs simultaneously, making them more flexible than traditional neurons. It can capture complex nonlinear relationships between voltage, current, temperature, and number of charging cycles, providing richer feature representations for subsequent calculations.

[0138] S3.2 Weight update based on quantum tunneling effect.

[0139] In traditional neural networks, weight updates often employ gradient descent. This patent introduces the quantum tunneling effect to improve the weight update calculation formula. Let the neuron... With neurons The connection weights between them are The loss function is In the In the next iteration, the weight update calculation formula is:

[0140]

[0141] in, It's the learning rate. It is a reference value for the weight. It is a parameter that controls the probability of tunneling.

[0142] In handling voltage Current ,temperature Number of charging cycles When considering relevant weights, use weights related to voltage. For example, in each iteration, the weights are updated according to the above calculation formula. If the current weights... Far from the reference value , The value of will decrease, resulting in a smaller weight update step size, which prevents the model from over-adjusting when it is far from the optimal solution; when the weights are close to the reference value, the update step size is close to the step size of traditional gradient descent, ensuring that the model can converge quickly when it is close to the optimal solution.

[0143] quantum neuron state With weight The interaction between neurons is reflected in the computational process of neurons. Taking the first... Taking a layer neuron as an example, its input quantum state is For a single neuron Its output quantum state The calculation involves weights Weighted summation:

[0144]

[0145] In the formula, the weights The input quantum state is determined For output quantum state The degree of contribution. Different weight values ​​will adjust the relative importance of each input quantum state component.

[0146] The introduction of the quantum tunneling effect allows the model to escape local optima during training. When predicting the health status of drone power batteries, the complexity of the data makes it easy to get trapped in local optima. This weight update method helps the model explore the solution space more effectively, find better weight combinations, and thus improve the model's prediction accuracy.

[0147] Neuronal synergy under the S3.3 quantum entanglement mechanism.

[0148] To enhance information exchange and synergy among different neurons, a quantum entanglement mechanism is introduced. Assume that neurons... and neurons Through weight Connected, neurons The output is Neuron The input is (Including voltage) Current ,temperature Number of charging cycles (Related characteristic information), after considering quantum entanglement, neurons The update calculation formula is:

[0149]

[0150] in, It is an activation function. Neuron and The entanglement coefficient between them reflects the degree of correlation between them.

[0151] To process voltage and current Taking two neurons as an example, if one neuron mainly processes voltage information and the other mainly processes current information, when their entanglement coefficient... When the voltage is large, the output of the neuron processing voltage information will have a significant impact on the neuron processing current information, and vice versa. This allows the model to better capture the cooperative relationship between voltage and current.

[0152] Quantum entanglement breaks the traditional, relatively independent information transmission mechanism between neurons, enabling them to share more information and collaboratively process complex data relationships. When predicting battery health, this mechanism helps models comprehensively consider the interactions between multiple factors such as voltage, current, temperature, and the number of charging cycles, thus improving the overall performance of the model.

[0153] Inspired by quantum measurement in the S3.4 attention mechanism.

[0154] The attention mechanism allows the model to automatically focus on the features most important for predicting battery health. Let the feature vector input to the model be... ,in Corresponding voltages Current ,temperature Number of charging cycles And other features. The feature vector after attention mechanism processing. for:

[0155]

[0156] in, It is the attention weight, inspired by the quantum measurement process, and its calculation formula is:

[0157]

[0158] here, and Is with characteristics and The corresponding quantum state, This is a temperature parameter used to control the smoothness of weight distribution. (Based on voltage...) and current For example, It is voltage corresponding quantum state , It is electric current corresponding quantum state , This represents the degree of overlap between two quantum states. A higher degree of overlap indicates a stronger correlation between the two features, and the corresponding attention weight... The larger.

[0159] Through this quantum measurement-inspired attention mechanism, the model can dynamically allocate weights, giving higher attention to features more relevant to battery health and lowering the weights of less relevant features. This helps the model focus on key information and improve the accuracy of battery health predictions.

[0160] S4: Model training and optimization.

[0161] Through model training and optimization, the model can more accurately predict the health status of the drone's power battery. The specific steps are as follows:

[0162] S4.1 Quantum Heuristic Optimization Algorithm.

