Method and system for measuring and calculating cycle aging of lithium battery under irregular working condition based on DCN-AM
The DCN-AM model captures the complex interaction influence in the aging process of lithium battery, solves the accuracy of lithium battery capacity attenuation under irregular working conditions, and achieves more accurate capacity attenuation prediction.
Patent Information
- Application Number
- CN202510788105.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-13
- Publication Date
- 2025-09-02
AI Technical Summary
The prior art fails to accurately reflect the complex interaction influence under irregular working conditions in the cyclic aging modeling of lithium batteries, resulting in large errors in capacity attenuation calculation.
Using a DCN-AM-based model, the complex interactions and dependencies in the aging process of lithium battery cyclic aging is captured through deep cross networks and attention mechanisms, and the battery operation data is used for accurate prediction.
It significantly improves the accuracy of the cyclic aging calculation of lithium batteries under irregular working conditions, reduces model errors, and improves the accuracy of battery capacity attenuation prediction.
Smart Images
Figure CN120577702A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of new energy technology, and in particular to a DCN-AM-based lithium battery cycle aging calculation method under irregular working conditions and a calculation system thereof. Background Art
[0002] As renewable energy penetration increases in the power grid, electrochemical energy storage systems, particularly lithium-ion batteries, offer strong support for balancing supply and demand, improving energy utilization, and enhancing grid reliability due to their high energy density and long lifespan. Understanding the marginal operating costs of batteries as they participate in electricity markets and operate as auxiliary systems is crucial; these costs are key to power market pricing and operational strategies. Research on capacity degradation during battery aging is crucial for a deeper understanding and accurate calculation of these marginal operating costs.
[0003] Traditional battery capacity degradation modeling often uses simplified and fixed test conditions. While this simplifies calculations, it introduces errors due to its inability to accurately reflect actual conditions. Existing technologies only consider the battery's depth of discharge, or only the depth of discharge and initial state of charge (SOC), and often assume that the depth and rate of charge and discharge are fixed. However, in actual operation, the depth and rate of charge and discharge of a battery change dynamically based on operating conditions. Battery capacity degradation modeling methods are primarily categorized as mathematical model-based, data-driven, and hybrid approaches. In recent years, the application of artificial intelligence technologies such as deep learning in data-driven battery capacity degradation research has increased, with deep neural networks (DNNs) being widely used. However, DNNs are not straightforward when learning complex interactions between features, resulting in inferior model performance compared to models specifically designed to capture feature interactions, such as deep cross networks (DCNs). Summary of the Invention
[0004] The present invention overcomes the shortcomings of the existing technology and provides a lithium battery cycle aging calculation method and calculation system under irregular working conditions based on DCN-AM (Deep & Cross Network with Attention Mechanism, DCN-AM). By considering the cross-influence between battery cycle aging characteristics and using the DCN-AM model to effectively capture complex interactions and dependencies, the accuracy of lithium battery cycle aging calculation under irregular working conditions is significantly improved.
[0005] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a method for calculating the cycle aging of lithium batteries under irregular working conditions based on DCN-AM, comprising: obtaining the operating data of the lithium battery; inputting the operating data into a trained lithium battery cycle aging calculation model based on DCN-AM to obtain the cycle capacity attenuation prediction result of the lithium battery, and completing the lithium battery cycle aging test; wherein, DCN-AM is a deep cross network attention mechanism.
[0006] In a preferred embodiment of the present invention, the operating data includes the initial state of charge (SOC), depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, initial battery state of health (SOH), and the corresponding battery capacity attenuation.
[0007] In a preferred embodiment of the present invention, a method for acquiring training data for a lithium battery cycle aging estimation model based on a deep cross network attention mechanism includes: Training data is generated through a battery cycle aging simulation system, where the training data covers repeated discharge and charge cycles under different operating conditions until the battery capacity drops to a preset threshold. The training data is collected, where the data dimensions include initial state of charge, depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, initial battery health status, and battery capacity decay.
