Aviation power battery digital twinborn fusion deep learning health prediction method

Abstract

The invention discloses an aviation power battery digital twinning fusion deep learning health prediction method comprising the following steps: constructing a digital twinning virtual battery model, collecting and processing battery data, and obtaining a new real-time data matrix; constructing a multi-task deep learning model of joint training, and inputting the new real-time data matrix into the multi-task deep learning model of joint training for processing to obtain an SOH value predicted by the multi-task deep learning model of joint training; and updating the battery data based on the SOH value predicted by the joint training multi-task deep learning model to obtain a new data scalar, inputting the new data scalar into the digital twinning virtual battery model, and updating the digital twinning virtual battery model. According to the invention, through deep combination of the digital twinning virtual battery model and the multi-task deep learning model of joint training, an efficient and reliable solution is provided for health management and life prediction of the aviation power battery.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health management, and particularly to a health prediction method for aviation power batteries integrating digital twin and deep learning. Background Art

[0002] Aviation power batteries are a key energy storage device in the modern aviation field and are widely used in equipment such as unmanned aerial vehicles and electric aircraft. Their performance directly affects the endurance, safety, and operating costs of aircraft. However, when used under high power, high load, and complex environmental conditions, aviation power batteries face significant challenges in health management and reliability, including problems such as capacity decay, shortened lifespan, and difficulty in fault prediction. And battery health management methods mainly focus on battery state monitoring (such as SOH) and RUL prediction. Therefore, accurately estimating the SOH and RUL of aviation power batteries can not only extend the battery's service life but also optimize the maintenance plan of the aircraft, improving the safety and economy of system operation.

[0003] Traditional battery modeling and health prediction methods can be divided into two categories: one is the physical model method based on electrochemical mechanisms (such as equivalent circuit models, pseudo-two-dimensional models), which realizes dynamic characterization by establishing mathematical equations for internal battery reactions. Its advantage lies in strong model interpretability and clear physical meaning; however, it relies on accurate physical parameter identification and complex electrochemical equation solving, with technical defects such as low computational efficiency and poor real-time performance. The other is the machine learning method based on big data analysis (such as neural networks, support vector machines), which realizes health state prediction by mining the statistical laws of battery operation data. Its advantage lies in the absence of explicit mechanism modeling and adaptability to complex nonlinear relationships; however, it has limitations such as the dependence of model generalization ability on the scale and quality of training data and insufficient interpretability of prediction results, making it difficult to guide the in-depth optimization of battery aging mechanisms.

[0004] The existing Chinese patent CN119438958A introduces a method for predicting the health state of lithium batteries. This method first establishes a lithium battery health state prediction model composed of an encoder of the Transformer layer, a multi-layer KAN network as the decoder, and an unscented Kalman filter connected. Then, the lithium battery operation data is input into the prediction model, and finally, the prediction result is output. Another Chinese patent CN119310487A proposes a method for evaluating the health state of battery energy storage. It collects the constant power charge and discharge process data of lithium batteries and obtains health factors highly correlated with the health state of battery energy storage through Pearson correlation analysis. Then, using an improved Kepler optimization algorithm and introducing a differential evolution strategy, the bidirectional long short-term memory network model is optimized. Finally, by inputting the health factors into the bidirectional long short-term memory network model for processing, the estimated value of the health state of the energy storage battery is output.

[0005] In the above patents, the battery health management method usually uses a single model for modeling. However, it is difficult for a single model to deeply mine the collected data and update the model parameters in real time. Its adaptability and efficiency are limited under complex working conditions, and traditional data-driven methods often struggle to fully utilize the long-term dependence characteristics in time series data. Summary of the Invention

[0006] To overcome the above defects of the prior art, an embodiment of the present invention provides a health prediction method for aviation power batteries that integrates digital twin and deep learning, which solves the problems mentioned in the background art.

[0007] To achieve the above object, the present invention provides the following technical solution: A health prediction method for aviation power batteries that integrates digital twin and deep learning, comprising the following steps: Step S1: Construct a digital twin virtual battery model; Step S2: Collect battery data, process the battery data to obtain a new real-time data matrix; Step S3: Construct a multi-task deep learning model for joint training; The multi-task deep learning model for joint training consists of a convolutional neural network layer, a multi-scale hierarchical bidirectional long short-term memory network layer, a multi-layer perceptron layer, and a task branch layer; Input the new real-time data matrix into the convolutional neural network layer for processing to obtain the output of the pooling layer; Step S4: Input the output of the pooling layer into the multi-scale hierarchical bidirectional long short-term memory network layer for processing to obtain an enhanced context vector; Step S5: Input the enhanced context vector into the multi-layer perceptron layer for processing to obtain the output of the second fully connected layer; Step S6: Input the output of the second fully connected layer into the task branch layer for processing to obtain the value predicted by the multi-task deep learning model for joint training; Step S7: And update the battery data based on the value predicted by the multi-task deep learning model for joint training to obtain a new data scalar, and input the new data scalar into the digital twin virtual battery model to update the digital twin virtual battery model.