[0163] Based on the quantum genetic algorithm, model parameters are optimized by simulating the evolutionary process of quantum states. In the quantum genetic algorithm, an individual is represented by multiple qubits, and the state of each qubit is... ,in For voltage Current ,temperature Number of charging cycles Relevant model parameters (such as weights) ), which is encoded into a string of qubits.

[0164] (1) Quantum Revolving Door Update

[0165] Quantum rotation gates are used to adjust the state of qubits, thereby updating model parameters. For the ... For a given individual in the population (a set of model parameters), suppose one of the qubits represents a parameter of... The corresponding quantum state is According to the update calculation formula of the quantum rotating gate:

[0166] =

[0167] in, It is the rotation angle, a function of the current model prediction error (the difference between the current error and the actual battery health state) and the degree to which this parameter affects the prediction result. (This is related to voltage.) Related weights For example, let the battery health state predicted by the model be... The actual health status is It can be defined as:

[0168]

[0169] in, It is a symbolic function. , and This is a hyperparameter used to balance the effects of prediction error and parameter gradient on the rotation angle. Thus, by adjusting the angle of the quantum rotation gate, the voltage-related weights are updated in a direction that reduces prediction error.

[0170] (2) Quantum mutation operation

[0171] To increase population diversity, a quantum mutation operation is introduced. This operation generates mutations with a certain probability. Mutate a qubit. For a single qubit... If a mutation occurs, an angle will be randomly selected. And perform the following operations:

[0172] =

[0173] For the current Relevant parameters (such as weights) The qubits corresponding to these qubits also follow the above rules when they mutate. Mutation operations help the model escape local optima, explore a wider solution space, and improve the model's generalization ability.

[0174] Quantum heuristic optimization algorithms leverage the properties of quantum states, using quantum rotation gates and quantum mutation operations to iteratively update model parameters related to voltage, current, temperature, and the number of charge cycles. This method can more effectively search for optimal parameter combinations and, compared to traditional optimization algorithms, better handle complex nonlinear problems, improving the accuracy of the model's prediction of battery health status.

[0175] S4.2 Regularization processing.

[0176] To prevent overfitting during training, a regularization term is introduced. Regularization (weight decay) in the loss function Add regularization terms The new loss function for:

[0177]

[0178] in, It is a regularization parameter. It is the sum of the squares of all model weights. When considering voltage... Current ,temperature Number of charging cycles When considering the relevant weights, this formula encompasses the weights corresponding to all these parameters. For the voltage-related weights... Current-related weights Temperature-related weights Weights related to the number of charging cycles All are included middle.

[0179] The predictive uncertainty of a model is related to its complexity. To balance the model's fitting ability and generalization ability, the regularization parameter is dynamically adjusted. The uncertainty of model predictions is expressed as the variance of the prediction results. To measure, regularization parameters Dynamically adjust according to the following formula:

[0180]

[0181] in, and It is the initial value. This is the ratio of the average fluctuation of the model's prediction error on the initial training set to the number of model parameters. This represents the variance of the prediction results for the initial training set.

[0182] Alternatively, empirical values ​​from similar models in related fields can be referenced. If relatively mature models already exist for other similar battery state predictions or complex system modeling... The value can be fine-tuned based on the characteristics of the current model and data. The purpose of this setting is to dynamically adjust the regularization parameters. Provide a reasonable starting point so that the model can be appropriately regularized in the early stages of training based on the uncertainty of the data, thus balancing the model's fitting ability and generalization ability.

[0183] Regularization, by constraining weights, prevents the model from overfitting the training data. Dynamically adjusting the regularization parameters can adaptively balance the model's fitting and generalization abilities, improving its performance under different data conditions. When predicting battery health, this dynamically adjusted regularization method helps the model more accurately reflect the battery's true state.

[0184] S4.3 Model Evaluation.

[0185] The training process of the model is monitored using a validation set, and the model that performs best on the validation set is selected as the final model. Root mean square error, coefficient of determination, and quantum state fidelity are used as evaluation metrics to comprehensively assess the model's predictive ability for battery health status. A new comprehensive evaluation metric is introduced. This allows for a more comprehensive assessment of model performance.

