Cloud-based lithium battery management method and system

Abstract

The invention relates to the technical field of lithium battery management, and discloses a cloud-based lithium battery management method and system, and the method comprises the steps: constructing a cloud collaborative management framework, and obtaining a dynamic adjustment factor and a degradation compensation coefficient; performing multi-dimensional state joint estimation, and predicting the health state and the residual life of the battery; a dynamic balance control strategy is set, and energy differences among the monomers are compensated; and closed-loop feedback and model iteration are carried out. The system comprises a distributed data acquisition unit, an edge computing unit, a cloud analysis platform, a communication gateway and a balance execution unit. According to the method, technologies such as cloud collaborative management, multi-dimensional data processing, dynamic balance control and model iterative optimization are utilized, accurate management of the lithium battery is realized, the battery state prediction accuracy can be improved, the service life of the battery can be prolonged, the system safety and stability can be enhanced, and the method is suitable for various lithium battery application scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery management, and particularly to a cloud-based lithium battery management method and system. Background Art

[0002] Lithium batteries are widely used in fields such as electric vehicles, energy storage systems, and portable electronic devices due to advantages such as high energy density, long cycle life, and low self-discharge rate. However, lithium battery management faces many difficulties, which have become key factors restricting its development and popularization.

[0003] In actual use of lithium batteries, there are differences in the characteristics of individual cells, which are caused by the inconsistency of manufacturing processes. During the charge and discharge process of the battery pack, this difference will gradually amplify, causing some cells to reach the charge and discharge limits prematurely, affecting the overall performance and life of the battery pack. Taking electric vehicles as an example, the inconsistency of individual cells in the battery pack may lead to a shortened driving range, premature battery scrapping, and increased usage costs. Although traditional battery management systems (BMS) have basic balancing functions, most of them adopt static balancing strategies, which are difficult to dynamically adjust according to the real-time state of the battery, resulting in poor balancing effects and unable to fundamentally solve the problem of inconsistent individual cells.

[0004] The accuracy of battery state monitoring and remaining life prediction is crucial but quite challenging. The state of health (SOH) and remaining useful life (RUL) of the battery are affected by many factors, such as charge and discharge current, temperature, number of cycles, etc., and these factors are interrelated and have a complex relationship. Current state monitoring and prediction methods often rely on only one or a few parameters and cannot comprehensively consider the operating conditions of the battery, resulting in large errors in prediction results. For example, in an energy storage system, inaccurate remaining life prediction may prevent the system from arranging maintenance or battery replacement in advance, resulting in failures at critical moments and affecting the stability of power supply.

[0005] With the continuous expansion of the application scenarios of lithium batteries, the difficulty of data processing and management has increased sharply. In large-scale energy storage power stations and electric vehicle networks, it is necessary to collect, transmit, and analyze a large amount of battery data in real time. Traditional local data processing methods have limited computing power and cannot quickly process this data, making it difficult to detect potential problems of the battery in a timely manner and respond. At the same time, due to the lack of an effective data integration and sharing mechanism, data between different devices and systems cannot be interconnected, forming "data islands", and the data value cannot be fully exploited, which limits the optimization of the overall performance of the battery. In addition, safety is an important issue that cannot be ignored in the application of lithium batteries. Abnormal conditions such as overcharging, over-discharging, overheating, and short-circuiting of the battery may cause serious accidents such as fires and even explosions. Existing anomaly detection and protection mechanisms have problems such as slow response speed and high false alarm rate, and cannot accurately judge and take effective measures at the first time when an abnormal situation occurs, posing a huge threat to the safety of personnel and property. For example, in some portable electronic devices, fire incidents caused by battery anomalies occur from time to time, causing serious damage to the lives and property of users. Summary of the Invention

[0006] The purpose of the present invention is to provide a cloud-based lithium battery management method and system to solve the problems proposed in the above background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solutions: A cloud-based lithium battery management method, the method includes: Construct a cloud collaborative management framework, including creating an efficient communication link between distributed battery data acquisition nodes and the cloud analysis platform; the cloud collaborative management framework integrates multi-source heterogeneous data and considers the historical degradation mode of the battery, and is used to obtain dynamic adjustment factors and degradation compensation coefficients; Perform multi-dimensional state joint estimation, including constructing a state estimation model, and obtaining real-time battery health status and remaining life prediction results by inputting multi-dimensional sensor data and environmental parameters into the state estimation model; the multi-dimensional sensor data includes single-cell voltage, current, temperature, and internal resistance; the environmental parameters include environmental temperature and humidity and charge and discharge cycle times; Set a dynamic equalization control strategy, including implementing the dynamic equalization control strategy according to the health status and remaining life prediction results output by the state estimation model; the dynamic equalization control strategy includes dynamically optimizing the equalization parameters to realize real-time compensation for the energy difference between single cells in the battery pack; Perform closed-loop feedback and model iteration; the closed-loop feedback includes periodically collecting the actual operation data of the battery pack and synchronizing it to the cloud analysis platform; The formula for the degradation compensation coefficient is:

[0008] Among them, is the degradation compensation coefficient of time t; is the reference degradation rate coefficient; is the weight of the k-th battery parameter; is the real-time measured value of the k-th parameter; is the reference threshold of the corresponding parameter; is the cumulative charge and discharge depth; is the depth attenuation correction factor.

