Remote monitoring management method and system for new energy automobile power battery

Abstract

The invention relates to a remote monitoring management method and system for a new energy automobile power battery. The method comprises the following steps: constructing a layered monitoring network structure according to a vehicle-mounted terminal layer, a regional monitoring center layer and a central management platform layer, and collecting a hierarchical data transmission sequence of each monitoring region in a power battery; performing battery performance feature extraction and reconstruction verification on the hierarchical data transmission sequence of each monitoring area by using a feature enhanced automatic encoder to obtain a battery health prediction model of each monitoring area; performing distributed training in each monitoring area based on the battery health prediction model, and determining a model fusion weight according to data quality and prediction accuracy to obtain a global battery monitoring model set; and acquiring a cooperative monitoring data stream of the power battery, and performing multi-dimensional analysis on the cooperative monitoring data stream based on the global battery monitoring model set to obtain a battery health level assessment result. According to the invention, the accuracy and reliability of the health level evaluation result of the power battery are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery remote monitoring, and in particular to a remote monitoring management method and system for a power battery of a new energy vehicle. Background Art

[0002] With the rapid development of the new energy vehicle industry, the safety, reliability and service life of power batteries, as core components, directly affect the overall performance and user experience of new energy vehicles. In order to timely grasp the health status of power batteries, remote monitoring technology is widely used in the field of power battery management of new energy vehicles. However, due to the wide distribution range and complex use environment of new energy vehicles, traditional centralized monitoring methods are difficult to adapt to large-scale and multi-scenario monitoring needs, resulting in unstable monitoring data quality and insufficient model generalization capabilities.

[0003] Current power battery remote monitoring systems generally have problems with low data transmission efficiency and uneven allocation of computing resources. Especially when network conditions are unstable, real-time transmission of key parameters is difficult to guarantee, affecting the timeliness and accuracy of monitoring. At the same time, battery performance varies significantly under different usage environments, and a single monitoring model is difficult to accurately describe the health status of the battery, resulting in inaccurate warnings and high false alarm rates. Summary of the invention

[0004] The main purpose of the present invention is to provide a remote monitoring and management method and system for new energy vehicle power batteries. The present invention improves the accuracy and reliability of the health level assessment results of the power batteries.

[0005] To achieve the above object, the present invention provides a remote monitoring and management method for a new energy vehicle power battery, comprising the following steps: A hierarchical monitoring network structure is constructed based on the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and processed hierarchically to obtain a hierarchical data transmission sequence for each monitoring area. The feature-enhanced autoencoder is used to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area to obtain the battery health prediction model of each monitoring area. Based on the battery health prediction model, distributed training is performed in each monitoring area, and the model fusion weight is determined according to the data quality and prediction accuracy to obtain a global battery monitoring model set; The collaborative monitoring data stream of the power battery is obtained, and the collaborative monitoring data stream is analyzed in multiple dimensions based on the global battery monitoring model set to obtain a battery health level assessment result.