[0008] In a preferred embodiment of the present invention, a training method for a lithium battery cycle aging estimation model based on DCN-AM comprises the following steps: Obtain historical operating data of lithium battery cycle aging as training samples; use the historical data to train the lithium battery cycle aging measurement model based on DCN-AM, and adjust the model parameters by optimizing the loss function until the model converges.
[0009] In a preferred embodiment of the present invention, a lithium battery cycle aging estimation model based on DCN-AM includes an input layer connected in sequence, a cross network and a deep network arranged in parallel, an attention mechanism layer and an output layer.
[0010] In a preferred embodiment of the present invention, a cross network is used to learn cross combinations between features, and the output x of the l+1 cross layer of the cross network is l+1 The calculation method is: , where x0 is the base layer containing the first-order original features, x l and x l+1 are the input and output of the l+1th cross layer, W l and b l are the weight matrix and bias vector learned by the l+1th cross layer respectively.
[0011] In a preferred embodiment of the present invention, a deep network is used to capture highly nonlinear features in the data, and the output of the first depth layer of the deep network is The calculation method is: in, and are the input and output of the lth depth layer, W l and b l are the weight matrix and bias vector learned by this layer, and f is the activation function.
[0012] In a preferred embodiment of the present invention, the attention mechanism layer is used to fused the output features x of the cross network and the deep network. final Weighted, calculated as: attention weight , output = in, Yes Perform exponential operations to make numerical differences more significant and highlight relative sizes. Yes j ranges from 1 to Do the sum and normalize. is the attention weight.
[0013] In a preferred embodiment of the present invention, the activation function f in the deep network is a ReLU function; The discharge rate and charge rate in the real-time operating data of the lithium battery are C-rates, whose amplitudes are discrete values; the ambient temperature is a discrete value in degrees Celsius; the initial state of charge, depth of discharge, depth of charge and initial battery health status are continuous quantities; The loss function of the model is Mean Squared Error (MSE), and the Adaptive Moment Estimation (ADAM) algorithm is used for training.
[0014] In a preferred embodiment of the present invention, a system for calculating cycle aging of lithium batteries under irregular working conditions based on DCN-AM includes a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of a method for calculating cycle aging of lithium batteries under irregular working conditions based on DCN-AM.
[0015] In a preferred embodiment of the present invention, a device for calculating the cycle aging of a lithium battery under irregular operating conditions includes: a data acquisition module for acquiring real-time operating data of the lithium battery, the real-time operating data including the initial state of charge, depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature and initial battery health status; an aging calculation module for inputting the real-time operating data into a trained DCN-AM-based lithium battery cycle aging measurement model to obtain a single cycle capacity attenuation prediction result of the lithium battery.
[0016] The present invention solves the defects existing in the technical background, and the beneficial technical effects of the present invention are: The present invention's DCN-AM-based method and system for calculating lithium battery cycle aging under irregular operating conditions significantly improves the accuracy of lithium battery cycle aging calculations under irregular operating conditions by considering the cross-influence of battery cycle aging characteristics and effectively capturing complex interactions and dependencies using the DCN-AM model. This method can calculate the capacity decay caused by single-cycle aging of lithium batteries under irregular operating conditions, addressing the potential for errors introduced by existing methods when modeling battery capacity degradation due to simplified test conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The present invention will be further described below with reference to the accompanying drawings and examples.
[0018] Figure 1 It is a scatter plot matrix of data generated in the lithium battery cycle aging measurement method under irregular working conditions based on DCN-AM of the present invention, showing the relationship between each variable and the distribution of each variable itself; Figure 2 It is a scatter plot matrix of data generated in the lithium battery cycle aging measurement method under irregular working conditions based on DCN-AM of the present invention, showing the relationship between each variable and the distribution of each variable itself; Figure 3 This is a schematic diagram of the structure of the DCN-AM network model proposed in the present invention for the method of measuring lithium battery cycle aging under irregular working conditions based on DCN-AM, showing the cross network, deep network, and attention layer it contains; Figure 4 It is a visualization diagram of a single cross-layer operation in the DCN-AM-based lithium battery cycle aging measurement method under irregular working conditions of the present invention; Figure 5 It is a loss curve diagram during the DCN-AM model training process in the lithium battery cycle aging measurement method under irregular working conditions based on DCN-AM of the present invention; Figure 6 This is the loss curve diagram during the training process of the comparative DNN model; Figure 7 This is a flow chart of the method for calculating the cycle aging of lithium batteries under irregular working conditions based on DCN-AM of the present invention. DETAILED DESCRIPTION
[0019] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. These drawings are simplified schematic diagrams that only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.