[0186] Root mean square error (RMSE) measures the average error between model predictions and actual values. The formula is:

[0187]

[0188] in, It is the number of samples in the validation set. The model is for the first The predicted value for each sample, It is the first The actual battery health status value of each sample.

[0189] Coefficient of determination ( This is used to evaluate the goodness of fit of the model to the data, and the calculation formula is:

[0190]

[0191] in, It is the average of the actual values. The closer the value is to 1, the better the model fits the data, meaning the model can explain the changes in battery health to a greater extent.

[0192] Quantum state fidelity is used to evaluate the similarity between the quantum state predicted by the model and the actual quantum state (the quantum state encoded based on the battery's actual health state). Assume the quantum state corresponding to the battery health state predicted by the model is... The quantum state corresponding to the actual battery health state is Quantum state fidelity for:

[0193]

[0194] Introduced comprehensive evaluation indicators The root mean square error, coefficient of determination, and quantum state fidelity are considered comprehensively, and weights are assigned to each indicator. , , The calculation formula is as follows:

[0195]

[0196] in, .

[0197] In practical applications, adjustments can be made according to different needs and scenarios. , , The value of this comprehensive evaluation index. This method integrates information from three metrics—root mean square error, coefficient of determination, and quantum state fidelity—to comprehensively evaluate the model's predictive ability for battery health status from different perspectives. Monitoring is performed on the validation set. Indicators allow for a more comprehensive comparison of the performance of different models, thereby selecting the optimal model and improving the reliability and accuracy of predictions.

[0198] S5: Health Status Prediction and Assessment.

[0199] After training and optimizing the model, the trained quantum-inspired deep learning model is used to predict the health status of the drone's power battery, and the prediction results are evaluated to ensure the accuracy and reliability of the model. The specific steps are as follows:

[0200] S5.1 Health Status Prediction.

[0201] Newly collected battery data (voltage) Current ,temperature Number of charging cycles Following the previous data preprocessing and feature extraction methods, the corresponding quantum state codes are obtained. , , , And combine into a whole quantum state. .

[0202] A well-trained quantum-inspired deep learning model can be represented as a function. The processed quantum state Input the data into the model to obtain a predicted value for the battery's health status. The formula for its calculation is:

[0203]

[0204] More specifically, the computation of neurons within the model involves the evolution and processing of qubit states. Assume the model has... Layer neurons, first Output quantum state of layer neurons ( ) and input quantum state (Input layer) ) and weight Related, for a single neuron In the The calculation process of a layer can be represented as follows:

[0205]

[0206] in, It is the first The number of input channels of layer neurons, It is the probability amplitude. It is the activation function of the neuron (using the quantum rotation activation function). It is a bias term. Through layer-by-layer calculations, the quantum state of the output layer is finally obtained, and then the predicted health indicator value is obtained through specific measurement or transformation operations. Quantum rotation activation function Design based on quantum rotating gate operation. Assume the input quantum state... Quantum rotation activation function for:

[0207]

[0208] in, It is a rotation angle that is dynamically adjusted based on the input quantum state or network parameters. Compared with classical activation functions (such as ReLU), the quantum rotation activation function can better utilize the characteristics of quantum states. When processing quantum state inputs in battery data, it can adjust the probability amplitude distribution at the quantum level by rotating the quantum state, more effectively capturing complex nonlinear relationships in the data and improving the model's ability to predict battery health status.

[0209] The above calculation describes the process of transforming newly acquired battery data into a predicted health status value. First, the data is converted into a quantum state input model. Within the model, neurons perform complex calculations based on the states of qubits. Through the processing of multiple layers of neurons, data features are continuously extracted and integrated, ultimately outputting the predicted health status value. This quantum-inspired computational method fully utilizes the superposition and entanglement properties of quantum states, enabling more effective handling of complex relationships in battery data and improving prediction accuracy.

[0210] S5.2 Evaluation of prediction results.

[0211] To assess the accuracy and reliability of the prediction results, a comprehensive evaluation is conducted again, including the root mean square error (RMSE) and coefficient of determination mentioned earlier. In addition to quantum state fidelity, an evaluation index based on quantum state distance is also added.