[0009] Preferably, the construction of the cloud collaborative management framework includes: using Kubernetes to achieve elastic scheduling of cloud resources; adopting a time series database to store the battery data stream uploaded by multiple nodes.

[0010] Preferably, the multi-source heterogeneous data includes battery operation data, environmental monitoring data, and equipment maintenance logs; the calculation of the dynamic adjustment factor is based on the federated learning framework, and the global model is updated by aggregating the local model parameters of the edge nodes.

[0011] Preferably, the training process of the state estimation model includes: Data collection: Collect 10,000 groups of battery charge and discharge cycle data from distributed battery data collection nodes; the data includes single-cell voltage, current, temperature, internal resistance, and corresponding environmental temperature and humidity; Feature engineering: Perform sliding window segmentation on the charge and discharge cycle data, extract time-domain statistical features, frequency-domain energy features, and charge and discharge curve shape features; perform piecewise polynomial fitting on the temperature data to eliminate noise interference; Model construction: Adopt a dynamic graph neural network as the core architecture, and the dynamic graph neural network includes a time series embedding layer, a graph structure adaptive update layer, and a multi-task output layer; Federated training: Through the edge-cloud collaborative training mechanism, use local data to update model parameters and encrypt and upload them to the cloud for parameter aggregation to generate a global state estimation model.

[0012] Preferably, the graph structure adaptive update layer of the dynamic graph neural network dynamically adjusts the correlation weights between battery cells through an attention mechanism, and the formula is:

[0013] Among them, is the correlation weight between cell i and cell j at time t; and are the query matrix and the key matrix respectively; is the feature dimension.

[0014] Preferably, the dynamic balancing control strategy includes: constructing a balancing objective function based on mixed integer programming, with the optimization objective being to minimize the weighted sum of the voltage difference between monomers within a group and the balancing energy consumption, and the constraint conditions including the maximum allowable balancing current and the temperature safety threshold.

[0015] Preferably, the balancing objective function is:

[0016] Wherein, and are weight coefficients; is the voltage of the i-th monomer; is the average voltage of the group; is the balancing current; is the resistance of the balancing circuit.

[0017] Preferably, the closed-loop feedback and model iteration include: updating the state estimation model through an online incremental learning algorithm, specifically adopting a joint framework of Kalman filtering and variational autoencoder to fuse new data in real time and correct the model parameters.

[0018] Preferably, the cloud analysis platform is also deployed with an anomaly detection module, which identifies voltage dips, temperature anomalies, and internal resistance mutation events based on the isolation forest algorithm and dynamic threshold comparison, and triggers a hierarchical warning strategy.

[0019] Preferably, the present invention also includes a cloud-based lithium battery management system, including: Distributed data acquisition units, deployed at each battery node, for real-time acquisition of monomer voltage, current, temperature, and internal resistance data; Edge computing units, connected to the data acquisition units, for local data preprocessing, feature extraction, and model lightweight inference; Cloud analysis platform, including a microservice cluster, a federated learning coordinator, and a control strategy generation module, for integrating multi-node data, training a global model, and generating dynamic balancing instructions; Communication gateways, adopting a hybrid networking protocol of LoRa and 5G to achieve low-latency data transmission between the edge unit and the cloud platform; Balancing execution units, receiving the balancing instructions issued by the cloud, and adjusting the energy distribution between monomers through a multi-channel DC-DC converter and a MOSFET switch array.

[0020] Compared with the prior art, the beneficial effects of the present invention are: The present invention constructs a state estimation model. By inputting multi-dimensional sensor data and environmental parameters, it can accurately obtain the real-time battery health state and the prediction result of the remaining life. The multi-dimensional sensor data covers cell voltage, current, temperature and internal resistance, and the environmental parameters include environmental temperature and humidity and the number of charge and discharge cycles, comprehensively and synthetically reflecting the actual operating conditions of the battery. The training process of the state estimation model is rigorous. 10,000 groups of data are collected from distributed battery data acquisition nodes and optimized through multiple links such as data acquisition, feature engineering, model construction and federated training. Among them, feature engineering performs sliding window segmentation on the data, extracts various features and eliminates noise interference; the model construction uses a dynamic graph neural network, and its graph structure adaptive update layer dynamically adjusts the correlation weights between battery cells through an attention mechanism, improving the accuracy of the model. The accurate state assessment and life prediction provide a scientific basis for users to reasonably arrange the battery usage and maintenance plan, avoiding losses caused by unexpected battery failures.

[0021] Implement a dynamic equalization control strategy according to the output result of the state estimation model. Based on mixed integer programming, construct an equalization objective function. The optimization goal is to minimize the weighted sum of the voltage difference between cells within the group and the equalization energy consumption, while considering constraints such as the maximum allowable equalization current and the temperature safety threshold. This strategy can compensate for the energy difference between cells in the battery pack in real time, ensuring that each single cell maintains a relatively consistent state during the charge and discharge process. In the application scenario of electric vehicles, it can effectively reduce the inconsistency of single cells in the battery pack, improve the overall performance of the battery pack, extend the service life of the battery, and reduce the usage cost.