[0020] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, bottom, top, etc.), the directional indications are only used to explain the relative positional relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. The terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. Unless otherwise clearly specified and defined, the terms "set", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected, or indirectly connected through an intermediate medium, or it can be a communication between the internal parts of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. Example 1
[0021] A method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM, comprising: Obtain the operating data of the lithium battery (i.e., the operating data of the lithium battery to be measured); the operating data includes the initial state of charge (SOC), depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, initial battery state of health (SOH), and the corresponding battery capacity decay.
[0022] The operating data is input into a trained lithium battery cycle aging estimation model based on the Deep & Cross Network with Attention Mechanism (DCN-AM). DCN-AM is a deep cross network attention mechanism. This method obtains the cycle capacity decay prediction results of the lithium battery and completes the lithium battery cycle aging test. Specifically, the method for obtaining training data for the lithium battery cycle aging estimation model based on the deep cross network attention mechanism includes: generating training data through a battery cycle aging simulation system, the training data covering repeated discharge and charge cycles under different operating conditions until the battery capacity drops to a preset threshold; collecting the training data, the data dimensions of which include initial state of charge, depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, initial battery health status, and battery capacity decay.
[0023] Specifically, the training method of the DCN-AM-based lithium battery cycle aging calculation model includes the following steps: obtaining historical operating data of lithium battery cycle aging as training samples; using the historical data to train the DCN-AM-based lithium battery cycle aging calculation model, and adjusting the model parameters by optimizing the loss function until the model converges.
[0024] Among them, the lithium battery cycle aging measurement model based on DCN-AM includes an input layer connected in sequence, a cross network and a deep network set in parallel, as well as an attention mechanism layer and an output layer.
[0025] The cross network is used to learn the cross combination between features, and the output x of the l+1th cross layer of the cross network is l+1 The calculation method is: , where x0 is the base layer containing the first-order original features, x l and x l+1 are the input and output of the l+1th cross layer, W l and b l are the weight matrix and bias vector learned by the l+1th cross layer respectively.
[0026] Among them, the deep network is used to capture the highly nonlinear features in the data, and the output of the lth depth layer of the deep network is The calculation method is: in, and are the input and output of the lth depth layer, W l and b l are the weight matrix and bias vector learned by this layer, and f is the activation function.
[0027] Among them, the attention mechanism layer is used to integrate the output features x of the cross network and the deep network.final Weighted, calculated as: attention weight , output = in, Yes Perform exponential operations to make numerical differences more significant and highlight relative sizes. Yes j ranges from 1 to Do the sum and normalize. is the attention weight.
[0028] The activation function f in the deep network is a ReLU function; the discharge rate and charge rate in the real-time operating data of the lithium battery are C rates, whose amplitudes are discrete values; the ambient temperature is a discrete value in degrees Celsius; the initial state of charge, depth of discharge, depth of charge, and initial battery health status are continuous quantities; the loss function of the model is mean squared error (MSE), and the adaptive moment estimation algorithm (ADAM) is used for training. Example 2
[0029] Based on Example 1, the lithium battery cycle aging measurement model structure based on DCN-AM includes an input layer, a cross network, a deep network, and an attention layer (AM layer).
[0030] Input layer: The collected battery aging test data dimensions are used as input features, including initial SOC (variable 0), discharge depth (variable 1), charge depth (variable 2), discharge rate (variable 3), charge rate (variable 4), ambient temperature (variable 5), and initial battery health status (variable 6).
