A digital twin battery construction method based on electromagnetic detection technology

Abstract

The application discloses a kind of digital twin battery construction methods based on electromagnetic detection technology, first, different models different batches of energy storage batteries are collected, electromagnetic detection equipment is used to detect the electromagnetic signal of each energy storage battery under different charge-discharge states, and the electrical, thermal and corresponding energy storage battery state performance parameters are collected synchronously;Then carry out isomeric data synchronous calibration, respectively through pulse detection and quality assessment to clean up the collected electromagnetic data, delete low signal-to-noise ratio data;Based on the deep neural network model of multi-task learning, define to estimate different battery state performance parameters as subtask, design its respective loss function, establish the deep neural network model based on feature weight sharing, use joint training to realize the mapping relationship of multi-source signal and battery state evolution by using the correlation of subtask, to construct digital twin battery model, solve the problem of inaccurate real-time evaluation of battery state, realize the intelligent management of energy storage battery full life cycle.

Description

Technical Field

[0001] This invention belongs to the field of electromagnetic detection and digital twin technology, and mainly relates to a method for constructing a digital twin battery based on electromagnetic detection technology. Background Technology

[0002] Due to their advantages such as high energy density and long cycle life, energy storage batteries are widely used in various fields. However, the safety of their operation has become one of the pain points that restricts their widespread application. The key lies in accurately predicting the battery status and maintaining or replacing the battery in a timely manner.

[0003] As the number of charge-discharge cycles of energy storage batteries increases, phenomena such as battery capacity decay and increased internal resistance will occur, and the safety and reliability of operation will gradually decrease. In order to ensure that the battery can operate safely and stably, it is necessary to collect several battery state data, including battery state of charge (SOC), state of health (SOH), remaining life, remaining cycle count, charging time, and discharging time, to monitor the degree of battery aging and predict the state of the battery throughout its entire life cycle.

[0004] Existing methods for constructing energy storage batteries primarily rely on physical testing and experiments to determine battery performance and lifespan. This approach is time-consuming, costly, and lacks satisfactory accuracy. Digital twins, on the other hand, fully utilize data from physical models, sensor updates, and operational history, integrating multi-disciplinary, multi-physical-quantity, multi-scale, and multi-probabilistic simulation processes to map data in a virtual space, thereby reflecting the entire lifecycle of the corresponding physical equipment. By processing historical battery state data using digital twin technology, a virtual model of the battery's operational state can be constructed, uncovering implicit battery health information and evolution patterns, enabling real-time updates and dynamic evolution of the battery's state. Summary of the Invention

[0005] This invention addresses the problems of time-consuming, costly, and inaccurate traditional methods in existing technologies by providing a digital twin battery construction method based on electromagnetic detection technology. First, it collects energy storage batteries of different models and batches. Electromagnetic detection equipment is used to detect electromagnetic signals of each battery under different charge and discharge states, collecting electromagnetic data and simultaneously collecting electrical, thermal, and corresponding battery state performance parameters. Next, heterogeneous data synchronous calibration is performed. The collected electromagnetic data is cleaned through pulse detection and quality assessment, removing low signal-to-noise ratio data. Based on a multi-task learning deep neural network model, estimating different battery state performance parameters is defined as a sub-task, and its respective loss function is designed. A deep neural network model based on feature weight sharing is established. Joint training is used to construct the mapping relationship between multi-source signals and battery state evolution by utilizing the correlation of sub-tasks, thereby constructing a digital twin battery model. This method can solve the problem of inaccurate real-time battery state assessment and realize intelligent management of the entire life cycle of energy storage batteries.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a method for constructing a digital twin battery based on electromagnetic detection technology, comprising the following steps:

[0007] S1: Collect historical state data of energy storage batteries: Collect energy storage batteries of different models and batches, use electromagnetic detection equipment to detect electromagnetic signals of each energy storage battery under different charging and discharging states, and simultaneously collect electrical, thermal and corresponding energy storage battery state performance parameters.

[0008] S2: Data preprocessing: First, perform heterogeneous data synchronization calibration, and then clean the electromagnetic data collected in step 1 by pulse detection and quality assessment respectively, and delete low signal-to-noise ratio data.