[0008] Further, in step S1, when constructing the digital twin virtual battery model, the specific process is as follows: Construct a dual RC equivalent circuit model as the digital twin virtual battery model. The digital twin virtual battery model consists of an ohmic resistance , two parallel loop RCs, and an open-circuit voltage ; The two parallel loop RCs include a first parallel loop RC 1 and a second parallel loop RC2 ; The first parallel RC circuit 1 includes a first resistor , a first capacitor ; The second parallel RC circuit 2 includes a second resistor and a second capacitor ; The output of the digital twin virtual battery model is the terminal voltage of the battery , the dynamic behavior, which is expressed as: ; In the formula, represents the terminal voltage of the battery at time step t; represents the battery current at time step t; and respectively represent the polarization voltages of the first parallel RC 1 and the second parallel RC 2 circuits at time step t; and The dynamic behavior is expressed as: ; ; ; ; In the formula, represents the time constant of the first parallel RC 1 circuit, indicating the fast polarization dynamics; represents the time constant of the second parallel RC 2 circuit, indicating the slow polarization dynamics; represents the base of the natural logarithm; represents the time variable.

[0009] Furthermore, the specific process of the new real-time data matrix in step S2 is as follows: The battery data is collected in real time through sensors and processed to construct a real-time data matrix , which is expressed as: ; In the formula, represents the battery current at time step ; represents the battery current at time step ; represents the time window length; represents the temperature at time step ; represents the time step The temperature at a moment; Indicates that the time step is The temperature at a moment; Indicates the battery capacity at time step t; Indicates that the time step is The battery capacity at a moment; Indicates that the time step is The battery capacity at a moment; Record the dynamic parameter S in the scalar of the real-time collected battery data, which means: ; In the formula, Indicates the depth of discharge; Indicates the initial battery capacity; Input the real-time data matrix and the dynamic parameter S in the scalar of the battery data into the virtual battery model of the digital twin for real-time state update, and obtain the new real-time data matrix , which means: ; In the formula, Indicates the terminal voltage of the battery at time step t; Indicates that the time step is The terminal voltage of the battery at a moment; Indicates that the time step is The terminal voltage of the battery at a moment; Indicates the new real-time data matrix The dimension of, A real number matrix of; m represents the feature dimension.

[0010] Furthermore, the output of the pooling layer in step S3, the specific process is: Input the new real-time data matrix into the input layer of the convolutional neural network layer to obtain the output of the input layer; Transmit the output of the input layer to the convolutional layer of the convolutional neural network layer for convolutional kernel and dynamic gated piecewise activation function operations to obtain the result of the convolutional kernel operation, process the result of the convolutional kernel operation to obtain the output result of the dynamic gated piecewise activation function, and process the result of the convolutional kernel operation and the output result of the dynamic gated piecewise activation function to obtain the output dimension of The output of the convolutional layer , Indicates the number of convolutional kernels, which means: ; In the formula, Indicates the Output of the th convolutional layer; Indicates the convolutional kernel size; Transfer the output of the convolutional layer to the pooling layer of the convolutional neural network layer for pooling operation, and obtain an output dimension of the output of the pooling layer , which means: ; ; In the formula, represents the max pooling function; represents the number of time points after pooling operation; represents the stride; represents the pooling layer window size.

[0011] Furthermore, the output result of the dynamic gating piecewise activation function, the specific process is as follows: Calculate the result of the convolutional kernel operation to obtain the mean value and the standard deviation ; Based on the mean value and the standard deviation generate a dynamic threshold, which means: ; ; In the formula, represents the first dynamic threshold; represents the learnable parameter of the first dynamic threshold; represents the second dynamic threshold; represents the learnable parameter of the second dynamic threshold; Based on the mean value and the standard deviation generate a gating parameter, which means: ; In the formula, ; represents the th piecewise slope; ; represents the th intercept; represents the lightweight fully connected network; Divide the interval of the result of the convolutional kernel operation by the threshold, and achieve smooth transition through the activation function to obtain the output result of the dynamic gating piecewise activation function, which means: ; ; In the formula, represents the output result of the dynamic gating piecewise activation function, according to the input Interval division is achieved through interpolation to smoothly transition between different linear segments; Represents the Sigmoid activation function; Represents the result after the convolution operation; Represents the exponential part.

