CN-120296396-B - Unmanned aerial vehicle power battery health state prediction method

CN120296396BCN 120296396 BCN120296396 BCN 120296396BCN-120296396-B

Abstract

The invention relates to the technical field of battery state prediction and discloses a method for predicting the health state of an unmanned aerial vehicle power battery. The method breaks through the limitation of the traditional method in complex nonlinear relation modeling and local optimization through the quantized coding, the quantum heuristic model architecture and the dynamic optimization mechanism, provides a high-precision and high-robustness solution for the health management of the unmanned aerial vehicle power battery, and has important engineering application value.

Inventors

LI JINBO
JIAO JUNYAN
YANG CHENG
Xu Jiugun
WANG QIN

Assignees

南京航空航天大学

Dates

Publication Date: 20260505
Application Date: 20250415

Claims (8)

1. A method for predicting the state of health of an unmanned aerial vehicle power battery, the method comprising the steps of: S1, data acquisition and quantization coding; S2, quantum feature extraction; S3, constructing a quantum heuristic deep learning model; S4, model training and optimization; S5, predicting and evaluating the health state; The process for constructing the quantum heuristic deep learning model comprises the following steps: Firstly, constructing a deep neural network containing quantum bit neurons, wherein the states of the neurons are expressed as superposition of a plurality of ground states, then introducing quantum tunneling effect to improve a weight updating rule, and enhancing the coordination of the neurons through a quantum entanglement mechanism; The process of constructing deep neural network containing qubit neurons is to let the neurons receive the voltage from Current flow Temperature (temperature) Number of times of charging The encoded quantum state inputs are respectively 、、、 For a device with Neurons of the input channels, their qubit neuron states The method comprises the following steps: ; Wherein, the Is a probability amplitude satisfying , Is that The ground state of the individual qubits; The process of introducing quantum tunneling effect improved weight updating rule includes setting neuron And neurons The connection weight between the two is The loss function is In the first place And in the next iteration, the weight updating calculation formula is as follows: ; Wherein, the Is the rate of learning to be performed, Is a reference value for the weight and, Is a parameter controlling the tunneling probability; The process of enhancing the coordination of neurons by quantum entanglement mechanism comprises setting neurons And neurons By weight Connected, neurons The output of (2) is Neurons The input of (2) is After considering quantum entanglement, neurons The updated calculation formula of (c) is: ; Wherein, the Is the function of the activation and, Is a neuron And Entanglement coefficients therebetween; The dynamic feature weight distribution process by adopting the attention mechanism inspired by quantum measurement comprises the steps of setting the feature vector input by a model as Wherein Respectively corresponding to voltages Current flow Temperature (temperature) Number of times of charging And other features, feature vectors processed by the attention mechanism The method comprises the following steps: ; Wherein, the Attention weight is inspired by the quantum measurement process, and the calculation formula is as follows: ; Wherein, the And Is and features And The corresponding quantum state of the light source is obtained, Is a temperature parameter for controlling the smoothness of the weight distribution.
2. The method for predicting the health state of a power battery of an unmanned aerial vehicle according to claim 1, wherein the data acquisition and quantization encoding process comprises the following steps: firstly, collecting the voltage of a battery in real time through a sensor of a unmanned aerial vehicle battery management system Current flow Temperature (temperature) Number of times of charging Respectively performing quantization coding on each parameter, and converting into corresponding quantum state 、、、 Combining the four quantum states into an integral quantum state Wherein, the method comprises the steps of, Representing tensor product operations.
3. The unmanned aerial vehicle power battery state of health prediction method of claim 2, wherein the quantized code is calculated as follows: ; Wherein, the , Representing the voltage, Representing the current, Indicating the temperature, Indicating the number of charging times; And The lower limit and the upper limit of the value range of each parameter are indicated.
4. The unmanned aerial vehicle power battery state of health prediction method of claim 1, wherein the quantum feature extraction process is: the method comprises the steps of calculating the quantum state relative change rate of each parameter, representing the quantum state difference of adjacent time steps, extracting multi-parameter quantum entanglement association characteristics by calculating the quantum entanglement degree of every two parameters, and calculating the quantum state probability distribution entropy for quantifying the uncertainty of the parameters.
5. The unmanned aerial vehicle power battery health state prediction method of claim 1, wherein the model training and optimization process comprises adopting quantum rotation gate and mutation operation, adjusting qubit state to optimize model parameters, dynamically adjusting regularization parameters Determining coefficients using root mean square error RMSE And quantum state fidelity And comprehensively evaluating the performance of the model.
6. The method for predicting the health of an unmanned aerial vehicle power battery according to claim 5, wherein the process of optimizing the model parameters by adjusting the qubit state using the quantum rotation gate and the mutation operation is as follows: For the first A certain individual in the generation group is set as a parameter expressed by one qubit The corresponding quantum state is According to the updated calculation formula of the quantum rotation gate: = ; Wherein, the Is the rotation angle; To increase the diversity of the population, a quantum mutation operation is introduced, wherein a certain mutation probability is adopted Mutating a qubit for a qubit If variation occurs, an angle is randomly selected And performs the following operations: = ; Dynamically adjusting regularization parameters The calculation formula of (2) is as follows: ; Wherein, the And Is an initial value of the value, Is the prediction result variance.
7. The method for predicting health of a battery of an unmanned aerial vehicle according to claim 1, wherein the predicting and evaluating of health comprises encoding newly collected battery data into a quantum state And inputting the trained model to obtain predicted value Introducing a comprehensive evaluation index Determining the root mean square error RMSE and coefficient Quantum state fidelity And overall quantum state distance And (5) performing comprehensive evaluation.
8. The method for predicting the health of a power battery of an unmanned aerial vehicle according to claim 7, wherein the predicted value is The calculation formula of (2) is as follows: ; The overall quantum state distance The calculation formula of (2) is as follows: ; Wherein, the 、、、 Is a weight coefficient, satisfies ; The comprehensive evaluation index The calculation formula of (2) is as follows: ; Wherein, the 。