[0009] S3, Digital Twin Battery Model Construction: Based on a deep neural network model for multi-task learning, the model defines the estimation of performance parameters of different battery states as a sub-task, designs its respective loss function, establishes a deep neural network model based on feature weight sharing, and uses joint training to construct the mapping relationship between multi-source signals and battery state evolution by utilizing the correlation of sub-tasks, thereby constructing a digital twin battery model.

[0010] Furthermore, the battery historical state data in step S1 includes at least internal resistance, state of charge (SOC), state of health (SOH), remaining cycle count, charging time, and discharging time.

[0011] Furthermore, the pulse detection in step S2 specifically employs an autocorrelation detection algorithm or a deep learning-based pulse detection algorithm to detect pulse signals; the quality assessment metrics include at least pulse integrity, pulse glitches, frequency stability, noise variance, signal multipath, and signal-to-noise ratio.

[0012] Furthermore, step S3 specifically includes:

[0013] Step S31: Establishment of a neural network model for multi-task learning: The neural network model can use time-frequency transformed data and perform feature extraction using a multi-layer convolutional neural network, or use the original data and perform feature extraction using a sequence coding network; multiple tasks share network weights;

[0014] Step S32: Model weight sharing: Under a unified neural network model, define shared layers for different tasks and specific layers for each task to construct a multi-task-based battery state evolution model; during model training, train the neural network using joint learning, where the parameters of the shared layers will be used and updated by multiple tasks, while the parameters of the task-specific layers will only be used and updated by the specific task.

[0015] Step S33: Model Evaluation and Adjustment: Train the model, evaluate and adjust the model using the validation set and test set, assess the model's generalization performance and predictive ability, apply the trained model to actual battery state monitoring, and establish a mapping between multi-source signal features and battery state evolution.

[0016] Furthermore, in step S1, the electrical signal is acquired in real time with precision down to the individual battery level using a voltage / current sensor; the thermal signal is acquired in real time after a temperature field is constructed using a temperature sensor deployed in the battery compartment; and the electromagnetic signal is measured by a radio frequency front-end deployed in the battery compartment to measure the electromagnetic signal generated by each individual battery during operation. Furthermore, the battery state evolution model in step S32 specifically includes:

[0017] (1) The RC equivalent circuit model of a single cell second-order battery is used to explain the internal dynamic behavior of the battery and to evaluate the state of charge (SOC) and state of health (SOH) of the battery.

[0018] (2) Single cell electrothermal coupling model, used to describe the internal electrochemical reaction, heat transfer and electrical transfer process of the battery. By calculating the relationship between the internal temperature and electrical performance of the battery, the battery's service life and state of health (SOH) can be predicted.

[0019] (3) Electromagnetic fitting model of single cell: By learning from the historical data of battery performance, a nonlinear mapping relationship between state of charge (SOC), state of health (SOH) and electromagnetic signal is established, and it is used to correct the predicted values ​​of electrothermal coupling model and single cell / system-second-order battery RC equivalent circuit model.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] (1) The present invention uses electromagnetic detection technology, which can monitor and diagnose the internal state of the battery without disassembling the battery, avoiding the cost and risk of disassembling and reassembling the battery; through digital twin technology, electromagnetic, electrical, thermal and other signals are associated with the battery state, which can realize high-precision battery state prediction and health management, and improve the battery's service life and safety; in addition, it can also realize real-time monitoring and tracking of the battery state, timely detection of changes and anomalies in the battery state, and improve the battery's safety and reliability.

[0022] (2) Compared with traditional methods that describe battery aging state and path based on mathematical or empirical models, the method of the present invention does not require the establishment of a white box model of capacity decay, avoids the complex aging mechanism and reaction inside the battery, and only uses mathematical statistical analysis methods to explore the intrinsic relationship between input and output. It has high flexibility and strong adaptability, and can better meet the requirements of online, high precision and fast prediction.

[0023] (3) The method of the present invention can also visualize battery status information, making it convenient for users to monitor and manage the status, and improving the efficiency and safety of battery use. Compared with traditional methods, it can adaptively optimize the battery status, and adjust charging, discharging and other operations according to the real-time monitored status information, thereby improving the battery's lifespan and performance, which is of great significance for improving the battery's lifespan and safety. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the steps of the digital twin battery construction method based on electromagnetic detection technology of the present invention.