[0012] Furthermore, the enhanced context vector in step S4 has the following specific process: Transmit the output of the pooling layer to the multi-scale hierarchical bidirectional long short-term memory network layer for processing to obtain the enhanced context vector; the multi-scale hierarchical bidirectional long short-term memory network layer includes a short-term layer and a long-term layer; Transmit the output of the pooling layer to the short-term layer for forward and backward propagation to obtain the short-term hidden state at time step t ; Input the short-term hidden state at time step t to the long-term layer through a skip connection and calculate to obtain the long-term hidden state at time step ; Concatenate the long-term hidden states at time step to obtain a bidirectional hidden state matrix with an output dimension of ; Represents the number of hidden layer units in a single direction; Calculate the attention score at time step t and the feature attention score with a feature dimension of for the bidirectional hidden state matrix , and then combine the spatio-temporal attention weights to generate the enhanced context vector , expressed as: ; ; ; In the formula, represents the activation function; represents the time attention weight matrix; represents the time attention bias term; represents the feature dimension attention weight matrix; represents the feature dimension attention bias term; represents the hidden states of all time steps in the long-term layer on the feature dimension ; represents the hidden state of the long-term layer at time step t on the feature dimension Hidden state value; Indicates the total number of time steps.

[0013] Furthermore, the output of the second fully connected layer in step S5, the specific process is as follows: The enhanced context vector and the dynamic parameter S in the scalar of battery data are concatenated, and the concatenated column vector with a dimension of is obtained; ; Indicates the number of dynamic parameters, expressed as: ; The concatenated column vector is input into the multi-layer perceptron layer for processing to obtain the output of the second fully connected layer ; The multi-layer perceptron layer includes a first fully connected layer and a second fully connected layer; The first fully connected layer includes the weight matrix of the first fully connected layer and the bias term of the first fully connected layer ; The second fully connected layer includes the weight matrix of the second fully connected layer and the bias term of the second fully connected layer ; The concatenated column vector is multiplied by the weight matrix of the first fully connected layer to obtain the result of the first matrix multiplication operation. The result of the first matrix multiplication operation is added to the bias term of the first fully connected layer and undergoes a non-linear transformation through the activation function ReLU to obtain the output of the first fully connected layer , expressed as: ; The output of the first fully connected layer is multiplied by the weight matrix of the second fully connected layer to obtain the result of the second matrix multiplication operation. The output of the first fully connected layer , the result of the second matrix multiplication operation and the bias term of the second fully connected layer are added together and undergo a non-linear transformation through the activation function ReLU to obtain the output of the second fully connected layer , expressed as: .

[0014] Furthermore, the value predicted by the multi-task deep learning model in step S6 during joint training, the specific process is as follows: The output of the second fully connected layer The input task branch layer is processed to obtain the value predicted by the multi-task deep learning model for joint training value; The task branch layer includes an SOH prediction branch; the SOH prediction branch includes the weight matrix of the first branch and the bias term of the first branch ; Multiply the output of the second fully connected layer by the weight matrix of the first branch to obtain the result of the first matrix multiplication operation. Add the result of the first matrix multiplication operation to the bias term of the first branch , and finally use the activation function for non-linear transformation to obtain the value predicted by the multi-task deep learning model for joint training value , which means: .

[0015] Furthermore, in step S7, the digital twin virtual battery model is updated. The specific process is as follows: According to the value predicted by the multi-task deep learning model for joint training value dynamically update the dynamic parameter S in the scalar of the battery data to obtain a new data scalar , where is the resistance value of the updated first resistor ; is the capacitance value of the updated first capacitor ; is the resistance value of the updated second resistor ; is the capacitance value of the updated second capacitor ; Input the new data scalar into the digital twin virtual battery model to update the digital twin virtual battery model, which means: ; ; ; ; In the formula, represents the sensitivity coefficient of the value predicted by the multi-task deep learning model for joint training to the degradation of the first resistor ; represents the sensitivity coefficient of the value predicted by the multi-task deep learning model for joint training to the degradation of the first capacitor ; The values predicted by the jointly trained multi-task deep learning model for the second resistor degradation sensitivity coefficient; The values predicted by the jointly trained multi-task deep learning model for the second capacitor degradation sensitivity coefficient.