[0006] The present invention also provides a remote monitoring and management device for a new energy vehicle power battery, comprising: A construction module is used to construct a hierarchical monitoring network structure according to the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and hierarchically processed to obtain a hierarchical data transmission sequence of each monitoring area; A feature extraction module is used to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area using a feature-enhanced automatic encoder to obtain a battery health prediction model for each monitoring area; A model fusion module, used to perform distributed training in each monitoring area based on the battery health prediction model, and determine the model fusion weight according to data quality and prediction accuracy to obtain a global battery monitoring model set; The analysis module is used to obtain the collaborative monitoring data stream of the power battery, and perform multi-dimensional analysis on the collaborative monitoring data stream based on the global battery monitoring model set to obtain a battery health level assessment result.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention constructs a hierarchical monitoring network structure and performs hierarchical data transmission, which reduces network bandwidth occupancy and improves data transmission efficiency while ensuring real-time transmission of key parameters; a feature-enhanced autoencoder is used to extract battery performance features, combined with residual connections and multi-head attention mechanisms, to enhance the ability to characterize the correlation between battery performance parameters; based on distributed training and model fusion strategies, the data resources of each monitoring area are fully utilized, and the generalization ability and prediction accuracy of the model are significantly improved; through dynamic allocation of computing resources and equipment status monitoring, balanced optimization of system load is achieved, and the reliability and stability of the monitoring system are improved; a multi-dimensional analysis method is used to evaluate the battery health status, comprehensively considering multiple features such as capacity attenuation and internal resistance change, to improve the accuracy and reliability of the evaluation results; a complete battery health level assessment system is established, and through multi-model integration and weighted fusion, accurate assessment and timely warning of battery health status are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1It is a schematic diagram of the steps of a remote monitoring management method for a power battery of a new energy vehicle in one embodiment of the present invention; Figure 2 This is a structural block diagram of a remote monitoring and management device for a new energy vehicle power battery in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 This embodiment provides a remote monitoring and management method for a new energy vehicle power battery, comprising the following steps: S1, constructing a hierarchical monitoring network structure based on the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, collecting the initial data set of each monitoring area in the power battery and performing hierarchical processing to obtain a hierarchical data transmission sequence of each monitoring area; Among them, the data acquisition module and the data cache module are deployed at the vehicle terminal layer. By setting the sampling time interval of the data acquisition module, the raw data of each monitoring area in the power battery are regularly sampled to obtain the raw sampling data, which covers the core performance parameters of the power battery, including voltage, current, temperature and SOC (state of charge). The raw sampling data of each monitoring area in the power battery is divided into regions according to the geographical location information. The collected data are classified and processed according to different geographical areas to form divided data, which are allocated to the corresponding regional monitoring center layer to form regional data groups. This step significantly reduces data redundancy and improves the relevance and availability of data in the region through a clear geographical positioning mechanism. Edge computing servers and data synchronization modules are configured at each monitoring area layer. The edge computing server preprocesses the regional data grouping. The preprocessing process includes data cleaning, denoising and feature extraction to obtain regional preprocessing data. The data synchronization module ensures the real-time and consistency of the data in the region, so that even in a complex network environment, the data can respond quickly through local processing to improve the robustness of the entire system. The regional preprocessing data is transmitted to the central management platform layer. At the central management platform level, a data backup module is configured to design a reasonable data storage solution, which ensures that data will not be lost due to emergencies during transmission. In this process, a two-way communication link is established according to the data storage solution to realize the channel for issuing control instructions and the channel for reporting status information. The implementation of the communication solution enables efficient transmission and feedback of data between different levels. In order to enhance the stability and adaptability of the system, the response mechanism in the event of network disconnection is considered. Based on the network communication solution, local caching of data during network disconnection and data retransmission processing after network recovery are realized to form the initial data set of each monitoring area in the power battery. The initial data set of each monitoring area is processed in a hierarchical manner. According to the characteristics, source and importance of the data, it is hierarchically organized to form a hierarchical data transmission sequence.

[0014] The data type of the initial data set of each monitoring area is identified, and different types of data in the power battery are divided into key parameter groups and auxiliary parameter groups according to their importance and purpose. The key parameter group includes voltage, current and power data, which directly affect the performance and operation safety of the power battery, while temperature, position and environmental data are divided into auxiliary parameter groups. A higher sampling frequency is set for the voltage and current data in the key parameter group to ensure that the key data can reflect the real-time status of the power battery with higher accuracy. For the temperature and position data in the auxiliary parameter group, downsampling is adopted to reduce the occupation of network bandwidth by non-critical data and reduce the processing pressure of the system to a certain extent. The multi-level sampling scheme formulated based on the importance of the data type optimizes the overall data transmission efficiency while ensuring the integrity of the key data. Based on the multi-level sampling scheme, the real-time bandwidth of the system data is monitored, and the network delay is calculated to determine the current network load status. The network load index dynamically reflects the operating status and transmission capacity of the system. According to these indicators, a high-priority transmission channel is set for the key parameter group to ensure that the key data has a higher bandwidth and lower delay during transmission; at the same time, a low-priority transmission channel is set for the auxiliary parameter group so that it can transfer resources to the key data when the bandwidth is limited. This transmission channel allocation scheme effectively improves the system's ability to respond to emergencies and gives priority to the transmission of critical data when resources are limited. The data is differentially encoded and compressed according to the transmission channel allocation scheme, and the data compression sequence is optimized so that it occupies less network resources during transmission. The use of differential coding can reduce data redundancy, especially in scenarios where the monitoring data has a strong correlation, while compression processing reduces the amount of data and optimizes transmission efficiency and cost. At the same time, the data compression sequence is monitored in real time for transmission success rate and packet loss rate. By monitoring the transmission parameters in real time, the transmission parameters are dynamically adjusted when the packet loss rate is high or the success rate is low, such as reallocating bandwidth resources or adjusting the compression ratio, thereby ensuring the stability and reliability of data transmission and ultimately obtaining a hierarchical data transmission sequence.