[0025] Figure 2 This is a schematic diagram of the digital twin battery system. Detailed Implementation

[0026] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0027] Example 1

[0028] A method for constructing a digital twin battery based on electromagnetic detection technology, such as Figure 1 As shown, it includes the following steps:

[0029] Step S1: Collect multiple energy storage batteries of different models and batches, deploy electromagnetic detection equipment on each, and detect their electromagnetic signals under different charge and discharge states. Simultaneously collect electrical, thermal, and corresponding energy storage battery state performance parameters.

[0030] Multiple energy storage batteries of different models and batches are collected. These batteries may come from different manufacturers or suppliers, or from different batches of the same manufacturer or supplier, and their electromagnetic signals are tested. Electromagnetic detection equipment, such as electromagnetic wave sensors or magnetic field sensors, is placed next to or inside the batteries to collect their electromagnetic signal data. During data collection, the battery's operating state needs to be controlled, such as its charge / discharge status and temperature, to ensure the representativeness of the collected data.

[0031] In this step, different electromagnetic detection instruments can be selected for data acquisition. For example, a magnetic field scanner based on the principle of magnetic field induction can be used to measure changes in the magnetic field in the battery's vertical, horizontal, front, and back directions, obtaining information such as the battery's internal current and electromagnetic field. Alternatively, a conductivity or resistivity meter can be used to measure the battery's conductivity and resistance to determine its state and performance.

[0032] When collecting data, data collection points and sampling intervals can be set according to actual conditions. The collected historical battery data should include at least the battery internal resistance, state of charge (SOC), state of health (SOH), remaining cycle count, charging time, and discharging time to obtain a comprehensive picture of state changes.

[0033] Step S2: Perform data preprocessing using methods such as pulse detection and quality assessment.

[0034] This step first performs heterogeneous data synchronization calibration. Through data mapping, the source data is converted into the target data format, and synchronization calibration is achieved through data alignment and interpolation. The calibrated data will then undergo impulse detection, followed by data cleaning based on defined quality assessment metrics, including at least impulse integrity, impulse glitches, frequency stability, noise variance, signal multipath, and signal-to-noise ratio. Impulse detection can employ autocorrelation detection algorithms or deep learning-based impulse detection algorithms, which can effectively mark and process data exceeding thresholds. By combining the autocorrelation function, impulse signal detection can be achieved by establishing and training a deep learning model with dynamically configured parameters. Then, based on the assessment metrics and set thresholds, data with low signal-to-noise ratios is deleted, thus cleaning the data. After impulse detection and quality assessment, data can be corrected or deleted based on the detection results. The cleaned data needs to be validated to ensure that the cleaning process did not introduce new errors or data biases.

[0035] Step S3: Based on the deep neural network model of multi-task learning, define the estimation of performance parameters of different battery states as sub-tasks, design their respective loss functions, establish a deep neural network model based on feature weight sharing, and use joint training to realize the mapping relationship between multi-source signals and battery state evolution by utilizing the correlation of sub-tasks, thereby constructing a digital twin battery model.

[0036] The method for building a deep neural network model for multi-task learning is as follows: Design and construct a deep neural network model based on multi-task learning; train the model using a prepared dataset; during training, model weight sharing can be employed; and evaluate the trained model using an independent test set to assess the accuracy of the battery state evolution model in identifying battery states. Apply the trained model to practical battery state monitoring and prediction tasks. By inputting new electromagnetic signal data, the model can predict the battery's state evolution, establishing a mapping between electromagnetic signal features and battery state evolution.

[0037] Model weight sharing refers to the mechanism in deep neural networks that allows the same model or parts of a model to share parameter weights across different tasks or inputs. Typically, each task has its own independent model trained, and parameter weights are learned independently for each task. However, through weight sharing, the parameter weights of certain layers or modules can be shared across multiple tasks or inputs.

[0038] In this step, a unified deep neural network structure such as Convolutional Neural Network (CNN), Recurrent Neural Network (GRU), or Long Short-Term Memory (LSTM) can be designed. Shared layers for different tasks can be defined within the neural network model to achieve parameter sharing across tasks. During training, multi-task learning techniques are incorporated to establish a mapping relationship between battery state and electromagnetic effective information features. The model parameters are then optimized using the backpropagation algorithm to improve the model's predictive ability.