[0016] Compared with the existing technologies, the present invention has the following beneficial effects: (1) The present invention constructs a digital twin virtual battery model and uses sensors to collect real-time battery data; digital twin can integrate the digital twin virtual battery model and real-time battery data, providing accurate support for the whole life cycle management of battery health status.

[0017] (2) Through the strong time series feature extraction ability of the multi-scale hierarchical bidirectional long short-term memory network layer of the present invention, the long-term dependence relationship and dynamic change trend in the new real-time data matrix can be effectively captured; at the same time, combined with the advantage of the convolutional neural network layer in extracting local features, the detailed information and local features in the multi-dimensional data of the battery can be fully mined, and then the collaborative modeling of time series global features and local features can be realized, so as to improve the expression ability and prediction accuracy of the jointly trained multi-task deep learning model for complex data.

[0018] (3) By adding a multi-layer perceptron layer after the convolutional neural network layer and the multi-scale hierarchical bidirectional long short-term memory network layer, the multi-layer perceptron layer has strong multi-modal feature fusion ability, which can map time series features and static features into a higher-dimensional or lower-dimensional space, further optimizing the feature representation and improving the prediction of the SOH branch of the digital twin virtual battery model; the values predicted by the jointly trained multi-task deep learning model dynamically update the dynamic parameters of the digital twin virtual battery model, which can effectively improve the modeling accuracy.

[0019] (4) Through the deep combination of the digital twin virtual battery model and the jointly trained multi-task deep learning model of the present invention, it can not only capture and reflect the health status of the aviation power battery in real time, but also optimize the input data quality through digital twin under complex working conditions, further improving the prediction accuracy, robustness and adaptability to complex working conditions of the digital twin virtual battery model, providing an efficient and reliable solution for the health management and life prediction of the aviation power battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] Such as Figure 1As shown, the present invention provides a technical solution: a health prediction method for aviation power battery digital twin fusion deep learning, comprising the following steps: Step S1: constructing a virtual battery model of a digital twin; Step S2: Collect battery data, process the battery data, and obtain a new real-time data matrix; Step S3: construct a multi-task deep learning model for joint training; the multi-task deep learning model for joint training consists of a convolutional neural network layer, a multi-scale hierarchical bidirectional long short-term memory network layer, a multi-layer perceptron layer, and a task branching layer; input the new real-time data matrix into the convolutional neural network layer for processing to obtain the output of the pooling layer; Step S4: Input the output of the pooling layer into the multi-scale hierarchical bidirectional long short-term memory network layer for processing to obtain an enhanced context vector; Step S5: input the enhanced context vector into the multi-layer perceptron layer for processing to obtain the output of the second fully connected layer; Step S6: The output of the second fully connected layer is input into the task branch layer for processing to obtain the predicted value of the jointly trained multi-task deep learning model. value; Step S7: and predict based on the jointly trained multi-task deep learning model The battery data is updated with the value to obtain a new data scalar, the new data scalar is input into the virtual battery model of the digital twin, and the virtual battery model of the digital twin is updated.

[0022] Among them, in step S1, a virtual battery model of a digital twin is constructed, and the specific process is as follows: Construct a dual RC equivalent circuit model as the virtual battery model of the digital twin. The virtual battery model of the digital twin consists of an ohmic resistor , two parallel circuits RC and open circuit voltage Composition: two parallel circuits RC including a first parallel circuit RC 1 And the second parallel circuit RC 2 ; The first parallel circuit RC 1 Including the first resistor , the first capacitor ; The second parallel circuit RC 2 Including the second resistor and the second capacitor ; Used to reflect the slow dynamic process of the battery; The output of the digital twin virtual battery model is the battery terminal voltage , dynamic behavior, indicating: ; In the formula, denotes the terminal voltage of the battery at time step \(t\); denotes the battery current at time step \(t\); and respectively denote the polarization voltages of the first parallel RC 1 and the second parallel RC 2 at time step \(t\); and represent the dynamic behaviors, indicating: ; ; ; ; In the formula, denotes the time constant of the first parallel RC 1 circuit, representing the fast polarization dynamics; denotes the time constant of the second parallel RC 2 circuit, representing the slow polarization dynamics; denotes the base of the natural logarithm; denotes the time variable.