[0015] S2, using feature-enhanced autoencoders to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area to obtain a battery health prediction model for each monitoring area; Specifically, the hierarchical data transmission sequence is segmented according to the charging and discharging cycle. By detecting the changes in key indicators of the battery at different charging and discharging stages, the periodic characteristics of the battery performance are captured. At the same time, the segmented data is normalized so that all data are standardized to the same range, reducing the scale differences between feature values, improving the training effect of the model, and forming a standardized data sequence. A feature-enhanced autoencoder network is constructed based on the standardized data sequence. The core architecture of the autoencoder network includes an encoder module and a decoder module. The encoder module performs dimensionality reduction mapping on the input data through 4 convolutional layers. Each convolutional layer uses the ReLU activation function to introduce nonlinear characteristics. At the same time, the batch normalization layer is combined to accelerate the convergence and stable training of the model. The decoder module completes the feature reconstruction through 4 deconvolutional layers. Each deconvolutional layer uses the Tanh activation function to ensure that the numerical range of the output data is consistent with the input data, and the batch normalization layer is used to improve the accuracy of the reconstruction. In order to improve the feature extraction ability of the model, a residual connection module is inserted between the convolutional layer and the deconvolutional layer. The module contains three residual units, each of which consists of two convolutional layers and a short-circuit connection. The short-circuit connection allows the gradient to be directly propagated back to the shallow network, thereby alleviating the gradient vanishing problem in the deep network. In order to capture the long-range dependency characteristics in the hierarchical data, a multi-head attention layer is added between the residual units. With the help of the attention mechanism, different weights are assigned to different parts of the input sequence, so that the model can more accurately focus on the features that have an important impact on the battery performance. These designs together constitute a feature-enhanced network structure. The standardized data sequence is input into the feature enhancement network. The encoder module performs dimensionality reduction mapping on the data to extract key low-dimensional feature vectors from the high-dimensional space. These low-dimensional features can condense the main information of the input data and filter out redundant noise. The decoder module reconstructs the low-dimensional feature vectors to generate a reconstructed data sequence. On this basis, the reconstruction error between the reconstructed data sequence and the standardized data sequence is calculated to evaluate the reconstruction ability of the model. This is used as an optimization indicator to back-propagate the parameters of the feature enhancement network structure so that the network can better adapt to the feature distribution of the input data and obtain the optimized autoencoder network. The capacity decay prediction layer is constructed based on the low-dimensional feature vector. The prediction layer consists of three fully connected layers. Each fully connected layer uses the Sigmoid activation function to ensure that the output value is within a specific range. At the same time, a Dropout layer is added between the fully connected layers to randomly block the connections of some neurons to reduce the risk of overfitting and improve the generalization ability of the model. The capacity decay prediction layer uses the low-dimensional features extracted by the autoencoder to predict the capacity decay trend and remaining service life of the power battery, and finally builds a battery health prediction model for each monitoring area.

[0016] S3, based on the battery health prediction model, distributed training is performed in each monitoring area, and the model fusion weight is determined according to the data quality and prediction accuracy to obtain the global battery monitoring model set; It should be noted that the parameters of the battery health prediction model are initialized. By setting the initial parameter values ​​to globally consistent standardized values, the model's initial state in each monitoring area is consistent, avoiding training deviations caused by differences in initial conditions. The initialized model is distributed to each monitoring area, and combined with the region's specific hardware resources, network bandwidth, and data characteristics, an independent model training task is assigned to each area to form a regional training plan. The quality of the training data in each monitoring area is evaluated based on the regional training plan. By designing a data quality scoring mechanism, the regional training data is weighted from three dimensions: data integrity, data consistency, and data timeliness to form a comprehensive scoring index. Data integrity reflects the coverage of the sampled data to ensure that key data is not lost; data consistency evaluates the coordination of data in time and space to avoid unreliability caused by abnormal distribution; and data timeliness ensures the real-time and effectiveness of the data, so that the training data can truly reflect the dynamic changes in battery performance. These scores are weighted to form regional data weights. After the data quality scoring is completed, the training process of each monitoring area is calculated in parallel. The model parameters are gradually updated using the stochastic gradient descent method, and the computing resources of each region are coordinated through a distributed computing framework to improve training efficiency. During the training process, the prediction accuracy of the model is monitored in real time and used as an important indicator for evaluating the performance of the regional model. The prediction accuracy reflects the adaptability and accuracy of the model on the current regional data. According to the regional data weights and regional model performance indicators, a model fusion weight matrix is ​​constructed. The model parameters of each monitoring area are fused by weighted average, and the data and models that perform better during the training process are given higher weights to form the fusion model parameters. The fusion model parameters are globally verified to evaluate the performance and adaptability of the fusion model. The global verification process generates parameter verification results by testing the model on data sets that are not involved in the training. These verification results are used to check the generalization ability of the model on data from different regions and to evaluate whether the model can accurately predict the health status of the battery. Based on the verification, the model structure and parameter distribution are optimized and adjusted, such as reallocating weights or adjusting the model architecture for poor performance in specific regions to improve model performance. A global battery monitoring model set is formed.