[0039] Specifically, the following implementation methods are required:

[0040] Data preparation: Before performing deep learning, the processed data needs to be divided into training, validation, and test sets. The training and validation sets are used for model training and parameter tuning, while the test set is used to evaluate the model's predictive ability and performance.

[0041] Building a deep neural network model for multi-task learning: First, feature extraction is needed to improve the accuracy of model building and training. One approach is to perform time-frequency transformations on the data using methods such as wavelet transform, Fourier transform, and Kalman filtering, followed by feature extraction using a multi-layer convolutional neural network. Alternatively, sequence coding networks can be used to directly process the raw data, converting it into vectors for input and output. Based on feature extraction, a deep neural network model for multi-task learning is designed and built. The model is trained using a prepared dataset. During training, shared network weights are used across multiple tasks, a unified loss function is constructed, and the same pre-embedding layers and convolutional layers are used to handle common features across all tasks. Then, independent fully connected layers are used on specific tasks to achieve those specific tasks.

[0042] Model weight sharing: To enable models for different tasks to learn in multiple ways, it is necessary to share model parameters across tasks. Under a unified neural network model, shared layers and task-specific layers are defined to construct a multi-task-based battery state evolution model. During training, stochastic gradient descent (SGD) or other optimization algorithms can be used to train the neural network for joint learning. The parameters of the shared layers will be used and updated by multiple tasks, while the parameters of the task-specific layers will only be used and updated for that specific task. This method effectively improves the model's generalization performance while allowing models for different tasks to learn in multiple ways.

[0043] Model Evaluation and Tuning: The model is evaluated and tuned using validation and test sets to assess its generalization performance and predictive ability. If the model performs poorly, model parameters can be readjusted, or new training and validation sets can be selected. Finally, the trained model is applied to predict and monitor battery state, establishing a mapping between multi-source signal features and battery state evolution.

[0044] Figure 2 This is a schematic diagram illustrating the construction principle of a digital twin battery system. During actual battery operation, the data acquisition module first collects the battery's electrical, thermal, and electromagnetic signals. The data acquisition module includes voltage / current sensors from the battery management system (BMS), temperature sensors deployed in the battery compartment, and a radio frequency front-end deployed in the battery compartment.

[0045] The data then undergoes preprocessing, including heterogeneous data synchronization calibration and data cleaning. In specific operations, electromagnetic, electrical, and thermal signals are cleaned through pulse detection and quality assessment, and low signal-to-noise ratio data is removed.

[0046] After acquiring real-time data, a neural network model for multi-task learning is built, and the model is continuously revised based on the real-time data to achieve complete synchronization with the real physical battery. The specific steps include the following:

[0047] Establishing an initial model: Based on historical data and the physical characteristics of the battery, the digital twin battery system establishes a corresponding battery model, specifically including:

[0048] (1) The second-order RC equivalent circuit model of a single cell is based on a circuit modeling method, which treats the battery as a circuit with resistance and capacitance. This model is used to explain the internal dynamic behavior of the battery, such as battery charging and discharging, open circuit voltage, internal resistivity, etc., and can evaluate the battery's state of charge (SOC) and state of health (SOH).

[0049] (2) Single-cell electrothermal coupling model, which is used to describe the electrochemical reactions, heat transfer and electrical transfer processes inside the battery. Based on the thermodynamic principles of the battery, the model is input with the battery temperature signal and calculates the relationship between the internal temperature and electrical performance of the battery to predict the battery's lifespan, state of health (SOH), etc.

[0050] (3) Single-cell electromagnetic fitting model: By learning from historical battery performance data, a nonlinear mapping relationship is established between the state of charge (SOC), state of health (SOH), and electromagnetic signals. This model can accurately predict the battery's SOC and SOH in a short time, and has high flexibility and adaptability. It can be used to correct the predicted values ​​of the electrothermal coupling model and the single-cell / system-second-order battery RC equivalent circuit model.

[0051] These models are used to describe different aspects of battery characteristics and provide a basis for calculating and correcting battery parameters.

[0052] Calculating and correcting battery state parameters: After acquiring real-time data and building a battery model, the system calls the corresponding model to calculate and correct the battery state parameters. State of Health (SOH) and State of Charge (SOC) are important indicators for battery safety and performance evaluation. Calculating and correcting these parameters can help determine the current state of the battery and predict its remaining lifespan.

[0053] Finally, by combining the established digital twin model, parameters such as battery status are obtained and sent to the battery management system (BMS) to assist in decision-making.