[0023] Among them, the new real-time data matrix in step S2, the specific process is: Collect battery data in real time through the sensors in the aviation battery management system for processing, and construct a real-time data matrix , indicating: ; In the formula, denotes the battery current at time step ; denotes the battery current at time step ; \(T\) denotes the time window length; denotes the temperature at time step \(t\); denotes the temperature at time step ; denotes the temperature at time step ; denotes the battery capacity at time step \(t\); denotes the battery capacity at time step ; denotes the battery capacity at time step ; Records the dynamic parameter \(S\) in the scalar of the real-time collected battery data, indicating: ; In the formula, Indicates the depth of discharge; Indicates the initial battery capacity; Input the real-time data matrix and the dynamic parameter S in the scalar of battery data into the virtual battery model of the digital twin for real-time state update to obtain a new real-time data matrix , which means: ; In the formula, Indicates the terminal voltage of the battery at time step t; Indicates the time step is The terminal voltage of the battery at the moment; Indicates the time step is The terminal voltage of the battery at the moment; Indicates the new real-time data matrix The dimension of, A real number matrix of; m represents the feature dimension.

[0024] Among them, the output of the pooling layer in step S3, the specific process is: Input the new real-time data matrix Into the input layer of the convolutional neural network layer to obtain the output of the input layer; Transmit the output of the input layer to the convolutional layer of the convolutional neural network layer for convolutional kernel and dynamic gated piecewise activation function (DGPA) operations to obtain the result of the convolutional kernel operation, process the result of the convolutional kernel operation to obtain the output result of the dynamic gated piecewise activation function, and process the result of the convolutional kernel operation and the output result of the dynamic gated piecewise activation function to obtain an output dimension of The output of the convolutional layer , Indicates the number of convolutional kernels, which means: ; In the formula, Indicates the Output of the th convolutional layer; Indicates the convolutional kernel size; Transmit the output of the convolutional layer To the pooling layer of the convolutional neural network layer for pooling operation, and the pooling operation selects the maximum pooling function to obtain an output dimension of The output of the pooling layer , which means: ; ; In the formula, MaxPooling1D represents the maximum pooling function; Indicates the number of time points after the pooling operation; Indicates the step size; Indicates the window size of the pooling layer.

[0025] Among them, the output result of the dynamic gating piecewise activation function, the specific process is as follows: Calculate the result of the convolutional kernel operation to obtain the mean value and the standard deviation ; Based on the mean value and the standard deviation Generate a dynamic threshold, expressed as: ; ; In the formula, Indicates the first dynamic threshold; Indicates the learnable parameter of the first dynamic threshold; Indicates the second dynamic threshold; Indicates the learnable parameter of the second dynamic threshold; Based on the mean value and the standard deviation Generate a gating parameter, expressed as: ; In the formula, ; Indicates the th piecewise slope; ; Indicates the th intercept; Indicates the lightweight fully connected network; Divide the interval of the result of the convolutional kernel operation by the threshold, and achieve smooth transition through the activation function to obtain the output result of the dynamic gating piecewise activation function, expressed as: ; ; In the formula, Indicates the output result of the dynamic gating piecewise activation function, and through interpolation, smooth transition of different linear segments is achieved according to the interval division of the input ; Indicates the Sigmoid activation function; Indicates the result after the convolutional operation; Indicates the exponential part.

[0026] Among them, the enhanced context vector in step S4, the specific process is as follows: The multi-scale hierarchical bidirectional long short-term memory network layer BiLSTM includes a short-term layer and a long-term layer; the time step of the short-term layer t = 10; the time step of the long-term layer = 50; Transfer the output of the pooling layer to the short-term layer for forward and backward propagation to obtain the short-term hidden state at time step t ; Input the short-term hidden state at time step t into the long-term layer through skip connection and calculate the long-term hidden state at the time step ; Concatenate the long-term hidden states at the time step to obtain a bidirectional hidden state matrix with an output dimension of , where represents the number of hidden layer units in a single direction; Calculate the attention scores at time step for the bidirectional hidden state matrix and the feature attention scores for the feature dimension . Then, combine with the spatio-temporal attention weights to generate an enhanced context vector , expressed as: ; ; ; ; ; In the formula, represents the activation function; represents the time attention weight matrix; represents the time attention bias term; represents the feature dimension attention weight matrix; represents the feature dimension attention bias term; represents the hidden states of all time steps in the long-term layer on the feature dimension ; represents the hidden state value of the long-term layer at time step t on the feature dimension ; represents the total number of time steps.