[0022] Construct a cloud collaborative management framework. Use Kubernetes to achieve elastic scheduling of cloud resources, and use a time series database to store the battery data stream uploaded by multiple nodes. Kubernetes can dynamically allocate computing resources according to the requirements of battery management tasks, improving resource utilization; the time series database is optimized for time series data, efficiently storing and querying battery data. At the same time, the cloud collaborative management framework integrates multi-source heterogeneous data, including battery operation data, environmental monitoring data and equipment maintenance logs, etc. Based on the federated learning framework, calculate the dynamic adjustment factor, aggregate the local model parameters of edge nodes to update the global model, make full use of the data of each node, and improve the generalization ability of the model and the management accuracy.

[0023] Through closed-loop feedback and model iteration, actual operation data of the battery pack is periodically collected and synchronized to the cloud analysis platform. An online incremental learning algorithm is adopted, and a joint framework of Kalman filtering and variational autoencoder is used to fuse new data in real time and correct model parameters. This enables the state estimation model to continuously adapt to the changes in the actual operating state of the battery. As time goes by and data accumulates, the accuracy and reliability of the model are continuously improved. The cloud analysis platform deploys an anomaly detection module. Based on the isolation forest algorithm and dynamic threshold comparison, it identifies events such as sudden voltage drops, abnormal temperatures, and sudden internal resistance mutations, and triggers a hierarchical alarm strategy to timely detect and handle potential problems in battery operation, ensuring the safe and stable operation of the battery. In the energy storage system, the anomaly detection module can issue an alarm at the first time when an abnormal situation occurs, enabling the operation and maintenance personnel to take measures in time to avoid accidents.

[0024] The present invention adopts an architecture design of a distributed data acquisition unit, an edge computing unit, a cloud analysis platform, a communication gateway, and an equalization execution unit. The distributed data acquisition unit collects battery data in real time. The edge computing unit performs local data preprocessing, feature extraction, and lightweight model inference to reduce the computing pressure on the cloud. The communication gateway uses a hybrid networking protocol of LoRa and 5G to achieve low-latency data transmission. The equalization execution unit receives cloud instructions to adjust the energy distribution between monomers. This system architecture is flexible and highly scalable, suitable for various lithium battery application scenarios, such as electric vehicles, energy storage power stations, portable electronic devices, etc., and can be customized and optimized according to different scenario requirements. Brief Description of the Drawings

[0025] Figure 1 It is the working principle diagram of the lithium battery management method described in the present invention; Figure 2 It is the flowchart of multi-source heterogeneous data processing and dynamic adjustment factor calculation; Figure 3 It is the flowchart of state estimation model training; Figure 4 It is the working principle diagram of anomaly detection and alarm in the cloud analysis platform. Detailed Embodiments

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0027] Please refer to Figures 1 - 4 , the present invention provides a technical solution: a cloud-based lithium battery management method, the method includes: Build a cloud collaborative management framework. Create an efficient communication link between distributed battery data acquisition nodes and the cloud analysis platform. This communication link can adopt various communication technologies, such as Bluetooth, Wi-Fi, ZigBee, etc. in wireless communication technologies, or Ethernet in wired communication technologies. Select a suitable communication method according to the actual application scenario and requirements to ensure stable and fast data transmission. This cloud collaborative management framework integrates multi-source heterogeneous data, including but not limited to battery operation data (such as charge and discharge current, voltage, etc.), environmental monitoring data (environmental temperature and humidity, etc.), and equipment maintenance logs. At the same time, considering the battery's historical degradation pattern, obtain dynamic adjustment factors and degradation compensation coefficients through specific algorithms. The calculation formula for the degradation compensation coefficient is:

[0028] where, is the degradation compensation coefficient at time t; is the reference degradation rate coefficient, and its value can be set according to a large amount of experimental data and battery characteristics, generally obtained through statistical analysis of the degradation test data of different types of batteries in a standard environment; is the weight of the k-th battery parameter, and these weights can be determined by multi-criteria decision-making methods such as the Analytic Hierarchy Process (AHP), comprehensively considering the influence degree of each parameter on battery degradation; is the real-time measured value of the k-th parameter, which is obtained by real-time acquisition of sensors distributed in various parts of the battery; is the reference threshold corresponding to the parameter, and these thresholds can be set according to the battery's technical specification and industry standards; is the cumulative charge and discharge depth, which is obtained by integrating the change in battery charge and discharge during the charging and discharging process; is the depth attenuation correction factor, and its value is also determined based on experimental data and battery characteristics, used to correct the influence of the cumulative charge and discharge depth on the degradation compensation coefficient.

[0029] Perform multi-dimensional state joint estimation. Construct a state estimation model and input multi-dimensional sensor data (single-cell voltage, current, temperature, and internal resistance) and environmental parameters (environmental temperature and humidity, and charge and discharge cycle times) into the model. Among them, the single-cell voltage, current, temperature, and internal resistance data can be collected through corresponding sensors. For example, a high-precision resistor voltage-divider type voltage sensor is used for the voltage sensor, a Hall current sensor can be selected for the current sensor, a thermistor type temperature sensor is used for the temperature sensor, and the internal resistance can be measured by the AC impedance method or the DC discharge method. The environmental temperature and humidity can be collected using a temperature and humidity sensor, and the charge and discharge cycle times can be obtained by recording the charge and discharge process times of the battery through the battery management system. Obtain the real-time battery health status and remaining life prediction results through this model, providing an important basis for subsequent battery management decisions.