[0054] In summary, the method of this invention employs electromagnetic detection technology, enabling the monitoring and diagnosis of the internal state of the battery without disassembling or disconnecting the power, thus avoiding the costs and risks associated with disassembling and reassembling the battery. Through digital twin technology, electromagnetic, electrical, and thermal signals are correlated with the battery state, enabling high-precision battery state prediction and health management, thereby improving battery lifespan and safety. Furthermore, it allows for real-time monitoring and tracking of the battery state, promptly detecting changes and anomalies, and enhancing battery safety and reliability.

[0055] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a digital twin battery based on electromagnetic detection technology, characterized in that, Includes the following steps: S1, collect historical state data of energy storage batteries: collect energy storage batteries of different models and batches, use electromagnetic detection equipment to detect electromagnetic signals of each energy storage battery under different charging and discharging states, collect electromagnetic signals, and simultaneously collect electrical signals, thermal signals and corresponding energy storage battery state performance parameters. S2, Data preprocessing: First, perform heterogeneous data synchronization calibration, and then clean the electromagnetic data collected in step S1 by pulse detection and quality assessment respectively, and delete low signal-to-noise ratio data. S3, Digital Twin Battery Model Construction: Based on a deep neural network model for multi-task learning, the model defines the estimation of performance parameters of different battery states as sub-tasks, designs their respective loss functions, establishes a deep neural network model based on feature weight sharing, and uses joint training to construct the mapping relationship between multi-source signals and battery state evolution by utilizing the correlation of sub-tasks, thereby constructing a digital twin battery model. S31, Establishment of a neural network model for multi-task learning: The neural network model uses time-frequency transformed data and performs feature extraction using a multi-layer convolutional neural network, or uses the original data and performs feature extraction using a sequence coding network; multiple tasks share network weights; S32, Model Weight Sharing: Under a unified neural network model, shared layers for different tasks and task-specific layers are defined to construct a multi-task-based battery state evolution model. During model training, the neural network is trained using joint learning. The parameters of the shared layers are used and updated by multiple tasks, while the parameters of the task-specific layers are used and updated only by that specific task. Specifically, the battery state evolution model includes: (1) The RC equivalent circuit model of a single cell second-order battery is used to explain the internal dynamic behavior of the battery and to evaluate the state of charge (SOC) and state of health (SOH) of the battery. (2) Single cell electrothermal coupling model, used to describe the electrochemical reaction, heat transfer and electrical transfer process inside the battery. By calculating the relationship between the internal temperature and electrical performance of the battery, the battery's service life and state of health (SOH) can be predicted. (3) Electromagnetic fitting model of single cell: by learning from the historical data of battery performance, a nonlinear mapping relationship between state of charge (SOC), state of health (SOH) and electromagnetic signal is established, and it is used to correct the predicted values ​​of electrothermal coupling model and single cell / system-second-order battery RC equivalent circuit model. S33, Model Evaluation and Adjustment: The model is trained, evaluated and adjusted using validation and test sets, the generalization performance and predictive ability of the model are assessed, and the trained model is applied to actual battery state monitoring to establish a mapping between multi-source signal features and battery state evolution.

2. The method for constructing a digital twin battery based on electromagnetic detection technology as described in claim 1, characterized in that: The battery historical state data in step S1 includes at least internal resistance, state of charge (SOC), state of health (SOH), remaining cycle count, charging time, and discharging time.

3. The method for constructing a digital twin battery based on electromagnetic detection technology as described in claim 2, characterized in that: The pulse detection in step S2 specifically involves using an autocorrelation detection algorithm or a deep learning-based pulse detection algorithm to detect pulse signals; the quality assessment indicators include at least pulse integrity, pulse glitches, frequency stability, noise variance, signal multipath, and signal-to-noise ratio.

4. The method for constructing a digital twin battery based on electromagnetic detection technology as described in claim 3, characterized in that: In step S1, the electrical signal is collected in real time with precision down to the individual battery level by the voltage / current sensor of the battery management system (BMS). The thermal signal is collected in real time after a temperature field is constructed by temperature sensors deployed in the battery compartment. The electromagnetic signal is measured by a radio frequency front-end deployed in the battery compartment to measure the electromagnetic signal generated by a single battery during operation.

Images