[0212] (1) Evaluation index based on quantum state distance

[0213] Employing the complementary quantity of fidelity—quantum state distance This is used to measure the difference between the predicted quantum state and the actual quantum state. For voltage... Let the quantum state encoded by the actual voltage value be... The quantum state encoded by the predicted voltage value is Then the voltage-dependent quantum state distance for:

[0214]

[0215] Similarly, for current ,temperature Number of charging cycles Its quantum state distance , , They are respectively:

[0216]

[0217]

[0218]

[0219] The overall quantum state distance is defined by comprehensively considering the quantum state distance of four parameters. for:

[0220]

[0221] in, , , , These are weighting coefficients, satisfying... .

[0222] (2) Adjustment of comprehensive evaluation indicators

[0223] Introduce a comprehensive evaluation index The root mean square error (RMSE) and coefficient of determination ( ) Quantum state fidelity Distance to the overall quantum state A comprehensive consideration was taken into account. Given the varying importance of different indicators, weights were assigned to each indicator. , , , Then the comprehensive evaluation indicators The formula for calculation is:

[0224]

[0225] in, By adjusting these weights, the importance of different evaluation indicators can be highlighted.

[0226] The quantum state distance-based evaluation metric measures the difference between model predictions and actual values ​​at the quantum level in more detail from the perspective of quantum state differences. The overall quantum state distance comprehensively considers the quantum state differences of four parameters: voltage, current, temperature, and number of charge cycles, thus comprehensively reflecting the accuracy of the model's predictions for various battery parameters. (Comprehensive evaluation metric) This approach integrates multiple evaluation metrics and performs weighted calculations based on the importance of each metric, providing a more comprehensive and representative numerical value for model evaluation. These evaluation metrics allow for a more accurate assessment of the model's performance in predicting battery health status, providing a basis for model improvement and optimization.

[0227] S5.3 Model Update and Calibration.

[0228] As new data accumulates, the model needs to be updated and calibrated regularly to adapt to changes in battery characteristics under different usage stages and environmental conditions.

[0229] (1) Data update

[0230] Let the newly accumulated battery data be... These data are then combined with the previous training data to form a new training set. (The original training set size was) ).

[0231] (2) Model retraining

[0232] The model is retrained using a new training set. During retraining, the quantum heuristic optimization algorithm is used to adjust the model parameters, while the regularization parameters are adjusted based on the characteristics of the new data. Assume the prediction uncertainty of the new data is... Then the regularization parameter The adjustment calculation formula is:

[0233]

[0234] in, These are the regularization parameters before retraining. It is the initially set reference variance.

[0235] (3) Model calibration

[0236] The retrained model was calibrated using actual battery performance tests and fault data. Let the battery health state obtained from the actual tests be... The model predicts the health status as The model is fine-tuned based on the differences between the two. This is achieved by minimizing the mean squared error loss function that minimizes the difference between them. accomplish:

[0237]

[0238] in, This refers to the number of actual test samples. The model parameters are adjusted using the backpropagation algorithm to minimize the loss function, thus completing the model calibration.

[0239] Model updates and calibrations are crucial steps in ensuring the long-term accuracy and reliability of the model. Data updates provide the model with more information, enabling it to learn about changes in battery characteristics at different stages. During retraining, regularization parameters are adjusted based on new data to balance the model's fitting and generalization abilities. Model calibration utilizes actual test data to fine-tune the model, further improving the match between model predictions and real-world conditions, ensuring the model can consistently and accurately predict the health status of the drone's power battery.