[0030] Set a dynamic equalization control strategy. Implement this strategy according to the health status and remaining life prediction results output by the state estimation model, dynamically optimize the equalization parameters, and achieve real-time compensation for the energy difference between single cells in the battery pack. For example, when the state estimation model predicts that the remaining power of a certain single-cell battery is significantly lower than that of other single-cell batteries and may affect the performance and life of the entire battery pack, the dynamic equalization control strategy will be activated. By adjusting relevant parameters, the single-cell battery with higher power transfers energy to the single-cell battery with lower power to balance the energy difference between single cells in the battery pack.

[0031] Perform closed-loop feedback and model iteration. Closed-loop feedback includes periodically collecting the actual operation data of the battery pack and synchronizing it to the cloud analysis platform. For example, collect the operation data of the battery pack every certain period (such as 5 minutes), including the voltage, current, temperature, internal resistance, etc. of each single-cell battery, as well as environmental parameters such as environmental temperature and humidity and charge and discharge cycle times. These data are transmitted to the cloud analysis platform through a communication link for updating the state estimation model, enabling the model to better adapt to the actual operation of the battery and continuously improving the accuracy and effectiveness of battery management.

[0032] The present invention will be further described below in conjunction with Embodiments 1 to 5: Embodiment 1: This embodiment details the construction process of the cloud collaborative management framework, including the elastic scheduling of cloud resources and the storage method of battery data streams, aiming to improve the utilization efficiency of cloud resources, ensure the efficient and orderly storage of battery data streams uploaded by multiple nodes, and provide basic support for subsequent data analysis and processing.

[0033] When building a cloud collaborative management framework, Kubernetes is used to achieve elastic scheduling of cloud resources. Kubernetes is an open-source container orchestration platform that can automate the deployment, scaling, and management of containerized applications. In the present invention, various applications related to lithium battery management (such as data processing programs, model training programs, etc.) are encapsulated in the form of containers. Kubernetes, through its resource scheduler, intelligently allocates these containers to appropriate computing nodes according to the resource requirements of different applications (such as CPU, memory, storage, etc.) and the usage of current cloud resources. For example, when a large amount of new battery data is uploaded and needs to be processed in real time, Kubernetes can automatically detect the increased load of the data processing program and then dynamically allocate more computing resources to it, such as increasing the number of CPU cores or the memory capacity, to ensure that the data processing task can be completed in a timely manner. When the data processing task decreases, Kubernetes can reclaim these redundant resources and allocate them to other applications with requirements, thereby improving the overall utilization efficiency of cloud resources.

[0034] A time-series database is used to store the battery data streams uploaded by multiple nodes. The time-series database is specifically optimized for time-series data and is very suitable for storing data that changes over time generated during the operation of the battery. Taking InfluxDB as an example, it is a widely used open-source time-series database. In practical applications, each distributed battery data acquisition node marks the collected battery data (such as single-cell voltage, current, temperature, internal resistance, etc.) according to the timestamp and then transmits it to the InfluxDB database in the cloud through the network. InfluxDB uses specific data structures and storage algorithms to efficiently store and query these time-series data. For example, when it is necessary to query the voltage change of a certain battery pack within a certain period of time, InfluxDB can quickly locate the relevant data according to the timestamp and return the query result in an efficient manner, greatly improving the speed and efficiency of data query.

[0035] Embodiment 2: This embodiment mainly illustrates the specific types of multi-source heterogeneous data and the calculation method of the dynamic adjustment factor, enabling the present invention to make full use of multi-faceted data information, achieve more accurate model optimization through the federated learning framework, and improve the intelligence level of the battery management system.

[0036] Multi-source heterogeneous data includes battery operation data, environmental monitoring data, and equipment maintenance logs. Battery operation data covers various parameters during the charging and discharging processes of the battery, such as single-cell voltage, current, charging and discharging power, etc. These data directly reflect the working state of the battery. For example, the change in single-cell voltage can intuitively show the charging and discharging level of the battery. When the voltage is below a certain threshold, it indicates that the battery power is low and needs to be charged. Current data can reflect the charging and discharging rate of the battery. Excessive current may damage the battery, so it is necessary to monitor and control it in real time.

[0037] Environmental monitoring data mainly includes environmental temperature and humidity. Environmental temperature has a great impact on the performance and lifespan of the battery. Different batteries have the best working performance within different temperature ranges. For example, lithium-ion batteries have higher charging and discharging efficiency and relatively longer battery cycle life at an environmental temperature of 20°C - 30°C. When the environmental temperature is too high or too low, the internal resistance of the battery will increase, the charging and discharging capacity will decrease, and even safety problems may occur. Environmental humidity also affects the battery. Excessive humidity may cause internal short circuits in the battery, reducing the safety and reliability of the battery. Therefore, it is very necessary to monitor the environmental temperature and humidity in real time and use them as important reference data for battery management.