[0027] Among them, the output of the second fully connected layer in step S5 is as follows: Concatenate the enhanced context vector with the dynamic parameter S in the scalar of the battery data to obtain a concatenated column vector with a dimension of ; represents the number of dynamic parameters, expressed as: ; ​​Concatenate column vectors Input it into a multi-layer perceptron layer for processing to obtain the output of the second fully connected layer ; The multi-layer perceptron layer includes a first fully connected layer and a second fully connected layer; The first fully connected layer includes the weight matrix of the first fully connected layer and the bias term of the first fully connected layer ; The second fully connected layer includes the weight matrix of the second fully connected layer and the bias term of the second fully connected layer ; Concatenate column vectors Perform matrix multiplication with the weight matrix of the first fully connected layer to obtain the result of the first matrix multiplication operation. Add the result of the first matrix multiplication operation to the bias term of the first fully connected layer and perform a non-linear transformation through the activation function ReLU to obtain the output of the first fully connected layer , which is expressed as: ; Perform matrix multiplication on the output of the first fully connected layer with the weight matrix of the second fully connected layer to obtain the result of the second matrix multiplication operation. Add the output of the first fully connected layer , the result of the second matrix multiplication operation and the bias term of the second fully connected layer and perform a non-linear transformation through the activation function ReLU to obtain the output of the second fully connected layer , which is expressed as: .

[0028] Among them, the value predicted by the multi-task deep learning model in step S6 is specifically as follows: Input the output of the second fully connected layer into the task branch layer for processing to obtain the value predicted by the multi-task deep learning model in joint training; The task branch layer includes a SOH prediction branch (state of health) and an RUL prediction branch (remaining useful life); the SOH prediction branch includes the weight matrix of the first branch and the bias term of the first branch ; Perform matrix multiplication on the output of the second fully connected layer with the weight matrix of the first branch Perform matrix multiplication to obtain the result of the first matrix multiplication, and add the bias term of the first branch to the result of the first matrix multiplication , and finally perform a non-linear transformation using an activation function to obtain the value predicted by the jointly trained multi-task deep learning model value , indicating that: .

[0029] Among them, the RUL prediction branch includes the weight matrix of the second branch and the bias term of the second branch ; Multiply the output of the second fully connected layer by the weight matrix of the second branch to perform matrix multiplication to obtain the result of the second matrix multiplication, and add the bias term of the second branch to the result of the second matrix multiplication , and finally perform a non-linear transformation using an activation function to obtain the RUL value predicted by the jointly trained multi-task deep learning model , indicating that: .

[0030] Among them, in step S7, the digital twin virtual battery model is updated, and the specific process is as follows: According to the value predicted by the jointly trained multi-task deep learning model value dynamically update the dynamic parameter S in the scalar of the battery data to obtain a new data scalar , where is the resistance value of the updated first resistor , is the capacitance value of the updated first capacitor ; is the resistance value of the updated second resistor ; is the capacitance value of the updated second capacitor ; Input the new data scalar into the digital twin virtual battery model to update the digital twin virtual battery model, indicating; ; ; ; ; In the formula, represents the value predicted by the jointly trained multi-task deep learning model value for the first resistor Degradation sensitivity coefficient; Indicating the Value pair for the first capacitor Degradation sensitivity coefficient; Indicating the Value pair for the second resistor Degradation sensitivity coefficient; Indicating the Value pair for the second capacitor Degradation sensitivity coefficient.

[0031] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A health prediction method for aviation power batteries that integrates digital twins and deep learning, characterized in that, It includes the following steps: Step S1: Construct a digital twin virtual battery model; Step S2: Collect battery data, process the battery data to obtain a new real-time data matrix; Step S3: Construct a jointly trained multi-task deep learning model; the jointly trained multi-task deep learning model consists of a convolutional neural network layer, a multi-scale hierarchical bidirectional long short-term memory network layer, a multi-layer perceptron layer, and a task branch layer; input the new real-time data matrix into the convolutional neural network layer for processing to obtain the output of the pooling layer; Step S4: Input the output of the pooling layer into the multi-scale hierarchical bidirectional long short-term memory network layer for processing to obtain an enhanced context vector; Step S5: Input the enhanced context vector into the multi-layer perceptron layer for processing to obtain the output of the second fully connected layer; Step S6: Input the output of the second fully connected layer into the task branch layer for processing to obtain the value predicted by the multi-task deep learning model for joint training; Step S7: And update the battery data based on the values predicted by the multi-task deep learning model trained jointly to obtain new data scalars, and input the new data scalars into the virtual battery model of the digital twin to update the virtual battery model of the digital twin. The value updates the battery data to obtain new data scalars, and the new data scalars are input into the virtual battery model of the digital twin to update the virtual battery model of the digital twin.