[0017] S4, obtain the collaborative monitoring data stream of the power battery, and perform multi-dimensional analysis on the collaborative monitoring data stream based on the global battery monitoring model set to obtain the battery health level assessment result.

[0018] Specifically, the computing load of the global battery monitoring model set is divided into time windows. By allocating computing tasks to fixed time periods, the dynamic changes in computing resource usage can be effectively captured. Based on the time window, the computing resource usage of each monitoring area is statistically analyzed to generate a resource occupancy state matrix. Each row and column of the matrix corresponds to a specific time period and resource type, such as processor occupancy, memory usage, and network bandwidth. The resource occupancy state matrix is ​​subjected to singular value decomposition to map complex multidimensional data into a low-dimensional space to obtain a feature vector that characterizes the core pattern of resource occupancy in each time window. Based on the feature vector, the computing tasks are prioritized and a task scheduling sequence is generated to ensure that key tasks can obtain resource allocation first. Under the guidance of the task scheduling sequence, a computing resource balanced allocation model is established, focusing on the three core resources of processor occupancy, memory usage, and network bandwidth, and a dynamic allocation strategy is formulated. By real-time monitoring of the resource usage status of each monitoring area and combining the importance and urgency of the task, a load balancing solution is generated. The role of the load balancing solution is to reduce the performance bottleneck caused by computing overload in some areas while ensuring efficient resource utilization. At the same time, the fault features of the running status data of each monitoring device are extracted. By collecting the core indicators of the equipment during operation, such as temperature, running time and data throughput, the constructed multi-layer perceptron network is used to classify and judge the equipment status, and the equipment health index is obtained to reflect the current status and potential fault risks of the monitoring equipment. The backup switching strategy of the monitoring equipment is designed according to the equipment health index. When the health of some devices is lower than the set threshold, switch to the backup device to ensure the continuity of the monitoring task. At the same time, the data flow during the switching process is optimized in combination with the load balancing solution to realize the device fault tolerance mechanism. The device fault tolerance mechanism is integrated with the load balancing solution to realize the unified scheduling of the collection, transmission and processing of monitoring data, and generate efficient and reliable collaborative monitoring data streams. Based on the global battery monitoring model set, the collaborative monitoring data stream is analyzed in multiple dimensions, and the time dynamic characteristics, spatial distribution characteristics and performance correlation of the data are analyzed to generate the battery health level evaluation results. The evaluation results reflect the current performance and potential risks of the battery by dividing the battery health status into multiple levels.