[0038] The equipment maintenance log records the maintenance situation of the battery equipment, including information such as maintenance time, maintenance content, and replaced parts. These information is of great value for analyzing the health status of the battery and predicting the remaining lifespan of the battery. For example, if it is found in the maintenance log that a certain battery pack frequently fails and has been repaired many times, it can be inferred that the health status of this battery pack may be poor and it is necessary to strengthen monitoring and management.

[0039] The calculation of the dynamic adjustment factor is based on the federated learning framework. Federated learning is a distributed machine learning technology that allows the global model to be updated by aggregating the local model parameters of edge nodes without sharing the original data. In the present invention, each edge computing unit serves as a participating node in federated learning. First, it uses the collected battery data locally for model training. For example, the edge computing unit uses local battery operation data, environmental monitoring data, etc. to train the local state estimation model and obtains local model parameters. Then, these local model parameters are uploaded to the cloud analysis platform in an encrypted manner. After the federated learning coordinator of the cloud analysis platform receives the local model parameters uploaded by each edge node, it aggregates these parameters according to a certain algorithm (such as the FedAvg algorithm) to generate global model parameters. These global model parameters are then sent to each edge node to update the local model of the edge node, thereby continuously optimizing the global model. In this way, both the privacy data of users are protected, and the data information of each edge node can be fully utilized to improve the accuracy and generalization ability of the model, and then a more reasonable dynamic adjustment factor can be obtained, providing more accurate support for battery management.

[0040] Embodiment 3: This embodiment details the training process of the state estimation model, including data collection, feature engineering, model construction, and federated training, etc., aiming to improve the accuracy and reliability of the state estimation model and provide a strong guarantee for accurately obtaining the real-time battery health status and remaining life prediction results.

[0041] In the data collection stage, 10,000 groups of battery charge and discharge cycle data are collected from distributed battery data collection nodes. These data cover multiple aspects such as single-cell voltage, current, temperature, internal resistance, and corresponding environmental temperature and humidity. In the actual collection process, to ensure the accuracy and reliability of the data, the collection equipment needs to be calibrated and maintained regularly. For example, for the voltage sensor, it is calibrated with a standard voltage source every once in a while (such as once a month) to ensure that the measured voltage value error is within the allowable range. At the same time, to obtain more comprehensive data, the collection time should cover the entire service life cycle of the battery, including different charge and discharge states, different environmental conditions, etc.

[0042] In the feature engineering stage, the charge and discharge cycle data is segmented using a sliding window. The size and step of the sliding window are selected according to the data characteristics and analysis requirements. For example, a sliding window with a size of 100 data points and a step of 10 data points is chosen. Through this sliding window segmentation, long-sequence time series data can be transformed into multiple short-sequence data blocks, facilitating subsequent feature extraction. For each data block, time-domain statistical features such as mean, variance, maximum value, and minimum value are extracted. The mean can reflect the average level of the data, while the variance can measure the degree of data dispersion. Taking the cell voltage data as an example, calculating its mean and variance within the sliding window can help understand the voltage stability during that time period. Meanwhile, frequency-domain energy features are extracted. The time-domain data is transformed into the frequency domain through Fourier transform, and then the energy distribution in different frequency bands is calculated. For example, for the battery charge and discharge current data, through frequency-domain analysis, it can be found that the energy changes in certain frequency bands are closely related to the battery's health status. In addition, the charge and discharge curve morphological features such as the slope of the charge curve and the inflection point of the discharge curve are also extracted. These morphological features can intuitively reflect the charge and discharge characteristics of the battery.

[0043] For temperature data, since it is easily affected by noise interference, a piecewise polynomial fitting method is used for processing. First, according to the changing trend of the temperature data, it is divided into multiple small segments. Then, for the data in each small segment, a polynomial function is used for fitting. For example, for a segment of temperature data, a quadratic polynomial is used for fitting, and the coefficients of the polynomial are determined through the least squares method , , , so that the fitting curve is as close as possible to the original temperature data, effectively eliminating noise interference.

[0044] In terms of model construction, a dynamic graph neural network (DGNN) is adopted as the core architecture. DGNN includes a time series embedding layer, a graph structure adaptive update layer, and a multi-task output layer. The role of the time series embedding layer is to convert time series data into a vector representation suitable for neural network processing. It learns the time features in the time series data and maps the data at each time step into a low-dimensional vector space, enabling the model to better capture the time dependence of the data.

[0045] The graph structure adaptive update layer dynamically adjusts the association weights between battery cells through an attention mechanism. The formula is:

[0046] where, is the association weight between cell i and cell j at time t; and are the query matrix and the key matrix respectively; is the feature dimension. In practical applications, this layer calculates the association weights between different cells through an attention mechanism based on the real-time status data of the battery cells. For example, when the voltage of a certain cell battery shows abnormal changes, the graph structure adaptive update layer will automatically increase the association weight between this cell and other cells, enabling the model to pay more attention to the impact of the status change of this cell on the entire battery pack.