[0031] Cross network: Input data x0 is fed into the cross network in parallel. The cross network directly learns the cross combination between features through a specific cross layer. The output of the l+1th cross layer is By formula Calculated, where x0 is the base layer containing the first-order original features, x l and x l+1 The input and output of the l+1th cross layer, W0 and b l are the learned weight matrix and bias vector.
[0032] Deep network: Input data x0 is fed into the deep network in parallel. Deep network is used to capture highly nonlinear interactions. The network output of its lth layer is By formula Calculated, where and are the input and output of the lth depth layer, is the weight matrix, is the bias vector, and f is the activation function (such as ReLU).
[0033] Feature fusion: the output x of the cross network l1 and the output of the deep network h l2 are connected to form a combined feature x final =[x l1 ;h l2 ].
[0034] Attention mechanism layer: for the fused feature x final Apply the attention mechanism. The attention mechanism calculates the attention weight matrix A and assigns different weights to different parts of the input data, allowing the model to focus on more critical information. The element a in the attention weight matrix i pass calculate.
[0035] Output layer: The final output (the predicted battery capacity attenuation, i.e. variable 7) is obtained by the attention weight matrix A and the input data x final Calculated, The model uses MSE as the loss function and is trained with ADAM. Example 3
[0036] This paper addresses the accuracy issue of lithium battery cycle aging calculation under irregular operating conditions and proposes a DCN-AM based calculation method. The method includes the following steps: (1) Generation of battery cycle aging data: The battery cycle aging data used for training in this paper comes from a battery cycle aging simulation system built in MATLAB. This lithium battery model can simulate the effects of ambient temperature, battery aging, and dynamic internal resistance, which are highly correlated with battery degradation. Based on this battery model, a battery cycle generator was designed and added to simulate discharge and charge cycles at dynamically changing discharge and charge rates and depths.
[0037] For each battery aging test, each cycle is simulated as discharging from a specific SOC to a lower SOC, followed by charging to a new SOC (this SOC value is typically different from the starting SOC), with varying discharge and charge rates. Data dimensions collected include: initial SOC (variable 0), depth of discharge (variable 1), depth of charge (variable 2), discharge rate (variable 3), charge rate (variable 4), ambient temperature (variable 5), initial SOH at each cycle (variable 6), and battery capacity fade (variable 7, which refers to the loss of battery capacity after a single charge / discharge cycle, relative to the battery's maximum capacity). Each battery aging test represents a group of batteries that are repeatedly charged and discharged under different operating conditions until the battery capacity drops to 80% of its maximum rated capacity. The charge and discharge rate (C-rate) is assumed to have a discrete range of [0.125, 0.25, 0.375, 0.5, 0.75, 1, 1.5, 2], and the temperature is assumed to have a discrete range of [15, 17, 20, 23, 25, 30, 35] degrees Celsius. Initial SOC, depth of discharge, depth of charge and SOH are continuous quantities.
[0038] The scatter plot matrix of the generated data is as follows Figure 1 、 Figure 2 As shown, Figure 1 and Figure 2 The relationships between different pairs of variables are shown. Figure 1 and Figure 2 It can be seen that variables 0, variable 1, variable 2, variable 6, and variable 7 are continuously distributed, while variables 3, variable 4, and variable 5 are discretely distributed, and there is an obvious nonlinear relationship between the variables.
[0039] (2) Structure and verification of DCN-AM model: The DCN-AM network model structure proposed by the present invention is as follows: Figure 3 As shown in the figure, it consists of three parts: cross network, deep network and attention layer.
[0040] Cross network: The structure of the cross network is as follows Figure 3 The operation of a single cross layer is as shown in the Cross network section. Figure 4 The core idea of the cross network is to effectively apply explicit feature crossover. The output x of the l+1th cross layer is l+1 The algorithms include: ;in, It is the base layer containing the first-order original features, usually set as the input layer, and d is the number of features. l ,x l+1 ∈R d Represent the input and output of the l+1th cross layer respectively. l ∈R d×dand b l ∈R d are the learned weight matrix and bias vector.