Description

Unmanned aerial vehicle power battery health state prediction method Technical Field The invention relates to the technical field of battery state prediction, in particular to a method for predicting the health state of a power battery of an unmanned aerial vehicle. Background Battery state of health prediction methods are typically based on traditional machine learning algorithms (e.g., support vector machine, random forest) or statistical models (e.g., kalman filtering). However, these methods have significant limitations in dealing with complex nonlinear relationships between battery parameters (voltage, current, temperature, number of charges, etc.), insufficient data representation capabilities, low model optimization efficiency, and experience-dependent feature engineering. In view of the above, complex nonlinearity problems are mostly addressed by quantum computing techniques. The relevance of multidimensional data can be efficiently represented based on superposition and entanglement of quantum states, and meanwhile, a quantum heuristic algorithm (such as a quantum genetic algorithm) shows remarkable advantages in global optimization and jump-out of a local optimal solution. However, the prior art has not effectively combined quantum computing with deep learning framework to solve the battery health prediction problem. Therefore, the invention provides a method for predicting the health state of the power battery of the unmanned aerial vehicle, which improves the technical problems. Disclosure of Invention Aiming at the defects of the prior art, the embodiment of the invention provides a method for predicting the health state of an unmanned aerial vehicle power battery. The technical aim of the invention is realized by the following technical scheme that the method for predicting the health state of the unmanned aerial vehicle power battery comprises the following steps: S1, data acquisition and quantization coding; S2, quantum feature extraction; S3, constructing a quantum heuristic deep learning model; S4, model training and optimization; s5, health state prediction and evaluation. As a preferable technical scheme of the invention, the data acquisition and quantization coding process comprises the following steps: firstly, collecting the voltage of a battery in real time through a sensor of a unmanned aerial vehicle battery management system Current flowTemperature (temperature)Number of times of chargingRespectively performing quantization coding on each parameter, and converting into corresponding quantum state、、、Combining the four quantum states into an integral quantum stateWherein, the method comprises the steps of,Representing tensor product operations. As a preferred embodiment of the present invention, the quantization coding is calculated as follows: Wherein, the ,Representing the voltage,Representing the current,Indicating the temperature,Indicating the number of charging times; And The lower limit and the upper limit of the value range of each parameter are indicated. As a preferable technical scheme of the invention, the quantum characteristic extraction process comprises the following steps: the method comprises the steps of calculating the quantum state relative change rate of each parameter, representing the quantum state difference of adjacent time steps, extracting multi-parameter quantum entanglement association characteristics by calculating the quantum entanglement degree of every two parameters, and calculating the quantum state probability distribution entropy for quantifying the uncertainty of the parameters. As a preferable technical scheme of the invention, the process for constructing the quantum heuristic deep learning model is as follows: The method comprises the steps of constructing a deep neural network containing quantum bit neurons, expressing the states of the neurons as superposition of a plurality of ground states, introducing quantum tunneling effect to improve a weight updating rule, enhancing the coordination of the neurons through a quantum entanglement mechanism, and finally dynamically distributing characteristic weights by adopting a attention mechanism inspired by quantum measurement. As a preferred technical scheme of the invention, the process of constructing the deep neural network comprising the quantum bit neurons comprises the steps of providing the neurons to receive the voltage fromCurrent flowTemperature (temperature)Number of times of chargingThe encoded quantum state inputs are respectively、、、For a device withNeurons of the input channels, their qubit neuron statesThe method comprises the following steps: Wherein, the Is a probability amplitude satisfying,Is thatThe ground state of the individual qubits; The process of introducing quantum tunneling effect improved weight updating rule includes setting neuron And neuronsThe connection weight between the two isThe loss function isIn the first placeAnd in the next iteration, the weight updating