2. The health prediction method of an aviation power battery digital twin integrated with deep learning according to claim 1, characterized in that: In step S1, the specific process of constructing the digital twin virtual battery model is as follows: Construct a double RC equivalent circuit model as the virtual battery model of the digital twin. The virtual battery model of the digital twin consists of an ohmic resistance , two parallel RC circuits, and an open-circuit voltage ; the two parallel RC circuits include a first parallel RC circuit 1 and a second parallel RC circuit 2 ; the first parallel RC circuit 1 includes a first resistor and a first capacitor ; the second parallel RC circuit 2 includes a second resistor and a second capacitor ; The output of the virtual battery model of the digital twin is the terminal voltage of the battery , the dynamic behavior, indicating: ； Wherein, represents the terminal voltage of the battery at time step t; represents the battery current at time step t; and respectively represent the polarization voltages of the first parallel circuit RC 1 and the second parallel circuit RC 2 at time step t; and the dynamic behavior, indicating: ； ； ； ； In the formula, represents the time constant of the first parallel RC 1 circuit and indicates the fast polarization dynamics; represents the time constant of the second parallel RC 2 circuit and indicates the slow polarization dynamics; represents the base of the natural logarithm; represents the time variable.

3. The health prediction method of an aviation power battery digital twin integrated with deep learning according to claim 2, characterized in that: In step S2, the specific process of the new real-time data matrix is as follows: The battery data is collected in real time by sensors for processing to construct a real-time data matrix , which means: ； In the formula, represents the battery current at the time step of ; represents the battery current at the time step of ; represents the time window length; represents the temperature at the time step of ; represents the temperature at the time step of ; represents the temperature at the time step of ; represents the battery capacity at the time step of t; represents the battery capacity at the time step of ; represents the battery capacity at the time step of ; Record the dynamic parameter S in the scalar of the real-time collected battery data, which represents: ； In the formula, represents the depth of discharge; represents the initial battery capacity; Input the real-time data matrix and the dynamic parameter S in the scalar of the battery data into the virtual battery model of the digital twin for real-time status update to obtain a new real-time data matrix , which means: ； In the formula, represents the terminal voltage of the battery at time step \(t\); represents the time step as the terminal voltage of the battery at the moment; represents the time step as the terminal voltage of the battery at the moment; represents the new real-time data matrix the dimension of, a real matrix; \(m\) represents the feature dimension.

4. The health prediction method of an aviation power battery digital twin integrated with deep learning according to claim 3, characterized in that: In step S3, the specific process of the output of the pooling layer is as follows: Input the new real-time data matrix into the input layer of the convolutional neural network layer to obtain the output of the input layer; Transfer the output of the input layer to the convolutional layer of the convolutional neural network layer, perform convolutional kernel and dynamic gated piecewise activation function operations to obtain the result of the convolutional kernel operation, process the result of the convolutional kernel operation to obtain the output result of the dynamic gated piecewise activation function, and process the result of the convolutional kernel operation and the output result of the dynamic gated piecewise activation function to obtain an output dimension of Output of the convolutional layer , Indicates the number of convolutional kernels, which means: ； In the formula, represents the output of the th convolutional layer; represents the convolutional kernel size; Transfer the output of the convolutional layer to the pooling layer of the convolutional neural network layer for pooling operation to obtain an output dimension of the output of the pooling layer , which means: ； ； In the formula, represents the max pooling function; represents the number of time points after the pooling operation; represents the stride; represents the window size of the pooling layer.

5. The health prediction method for an aviation power battery digital twin integrated with deep learning according to claim 4, wherein: The output result of the dynamic gating piecewise activation function, the specific process is as follows: Calculate the result of the operation on the convolution kernel to obtain the mean value and the standard deviation ; Based on the mean value and the standard deviation generate a dynamic threshold, which means: ； ； In the formula, represents the first dynamic threshold; represents the learnable parameter of the first dynamic threshold; represents the second dynamic threshold; represents the learnable parameter of the second dynamic threshold; Based on the mean value and the standard deviation generate gating parameters, indicating: ； In the formula, ; represents the th piecewise slope; ; represents the th intercept; represents the lightweight fully connected network; Divide the interval of the result of the convolutional kernel operation by a threshold and implement smooth transition through the activation function to obtain the output result of the dynamic gating piecewise activation function, which represents: ； ； In the formula, represents the output result of the dynamic gating segmented activation function, and according to the interval division of the input it realizes the smooth transition of different linear segments through interpolation; represents the Sigmoid activation function; represents the result after the convolution operation; represents the exponential part.