[0041] Deep network: To capture highly nonlinear interactions, a parallel deep network is introduced, whose structure is as follows Figure 5 As shown in the Deep network part. The output h of the lth layer network l+1 The algorithms include: ;in, are the input and output of the lth depth layer respectively. is the weight matrix, is the bias vector. It is the element activation function, which is set to ReLU in this embodiment.
[0042] Feature fusion and attention mechanism (AM layer): The input x0 is fed into the cross network and the deep network in parallel. Then, the output x of the cross network is l1 and the output of the deep network h l2 are concatenated to create the input x final =[x l1 ;h l2 Then, final Set up the attention mechanism. The basic principle of the attention mechanism is to assign different weights to different parts of the input data. These weights represent the importance of each part for completing a specific task. The core of the attention mechanism is to calculate an attention weight matrix A, which is used to weighted sum the important information. The input data is , then the element a in the attention weight matrix A i The algorithms include: .
[0043] The final output (predicted battery capacity decay) is obtained by the attention weight matrix A and the input data x final The algorithms include: .
[0044] During model training, variables 0, 1, 2, 3, 4, 5, and 6 were selected as feature variables, and variable 7 was selected as the output variable. The number of layers in both the DCN and DNN was set to 3. The number of hidden neurons in the DNN was 128, 64, and 8, respectively. Algorithms that use the mean squared error (MSE) as the network loss function include: , and use the adaptive moment estimation algorithm (ADAM) to train the neural network. is the true value (simulated battery capacity degradation), is the predicted value (the estimated battery capacity degradation), and N is the number of samples.
[0045] The loss curve of the DCN-AM model during training is as follows: Figure 5 As shown, the mean absolute error percentages on the training and validation sets are 98% and 97%, respectively.
[0046] For comparative analysis, only DNN is selected and trained using the same data. The loss value curve during training is as follows: Figure 6 As shown in Figure 2, the mean absolute error percentage of the DNN on the training set and validation set is 93% and 90%, respectively.
[0047] Due to the randomness in the neural network training process, the result is the best result selected from multiple training sessions. The comparison results show that the DCN-AM proposed in this paper has higher accuracy in calculating the battery cycle aging amount.
[0048] The present invention significantly improves the accuracy of lithium battery cycle aging calculation under irregular working conditions by more finely modeling the various influencing factors in the battery cycle aging process and using the DCN-AM model to effectively capture the complex interactions between these factors. Example 4
[0049] A system for calculating cycle aging of lithium batteries under irregular working conditions based on DCN-AM includes a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of a method for calculating cycle aging of lithium batteries under irregular working conditions based on DCN-AM. Example 5
[0050] A device for calculating the cycle aging of a lithium battery under irregular operating conditions comprises: a data acquisition module for acquiring real-time operating data of a lithium battery, wherein the real-time operating data includes initial state of charge, depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, and initial battery health state; and an aging calculation module for inputting the real-time operating data into a trained DCN-AM-based lithium battery cycle aging calculation model to obtain a single-cycle capacity attenuation prediction result of the lithium battery.
[0051] Working principle: The present invention's DCN-AM-based method and system for calculating lithium battery cycle aging under irregular operating conditions significantly improves the accuracy of lithium battery cycle aging calculations under irregular operating conditions by considering the cross-influence of battery cycle aging characteristics and effectively capturing complex interactions and dependencies using the DCN-AM model. This method can calculate the capacity decay caused by single-cycle aging of lithium batteries under irregular operating conditions, addressing the potential for errors introduced by existing methods when modeling battery capacity degradation due to simplified test conditions.
[0052] The present invention models the battery cycle aging process more accurately, takes into account multiple key factors that affect battery capacity decay, and does not simplify the discharge and charge processes in a single cycle, thereby more accurately reflecting the actual situation.