[0019] The collaborative monitoring data stream is input into multiple encoder modules in the global battery monitoring model set. Each encoder module performs dimensionality reduction mapping on the input data through the designed four convolutional layers. The convolutional layer uses the ReLU activation function to introduce nonlinear features and enhance the data's expressiveness in high-dimensional space. The main features of key monitoring data such as voltage, current and temperature are extracted, and the original complex high-dimensional data is converted into a structured feature mapping matrix, thereby reducing redundant information and improving the efficiency of subsequent processing. The feature mapping matrix of each encoder module is input into the residual connection module for feature enhancement. The residual connection module uses three residual units and a multi-head attention layer to deeply process the features. The residual unit alleviates the gradient vanishing problem through a short-circuit connection mechanism, and extracts the complex correlation between features by superimposing multiple convolution operations. The introduction of the multi-head attention layer enables the model to dynamically focus on the parts of the input features that are important to battery performance, such as specific voltage fluctuations or temperature anomalies, thereby improving the effectiveness and relevance of feature expression. Through this step, multiple enhanced feature vectors are obtained. The enhanced feature vector is input into multiple decoder modules in the global battery monitoring model set. The decoder module maps the low-dimensional features back to the high-dimensional space through 4 deconvolution layers, and uses the Tanh activation function to ensure that the numerical range of the reconstructed data is consistent with the input data. The deconvolution process is designed to generate a reconstructed data sequence. The mean square error of the reconstructed data sequence and the collaborative monitoring data stream is calculated, and the reconstruction error is normalized in combination with the batch normalization layer to generate multiple anomaly detection indicators to reflect potential anomalies in the data, such as battery voltage or temperature in certain areas exceeding the expected range. At the same time, the enhanced feature vector is input into the capacity decay prediction layer in the global battery monitoring model set, and the capacity decay trend of the battery is modeled through 3 fully connected layers. The Sigmoid activation function is used for each step of the fully connected layer to keep the output decay prediction value within a reasonable range. In order to avoid overfitting of the model, a Dropout layer is inserted between each fully connected layer to randomly block some neuron connections to enhance the generalization ability of the model. The output of the capacity decay prediction layer is a set of key prediction values ​​reflecting the future health status of the battery, which can accurately describe the long-term trend of battery performance. An evaluation matrix is ​​constructed based on multiple anomaly detection indicators and multiple attenuation prediction values, and the analysis results of all dimensions are integrated through weighted fusion operations. Combining the short-term monitoring capability of anomaly detection indicators with the long-term trend evaluation capability of capacity attenuation prediction values, each indicator is comprehensively scored according to its importance and weight to form an overall judgment on the battery health status. The comprehensive scoring results are parsed into a battery health level assessment, which is divided into multiple levels to reflect the current performance and potential risks of the battery.

[0020] In one example, a hierarchical monitoring network structure is constructed based on the vehicle terminal layer, the regional monitoring center layer, and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and hierarchically processed to obtain a hierarchical data transmission sequence for each monitoring area, including: A data acquisition module and a data cache module are configured at each vehicle terminal layer, and a sampling time interval is set for the data acquisition module to obtain the original sampling data of each monitoring area in the power battery; The original sampling data of each monitoring area in the power battery is divided into regions according to the geographical location information to obtain divided data, and the divided data is allocated to the corresponding regional monitoring center layer to obtain regional data grouping; Configure edge computing servers and data synchronization modules at each monitoring area layer, and pre-process regional data groups based on the edge computing servers to obtain regional pre-processed data; The regional pre-processed data is transmitted to the central management platform layer, and a data backup module is configured at the central management platform layer to obtain a data storage solution; According to the data storage scheme, two-way communication links are configured for data at each level, and control instruction issuing channels and status information reporting channels are established to obtain a network communication scheme. At the same time, based on the network communication scheme, the original sampling data of each monitoring area in the power battery is cached when the network is disconnected and the data is retransmitted after the network is restored to obtain the initial data set of each monitoring area in the power battery; The initial data set of each monitoring area is processed hierarchically to obtain a hierarchical data transmission sequence.