6. The health prediction method for an aviation power battery digital twin integrated with deep learning according to claim 5, characterized in that: In step S4, the specific process of the enhanced context vector is as follows: The output of the pooling layer is transmitted to a multi-scale hierarchical bidirectional long short-term memory network layer for processing to obtain an enhanced context vector; the multi-scale hierarchical bidirectional long short-term memory network layer includes a short-term layer and a long-term layer; Transfer the output of the pooling layer to the short-term layer for forward and backward propagation to obtain the short-term hidden state at time step t ; The short-term hidden state at time step t is input into the long-term layer through skip connections and the long-term hidden state at the time step is calculated. ; Concatenate the long-term hidden state of the time step to obtain a bidirectional hidden state matrix with an output dimension of ; ; represents the number of hidden layer units in a single direction; For the bidirectional hidden state matrix Calculate the attention scores at time step t and the feature dimension of the feature attention scores , and then combine the spatio-temporal attention weights to generate an enhanced context vector , which means: ； ； ； In the formula, represents the activation function; represents the temporal attention weight matrix; represents the temporal attention bias term; represents the feature dimension attention weight matrix; represents the feature dimension attention bias term; represents the hidden states of all time steps in the long-term layer in the feature dimension ; represents the hidden state value of the long-term layer at time step t in the feature dimension ; represents the total number of time steps.

7. A health prediction method for an aviation power battery digital twin integrated with deep learning according to claim 6, characterized in that: In step S5, the specific process of the output of the second fully connected layer is as follows: Concatenate the enhanced context vector with the scalar dynamic parameter S of the battery data, and the concatenated result is a concatenated column vector with a dimension of ; ; denotes the number of dynamic parameters, and it is expressed as: ； Concatenate column vectors and process them in a multi-layer perceptron layer to obtain the output of the second fully connected layer ; The multi-layer perceptron layer includes a first fully connected layer and a second fully connected layer; The first fully connected layer includes the weight matrix of the first fully connected layer and the bias term of the first fully connected layer ; The second fully connected layer includes the weight matrix of the second fully connected layer and the bias term of the second fully connected layer ; Concatenate the column vectors with the weight matrix of the first fully connected layer to perform matrix multiplication to obtain the result of the first matrix multiplication. Add the result of the first matrix multiplication to the bias term of the first fully connected layer and perform a non-linear transformation through the activation function ReLU to obtain the output of the first fully connected layer , which is expressed as: ； The output of the first fully connected layer and the weight matrix of the second fully connected layer are subjected to matrix multiplication operation to obtain the result of the second matrix multiplication operation. The output of the first fully connected layer , the result of the second matrix multiplication operation and the bias term of the second fully connected layer are added together and subjected to non-linear transformation through the activation function ReLU to obtain the output of the second fully connected layer , which is expressed as: 。 8. A health prediction method for an aviation power battery digital twin integrated with deep learning according to claim 7, characterized in that: The value predicted by the multi-task deep learning model jointly trained in step S6 The specific process is as follows: The output of the second fully connected layer is input to the task branch layer for processing, obtaining the value predicted by the multi-task deep learning model for joint training; The task branch layer includes an SOH prediction branch; The SOH prediction branch includes the weight matrix of the first branch and the bias term of the first branch ; The output of the second fully connected layer and the weight matrix of the first branch are subjected to matrix multiplication to obtain the result of the first matrix multiplication operation, and the result of the first matrix multiplication operation is added to the bias term of the first branch , and finally, a non-linear transformation is performed using an activation function to obtain the value predicted by the jointly trained multi-task deep learning model, indicating: 。 9. The health prediction method of digital twin fusion deep learning for an aviation power battery according to claim 8, wherein: In step S7, the specific process of updating the digital twin virtual battery model is as follows: Predicted by a multi-task deep learning model with joint training value Dynamically update the dynamic parameter S in the scalar for battery data to obtain a new data scalar , where is the resistance value of the first resistor after update ; is the capacitance value of the first capacitor after update ; is the resistance value of the second resistor after update ; is the capacitance value of the second capacitor after update ; Input the new data scalar into the virtual battery model of the digital twin to update the virtual battery model of the digital twin, indicating: ； ； ； ； In the formula, represents the sensitivity coefficient of the value pair of the first resistor to degradation predicted by the jointly trained multi-task deep learning model; represents the sensitivity coefficient of the value pair of the first capacitor to degradation predicted by the jointly trained multi-task deep learning model; represents the sensitivity coefficient of the value pair of the second resistor to degradation predicted by the jointly trained multi-task deep learning model; represents the sensitivity coefficient of the value pair of the second capacitor to degradation.

Images