[0047] The multi-task output layer then outputs the health status and remaining life prediction results of the battery based on the outputs of the previous layers. It predicts the health status and remaining life through different sub-network structures respectively. For example, for health status prediction, a fully connected neural network can be used, taking the feature vector output by the previous layer as input, and through the calculation of multiple layers of neurons, output a value representing the health status of the battery. For remaining life prediction, time series prediction models such as recurrent neural network (RNN) or long short-term memory network (LSTM) can be adopted, and based on the historical data and current status of the battery, predict the remaining life of the battery.

[0048] In the federated training stage, through the edge-cloud collaborative training mechanism, the model parameters are updated using local data and encrypted and uploaded to the cloud for parameter aggregation to generate a global state estimation model. The edge computing unit first trains the DGNN model locally using the collected data. For example, using the local battery charge and discharge cycle data and environmental data, the model is iteratively trained according to the training method described above to update the model parameters. Then, the updated model parameters are encrypted and uploaded to the cloud analysis platform through the communication link. After receiving the encrypted parameters uploaded by each edge node, the cloud analysis platform aggregates these parameters using a federated learning algorithm (such as the FedAvg algorithm). The basic idea of the FedAvg algorithm is to perform weighted averaging on the model parameters of each edge node according to certain weights to obtain the global model parameters. These global model parameters are then sent down to each edge node, and the edge node uses these global model parameters to update the local model, completing one round of federated training. Through multiple such federated trainings, the global state estimation model is continuously optimized to improve the accuracy and generalization ability of the model.

[0049] Embodiment 4: This embodiment focuses on the specific implementation details of the dynamic balancing control strategy. By constructing an accurate balancing objective function and setting reasonable constraint conditions, precise compensation for the energy difference between cells in the battery pack is achieved, thereby improving the overall performance and service life of the battery pack and ensuring that the battery can operate stably and efficiently under various working conditions.

[0050] The dynamic equilibrium control strategy constructs an equilibrium objective function based on mixed-integer programming, and takes the weighted sum of minimizing the voltage difference between cells within a group and the equilibrium energy consumption as the optimization objective. In actual battery application scenarios, due to differences in manufacturing processes, usage environments, etc. of battery cells, voltage inconsistencies will inevitably occur. Such voltage imbalance will not only reduce the overall performance of the battery pack, but also cause some battery cells to be overcharged and overdischarged, accelerating battery aging and thus shortening the service life of the battery pack. At the same time, when performing equilibrium control, the energy consumed during the equilibrium process cannot be ignored, and it is necessary to minimize the energy consumption as much as possible while ensuring the equilibrium effect.

[0051] The equilibrium objective function is set as:

[0052] Among them, and are weight coefficients, and their values are crucial, which determine the emphasis on voltage difference and equilibrium energy consumption during the equilibrium process. These two weight coefficients can be optimized and determined through a large amount of experimental data and actual application scenarios. For example, when testing a specific type of battery pack, set different and values, record the performance of the battery pack under different working conditions (such as charge and discharge efficiency, battery life, etc.), and then through data analysis and algorithm optimization, find the most suitable combination of weight coefficients for this battery pack.

[0053] represents the voltage of the i-th cell, which is obtained by real-time acquisition through voltage sensors distributed on each cell of the battery pack. These voltage sensors need to have high precision and high reliability to ensure that the collected voltage data can accurately reflect the true state of the single cell. For example, a voltage sensor with a precision of 0.01V can accurately measure the tiny changes in the voltage of a single cell.

[0054] represents the average voltage of the group, which is calculated by taking the arithmetic mean of the voltages of all cells in the battery pack. During the actual calculation process, a microprocessor or a dedicated calculation chip is used to quickly process the collected cell voltage data to obtain the average voltage value of the group.

[0055] is the equilibrium current, which is a key parameter for realizing energy transfer between cells. During the equilibrium process, by controlling the magnitude and direction of the equilibrium current, the transfer of electric charge from high-voltage cells to low-voltage cells is realized. The magnitude of the equilibrium current is restricted by various factors, such as the allowable charging current of the battery, the internal resistance of the battery, etc.

[0056] It is the balancing circuit resistance, and its value is related to the circuit design of the battery pack and the selected electronic components. When designing the battery management system, it is necessary to select appropriate resistance components according to actual requirements and cost considerations to ensure the stability and reliability of the balancing circuit.

[0057] Regarding the constraint conditions, they include the maximum allowable balancing current and the temperature safety threshold. The maximum allowable balancing current is set according to the characteristics of the battery and safety standards. If the balancing current is too large, it may cause irreversible damage to the battery, such as accelerating battery aging and triggering safety problems such as thermal runaway. For example, for a certain type of lithium-ion battery, its maximum allowable balancing current is 1A. When performing balancing control, it is necessary to ensure that the balancing current is always within this safe range.

[0058] The temperature safety threshold is also a crucial constraint condition. The battery generates heat during charging, discharging, and balancing processes. When the temperature is too high, it will not only affect the performance of the battery but also increase safety risks. Therefore, it is necessary to monitor the temperature of the battery in real time and set a reasonable temperature safety threshold. For example, when the battery temperature exceeds 45°C, automatically reduce the balancing current or suspend the balancing operation to prevent the battery from overheating. In practical applications, temperature sensors such as thermistors can be used to monitor the battery temperature in real time, and the temperature data is fed back to the controller of the battery management system, and the controller adjusts the balancing strategy according to the temperature situation.