[0021] In this example, a data acquisition module and a data cache module are configured in each vehicle terminal layer. The data acquisition module collects the operating parameters of the power battery in real time, such as voltage , Current ,temperature And other performance parameters. Sampling time interval By optimizing the formula Settings, where is the sampling frequency, which is adjusted according to monitoring requirements For example, a higher frequency is used when the battery is running at high load to capture subtle performance changes. The collected raw data is temporarily stored in the cache module to ensure data integrity in the event of network disconnection or delay. These raw sampled data are based on the geographic location information of the vehicle to which the power battery belongs. Regional division is carried out, and the geographical location information is obtained through the GPS module, and the coordinates Represents longitude, latitude and altitude respectively. The divided data forms a data set according to regional characteristics ,in Indicates the region number. For example, the data set for region 1 Including voltage, temperature and other parameter data from all vehicles in the area. Assigned to the corresponding regional monitoring center layer. The regional monitoring center layer generates regional data groups by summarizing and storing these data . Configure edge computing servers and data synchronization modules at each monitoring area layer. The edge computing server groups regional data Preprocessing is performed, including data cleaning, outlier removal, and feature extraction. For example, the voltage data is smoothed by the mean filter method: ,in is the smoothed voltage value, is the size of the sliding window. The edge computing server maintains data consistency with the vehicle terminal layer through the synchronization module to ensure that there will be no loss due to time delay or breakpoints during data transmission. Transmit to the central management platform layer. The central management platform layer configures the data backup module to generate a data storage solution. The optimization of the data storage solution is based on the multi-objective optimization formula: ,in represents the storage cost, Represents the data retrieval time. These parameters are weighed according to the needs of specific application scenarios to design a reasonable storage architecture. In order to ensure the reliable transmission of data between various levels, the data communication link is bidirectionally configured to establish a control instruction issuance channel and a status information reporting channel. For example, the control channel transmits management instructions through the TCP / IP protocol, while the status channel implements lightweight status information upload through the UDP protocol. Based on this network communication scheme, the data caching mechanism is started in the event of a network outage, the original sampling data is stored in the local cache module, and the data is supplemented through the supplementary transmission mechanism after the network is restored to generate a complete initial data set for the power battery. For the initial data set Perform hierarchical processing to build a hierarchical data transmission sequence. Categorize data according to its importance and usage scenario, for example, divide voltage and current data into key parameter groups , divide the temperature and environmental data into auxiliary parameter groups A multi-level sampling strategy is used to optimize the sampling frequency of different data groups, for example, the sampling frequency of the key parameter group Set to a higher value, and the sampling frequency of the auxiliary parameter group Set to a lower value to meet Apply compression coding methods, such as differential coding, to hierarchical data ,in is the differential encoding value, and The time and The generated hierarchical data transmission sequence Provides efficient and hierarchical data management.

[0022] In one example, the initial data sets of each monitoring area are processed hierarchically to obtain a hierarchical data transmission sequence, including: Identify the data type of the initial data set of each monitoring area, divide the voltage, current, and power data into key parameter groups, and divide the temperature, location, and environmental data into auxiliary parameter groups; The sampling frequency of the voltage and current data in the key parameter group is set, and the temperature and position data in the auxiliary parameter group are downsampled to obtain a multi-level sampling scheme; Based on the multi-level sampling scheme, real-time data bandwidth monitoring and network delay calculation are performed to obtain network load indicators; Based on the network load index, a high priority transmission channel is set for the key parameter group, and a low priority transmission channel is set for the auxiliary parameter group, so as to obtain a transmission channel allocation scheme; The data is differentially encoded and compressed according to the transmission channel allocation scheme to obtain a data compression sequence. The transmission success rate and packet loss rate of the data compression sequence are monitored in real time and the transmission parameters are dynamically adjusted to obtain a hierarchical data transmission sequence.

[0023] In this example, different data types in the initial data set are accurately identified and classified. Voltage , Current and power It is the core indicator of power battery performance and is divided into key parameter groups , directly reflects the battery's operating status and potential failure risk. ,Location And environmental factors data (such as humidity, vibration, etc.) are divided into auxiliary parameter groups , these data are used for environmental adaptability analysis and operating condition optimization, and their importance is relatively low. After completing the data classification, a multi-level sampling scheme is designed for different parameter groups to optimize the data transmission efficiency and resource utilization of the monitoring system. Voltage and current Data, sampling frequency Set it higher to ensure that the data can accurately capture the dynamic changes of the battery. satisfy . For the auxiliary parameter group Temperature and location data , sampling frequency Set it low to reduce data redundancy and save bandwidth resources. Differentiated sampling strategies are implemented through the formula express, is a sampling scheme, where Based on the multi-level sampling scheme, the data bandwidth occupancy and network transmission delay of the system are monitored in real time. The bandwidth occupancy is calculated by the formula Calculate, where is the total bandwidth requirement, is the transmission rate of each data stream, is the number of data streams. The delay is calculated by the formula Implementation, where is the total delay, is the queue delay (i.e. the waiting time of data in the buffer), The transmission delay (i.e. the propagation time of the data packet in the network) is calculated in real time. , dynamically evaluate the current network resource utilization. According to the network load index , set priority transmission channels for data in different parameter groups. Key parameter groups It is assigned a high priority channel to ensure that it can still be transmitted first in the case of network congestion. This process is achieved through a weighted allocation formula. Implementation, where is the weight of the key parameter group, and They are the data priority factors of the key parameter group and the auxiliary parameter group. The low priority channel is assigned to transmit when the network load is low to reduce interference with high priority data. After the transmission channel is assigned, the data is differentially encoded and compressed to further optimize the transmission efficiency. Differential encoding is done by formula Implementation, where is the differential encoding value, and The time and The differential coding significantly reduces data redundancy, which greatly reduces the amount of data transmitted. The encoded data is compressed by a data compression algorithm (such as Huffman coding or arithmetic coding) to obtain the final compressed sequence. . For compressed data sequences During transmission, real-time monitoring of the transmission success rate and packet loss rate The formula for calculating the transmission success rate is: ,in is the number of packets successfully transmitted, is the total number of data packets; the packet loss rate is Through monitoring and , dynamically adjust transmission parameters, such as adjusting bandwidth allocation or retransmission strategy, to ensure data integrity and reliability. The above process generates a hierarchical data transmission sequence ,in and They are compressed data sequences of key parameter groups and auxiliary parameter groups respectively.