[0053] Compared to methods using only deep neural networks, the proposed DCN-AM model-based calculation method can more accurately calculate the single-cycle aging of batteries under irregular operating conditions. Experimental results show that the DCN-AM model achieves a mean absolute error of 98% and 97% on the training and validation sets, respectively, surpassing the 93% and 90% achieved using the DNN model alone.
[0054] The DCN-AM model of the present invention can directly learn the cross-combination between features through its cross-network structure, and combined with the attention mechanism, it enables the network to not only extract deep information from the raw data, but also assign different weights to important information to enhance its impact on the final prediction results, and can effectively capture feature interactions; this is especially important for processes such as battery aging that are affected by the interaction of multiple factors.
[0055] The above specific implementation methods are specific support for the scheme ideas proposed in the present invention, and cannot be used to limit the scope of protection of the present invention. Any equivalent changes or equivalent modifications made on the basis of this technical scheme in accordance with the technical ideas proposed in the present invention still fall within the scope of protection of the technical scheme of the present invention.
Claims
1. A method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM, characterized in that: include: Obtain lithium battery operating data; The operating data is input into the trained lithium battery cycle aging measurement model based on DCN-AM to obtain the lithium battery cycle capacity attenuation prediction results and complete the lithium battery cycle aging test; among them, DCN-AM is a deep cross network attention mechanism.
2. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 1, characterized in that: The operating data includes initial state of charge, depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, initial battery health status, and corresponding battery capacity decay.
3. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 1, characterized in that: The method for obtaining training data for the lithium battery cycle aging measurement model based on the deep cross network attention mechanism includes: Training data is generated through a battery cycle aging simulation system, where the training data covers repeated discharge and charge cycles under different operating conditions until the battery capacity drops to a preset threshold. The training data is collected, where the data dimensions include initial state of charge, depth of discharge, depth of charge, discharge rate, charge rate, ambient temperature, initial battery health status, and battery capacity decay.
4. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 3, characterized in that: The training method of the DCN-AM-based lithium battery cycle aging measurement model includes the following steps: Obtain historical operating data of lithium battery cycle aging as training samples; use the historical data to train the lithium battery cycle aging measurement model based on DCN-AM, and adjust the model parameters by optimizing the loss function until the model converges.
5. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 4, characterized in that: The lithium battery cycle aging measurement model based on DCN-AM includes an input layer connected in sequence, a cross network and a deep network set in parallel, as well as an attention mechanism layer and an output layer.
6. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 5, characterized in that: The cross network is used to learn the cross combination between features, and the output x of the l+1th cross layer of the cross network l+1 The algorithm is: , where x0 is the base layer containing the first-order original features, x l and x l+1 are the input and output of the l+1th cross layer, W l and b l are the weight matrix and bias vector learned by the l+1th cross layer respectively.
7. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 6, characterized in that: The deep network is used to capture the highly nonlinear features in the data, and the output of the lth depth layer of the deep network The algorithm is: in, and are the input and output of the lth depth layer, W l and b l are the weight matrix and bias vector learned by this layer, and f is the activation function.
8. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 7, characterized in that: The attention mechanism layer is used to integrate the features x after the output of the cross network and the deep network final For weighted addition, the algorithm is: Attention weight ;output= ; in, Yes Perform exponential operations to make numerical differences more significant and highlight relative sizes. Yes j ranges from 1 to Do the sum and normalize. is the attention weight.
9. The method for calculating lithium battery cycle aging under irregular working conditions based on DCN-AM according to claim 8, characterized in that: The activation function f in the deep network is a ReLU function; The discharge rate and charge rate in the real-time operating data of the lithium battery are C rates, whose amplitudes are discrete values; the ambient temperature is a discrete value in degrees Celsius; the initial state of charge, depth of discharge, depth of charge and initial battery health status are continuous quantities; The loss function of the model is mean square error, and the adaptive moment estimation algorithm is used for training.
10. A lithium battery cycle aging measurement system under irregular working conditions based on DCN-AM, characterized in that: It includes a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of a method for measuring lithium battery cycle aging under irregular working conditions based on DCN-AM according to any one of claims 1 to 9.