[0059] Embodiment 5: This embodiment details the specific implementation methods of closed-loop feedback and model iteration, as well as the working principle and alarm strategy of the anomaly detection module. Through closed-loop feedback and model iteration, the state estimation model can adapt to the changes in the battery operating state in real time, continuously improving the accuracy of battery state monitoring and life prediction. The anomaly detection module can timely detect abnormal situations during the battery operation process and notify relevant personnel to take measures through a hierarchical alarm strategy to ensure the safe operation of the battery.

[0060] In the closed-loop feedback and model iteration link, an online incremental learning algorithm is used to update the state estimation model. Specifically, a joint framework of Kalman filter and variational autoencoder is utilized to achieve real-time fusion of new data and correction of model parameters. The Kalman filter is a commonly used optimal estimation method. Based on the state space model of the system, it can perform optimal estimation of the system state in the presence of noise. In the present invention, the Kalman filter is used to perform real-time estimation of the battery state (such as voltage, current, SOC, etc.). For example, according to the battery voltage and current data collected at the current moment, combined with the state estimation value at the previous moment, the optimal estimation value of the battery state at the current moment is calculated using the Kalman filter algorithm. At the same time, the Kalman filter can also estimate and compensate for measurement noise and system noise, improving the accuracy of state estimation.

[0061] The variational autoencoder (VAE) is a deep learning model that can learn the latent distribution of data and perform data reconstruction and generation. In this embodiment, the VAE is used for feature extraction and dimensionality reduction of battery data. The collected multi-dimensional battery data (such as single-cell voltage, current, temperature, internal resistance, etc.) is input into the VAE, and the VAE maps the high-dimensional data to a low-dimensional space by learning the latent features of the data, extracting the features that are most valuable for battery state monitoring and life prediction. These features can not only reduce the dimension of the data, reduce the computational complexity, but also improve the generalization ability of the model.

[0062] In practical applications, the Kalman filter works jointly with the variational autoencoder. When new battery data is collected, the data is first input into the VAE for feature extraction to obtain low-dimensional feature vectors. Then, these feature vectors are used as the observations of the Kalman filter, combined with the state prediction values of the Kalman filter, to update and estimate the state of the battery. At the same time, according to the new data and the estimation results, the model parameters of the Kalman filter and the VAE are adjusted and optimized to achieve online incremental learning of the model. In this way, the state estimation model can continuously adapt to the changes in the battery operating state and improve the accuracy of predicting the battery health state and remaining life.

[0063] The anomaly detection module deployed on the cloud analysis platform is based on the comparison between the isolation forest algorithm and the dynamic threshold to achieve the identification of abnormal events during the operation of the battery. The isolation forest algorithm is an unsupervised learning algorithm that models the data by constructing isolation trees. Under normal circumstances, data points will be located at the bottom layer of the isolation tree, while abnormal data points will be located at the upper layer of the isolation tree. When performing anomaly detection on battery data, the collected battery data (such as single-cell voltage, current, temperature, internal resistance, etc.) is input into the isolation forest model, and the model will calculate the anomaly score of each data point according to the position of the data point in the isolation tree. The higher the anomaly score, the more likely the data point is an abnormal point.

[0064] The dynamic threshold is adjusted dynamically according to the historical operating data and the current operating state of the battery. For example, a dynamic voltage threshold is calculated based on the voltage fluctuation range and change trend of the battery over a period of time. When the anomaly score of the single-cell voltage exceeds the dynamic threshold, it is determined that a voltage dip event has occurred. For temperature anomalies and internal resistance mutation events, similar methods are used for detection. By combining the isolation forest algorithm and the dynamic threshold comparison, abnormal situations during the operation of the battery can be identified more accurately, avoiding false positives or false negatives caused by unreasonable threshold settings.

[0065] When the anomaly detection module identifies events such as voltage sags, temperature anomalies, and internal resistance mutations, it triggers a hierarchical alarm strategy. The hierarchical alarm strategy is divided into different levels according to the severity of the anomaly events. For example, a level-1 alarm indicates a severe anomaly and immediate measures need to be taken; a level-2 alarm indicates a relatively severe anomaly and inspection and handling should be carried out as soon as possible; a level-3 alarm indicates a general anomaly and can be handled at an appropriate time. For example, when it is detected that the battery temperature exceeds the set high-temperature threshold and continues to rise, a level-1 alarm is triggered, and relevant operation and maintenance personnel and management personnel are notified by means such as text messages, emails, or system pop-ups, informing them that there is a severe temperature anomaly in the battery and the use of the battery needs to be stopped immediately, and cooling treatment and fault troubleshooting should be carried out. For abnormal situations such as voltage sags and internal resistance mutations, corresponding-level alarms will also be triggered according to their severity to ensure that problems in the battery operation process are discovered and handled in a timely manner, and the safe operation of the battery is guaranteed.