[0240] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the health status of a drone's power battery, characterized in that, The method includes the following steps: S1, Data Acquisition and Quantum Encoding; S2, Quantum Feature Extraction; S3. Constructing a quantum-inspired deep learning model; S4. Model training and optimization; S5. Health Status Prediction and Assessment; The process of constructing a quantum-inspired deep learning model is as follows: First, a deep neural network containing qubit neurons is constructed, with the neuron state represented as a superposition of multiple ground states; then, the quantum tunneling effect is introduced to improve the weight update rule; and the neuron collaboration is enhanced through the quantum entanglement mechanism; finally, a quantum measurement-inspired attention mechanism is used to dynamically allocate feature weights. The process of constructing a deep neural network containing qubit neurons is as follows: Assume that the neuron receives voltage... Current ,temperature Number of charging cycles The encoded quantum state inputs are respectively , , , For a person with The state of a neuron with input channels and its qubit neuron. for: ； in, It is the probability amplitude, satisfying , yes The ground state of 1 qubit; The process of introducing the quantum tunneling effect to improve the weight update rule is as follows: Let the neuron With neurons The connection weights between them are The loss function is In the In the next iteration, the weight update formula is: ； in, It's the learning rate. It is a reference value for the weight. It is a parameter that controls the probability of tunneling; The process of enhancing neuronal synergy through quantum entanglement is as follows: Let the neuron... and neurons Through weight Connected, neurons The output is Neuron The input is After considering quantum entanglement, neurons The update calculation formula is: ； in, It is an activation function. Neuron and The entanglement coefficient between them; The process of dynamically allocating feature weights using a quantum measurement-inspired attention mechanism is as follows: Let the feature vector input to the model be... ,in Corresponding voltages Current ,temperature Number of charging cycles Other features; feature vectors processed by the attention mechanism for: ； in, It is the attention weight, inspired by the quantum measurement process, and its calculation formula is: ； in, and Is with characteristics and The corresponding quantum state, It is a temperature parameter used to control the smoothness of weight distribution.

2. The method for predicting the health status of a UAV power battery according to claim 1, characterized in that, The data acquisition and quantum encoding process is as follows: First, the battery voltage is collected in real time by the sensors of the drone's battery management system. Current ,temperature Number of charging cycles Then, each parameter is quantized and encoded, converting it into its corresponding quantum state. , , , ; The four quantum states are combined into a single quantum state. ;in, This represents the tensor product operation.

3. The method for predicting the health status of a UAV power battery according to claim 2, characterized in that, The formula for calculating quantized encoding is as follows: ； in, , Indicates voltage, Indicates current, Indicates temperature, Indicates the number of charging cycles; and This indicates the lower and upper limits of the value range for each parameter.

4. The method for predicting the health status of a UAV power battery according to claim 1, characterized in that, The process of quantum feature extraction is as follows: First, the relative rate of change of quantum states of each parameter is calculated to characterize the quantum state difference between adjacent time steps; then, the quantum entanglement degree of each pair of parameters is calculated to extract the multi-parameter quantum entanglement correlation characteristics; finally, the quantum state probability distribution entropy is calculated to quantify the uncertainty of the parameters.

5. The method for predicting the health status of a UAV power battery according to claim 1, characterized in that, The model training and optimization process involves: using quantum rotation gates and mutation operations to adjust the qubit states and optimize the model parameters; and dynamically adjusting the regularization parameters. ; Using root mean square error (RMSE) and coefficient of determination and quantum state fidelity Comprehensive evaluation of model performance.

6. The method for predicting the health status of a UAV power battery according to claim 5, characterized in that, The process of optimizing model parameters by adjusting the state of qubits using quantum rotation gates and mutation operations is as follows: For the Let a given individual in the population have a parameter represented by one of its qubits. The corresponding quantum state is According to the update calculation formula of the quantum rotating gate: = ； in, It is the rotation angle; To increase population diversity, a quantum mutation operation is introduced: with a certain mutation probability... Mutate a qubit; for a qubit If a mutation occurs, an angle will be randomly selected. And perform the following operations: = ； Dynamically adjust regularization parameters The formula for calculation is: ; in, and It is the initial value. This represents the variance of the predicted results.

7. The method for predicting the health status of a UAV power battery according to claim 1, characterized in that, The process of predicting and assessing the health status is as follows: First, the newly acquired battery data is encoded into quantum states. And input it into the trained model to obtain the predicted value. ; Introduce another comprehensive evaluation indicator The root mean square error (RMSE) and coefficient of determination are... Quantum state fidelity Distance to the overall quantum state A comprehensive assessment will be conducted.

8. The method for predicting the health status of a UAV power battery according to claim 7, characterized in that, The predicted value The calculation formula is: ; The overall quantum state distance The calculation formula is: ； in, , , , These are weighting coefficients, satisfying... ; The comprehensive evaluation indicators The calculation formula is: ； in, .

Images