[0066] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0067] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cloud-based lithium battery management method, characterized in that, Including: Construct a cloud collaborative management framework, including creating an efficient communication link between distributed battery data acquisition nodes and the cloud analysis platform; The cloud collaborative management framework integrates multi-source heterogeneous data and considers the historical degradation mode of the battery to obtain dynamic adjustment factors and degradation compensation coefficients; Perform multi-dimensional state joint estimation, including constructing a state estimation model, and obtaining real-time battery health status and remaining life prediction results by inputting multi-dimensional sensor data and environmental parameters into the state estimation model; the multi-dimensional sensor data includes single-cell voltage, current, temperature, and internal resistance; the environmental parameters include environmental temperature and humidity and charge-discharge cycle times; Set a dynamic equalization control strategy, including implementing the dynamic equalization control strategy according to the health status and remaining life prediction results output by the state estimation model; the dynamic equalization control strategy includes dynamically optimizing the equalization parameters to achieve real-time compensation for the energy difference between single cells in the battery pack; Perform closed-loop feedback and model iteration; the closed-loop feedback includes periodically collecting the actual operation data of the battery pack and synchronizing it to the cloud analysis platform; The formula for the degradation compensation coefficient is: Among them, is the degradation compensation coefficient of time t; is the reference degradation rate coefficient; is the weight of the k-th battery parameter; is the real-time measured value of the k-th parameter; is the reference threshold of the corresponding parameter; is the cumulative charge and discharge depth; is the depth attenuation correction factor.

2. The method for managing a lithium battery based on the cloud according to claim 1, characterized in that, The construction of the cloud collaborative management framework includes: using Kubernetes to achieve elastic scheduling of cloud resources; adopting a time-series database to store the battery data stream uploaded by multiple nodes.

3. A cloud-based lithium battery management method according to claim 1, characterized in that, The multi-source heterogeneous data includes battery operation data, environmental monitoring data, and equipment maintenance logs; the calculation of the dynamic adjustment factor is based on a federated learning framework, and the global model is updated by aggregating the local model parameters of edge nodes.

4. A cloud-based lithium battery management method according to claim 1, characterized in that, The training process of the state estimation model includes: Data collection: Collect 10,000 groups of battery charge-discharge cycle data from distributed battery data acquisition nodes; the data includes single-cell voltage, current, temperature, internal resistance, and corresponding environmental temperature and humidity; Feature engineering: Perform sliding window segmentation on the charge-discharge cycle data, extract time-domain statistical features, frequency-domain energy features, and charge-discharge curve shape features; perform piecewise polynomial fitting on the temperature data to eliminate noise interference; Model construction: Adopt a dynamic graph neural network as the core architecture, and the dynamic graph neural network DGNN includes a time-series embedding layer, a graph structure adaptive update layer, and a multi-task output layer; Federated training: Through an edge-cloud collaborative training mechanism, use local data to update model parameters and encrypt and upload them to the cloud for parameter aggregation to generate a global state estimation model.

5. A cloud-based lithium battery management method according to claim 4, characterized in that, The graph structure adaptive update layer of the dynamic graph neural network dynamically adjusts the correlation weights between battery cells through an attention mechanism, and the formula is: Among them, is the correlation weight between monomer i and monomer j at time t; and are the query matrix and the key matrix respectively; is the feature dimension.

6. The method for managing a lithium battery based on cloud according to claim 1, wherein, The dynamic equalization control strategy includes: constructing an equalization objective function based on mixed integer programming, with the optimization objective being to minimize the weighted sum of the voltage difference between single cells in the group and the equalization energy consumption, and the constraint conditions including the maximum allowable equalization current and the temperature safety threshold.

7. A cloud-based lithium battery management method according to claim 6, characterized in that The equalization objective function is: Among them, and are weighting coefficients; is the voltage of the i-th monomer; is the average voltage of the group; is the balancing current; is the resistance of the balancing circuit.

8. A cloud-based lithium battery management method according to claim 1, characterized in that, The closed-loop feedback and model iteration include: updating the state estimation model through an online incremental learning algorithm, specifically adopting a joint framework of Kalman filter and variational autoencoder to fuse new data in real time and correct model parameters.

9. A cloud-based lithium battery management method according to claim 1, characterized in that The cloud analysis platform also deploys an anomaly detection module. The anomaly detection module compares with a dynamic threshold based on the isolation forest algorithm to identify voltage dips, temperature anomalies, and internal resistance mutation events, and triggers a hierarchical alarm strategy.

10. A cloud-based lithium battery management system, characterized in that, Including: Distributed data acquisition units, deployed at each battery node, for real-time acquisition of single-cell voltage, current, temperature, and internal resistance data; Edge computing units, connected to the data acquisition units, for local data preprocessing, feature extraction, and model lightweight inference; Cloud analysis platforms, including microservice clusters, federated learning coordinators, and control strategy generation modules, for integrating multi-node data, training global models, and generating dynamic balancing instructions; Communication gateways, using a hybrid networking protocol of LoRa and 5G, to achieve low-latency data transmission between edge units and cloud platforms; Balancing execution units, receiving the balancing instructions sent from the cloud, and adjusting the energy distribution between single cells through a multi-channel DC-DC converter and a MOSFET switch array.

Images