[0024] In one example, a feature-enhanced autoencoder is used to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area to obtain a battery health prediction model for each monitoring area, including: Performing time segmentation and normalization processing on the hierarchical data transmission sequence according to the charge and discharge cycle to obtain a standardized data sequence; Construct a feature-enhanced autoencoder network based on the standardized data sequence. The autoencoder network includes an encoder module and a decoder module. The encoder module contains 4 convolutional layers, each of which uses a ReLU activation function and a batch normalization layer. The decoder module contains 4 deconvolutional layers, each of which uses a Tanh activation function and a batch normalization layer. A residual connection module is set between the convolution layer and the deconvolution layer. The residual connection module contains three residual units. Each residual unit contains two convolution layers and a short-circuit connection. A multi-head attention layer is added between the residual units to obtain a feature enhancement network structure. The standardized data sequence is input into the feature enhancement network structure, and the encoder module is used to perform dimensionality reduction mapping to obtain a low-dimensional feature vector, and the decoder module is used to reconstruct the features to obtain a reconstructed data sequence; The reconstruction error is calculated for the reconstructed data sequence and the standardized data sequence, and the parameters of the feature enhancement network structure are optimized and updated based on the reconstruction error to obtain the optimized autoencoder network; The capacity decay prediction layer is constructed based on the low-dimensional feature vector. The capacity decay prediction layer contains three fully connected layers. Each fully connected layer uses the Sigmoid activation function, and a Dropout layer is added between the fully connected layers to obtain the battery health prediction model of each monitoring area.

[0025] In this example, the hierarchical data transmission sequence Contains voltage , Current ,temperature ,power and environmental data These data are segmented in time according to the charge and discharge cycle of the power battery, and the segment boundaries are determined by detecting the change in current direction and amplitude. represents the charging stage, and It indicates the discharge stage. For each charge and discharge cycle (No. Periods), extract the data subset within a fixed time length and perform normalization. The normalization formula is: ; in, is the original data, and are the minimum and maximum values ​​of the data respectively. The normalized data sequence Eliminate the scale differences between different parameters. Construct a feature-enhanced autoencoder network. The autoencoder network consists of an encoder module and a decoder module, where the encoder module is used for dimensionality reduction mapping and the decoder module is used to reconstruct the original data. The encoder module contains 4 convolutional layers, and the output of the convolution operation is defined as: ; in, is the input feature matrix, is the convolution kernel weight matrix, is the bias term, is the activation function. Each convolutional layer is followed by a batch normalization layer, using the formula: ; in, and are the mean and variance of the input data, respectively. and is a learnable parameter, is a small constant to prevent division by zero. The output of the encoder module is a low-dimensional feature vector , expressed as Encoder The decoder module contains 4 deconvolution layers, which map low-dimensional features back to high-dimensional space through deconvolution operations. Its output formula is: ; in, and are the weights and biases of the deconvolution kernel, Tanh is an activation function, which is used to ensure that the numerical range of the output data is consistent with the input. To enhance the feature extraction capability, a residual connection module is inserted between the convolution layer and the deconvolution layer. The residual connection module contains 3 residual units, each of which is expressed by the formula: ; Implement short-circuit connections to alleviate the gradient vanishing problem in deep networks. Introduce a multi-head attention layer between residual units to optimize the weight distribution of input features by calculating the correlation matrix between features, and enhance the model's attention to key data patterns. Input feature enhancement network structure, and obtain low-dimensional feature vector through encoder module dimensionality reduction , and reconstructed into a reconstructed data sequence through the decoder module To evaluate the quality of reconstruction, the reconstruction error is calculated : ; in, is the number of samples, and are the reconstruction and input data The network parameters are optimized by gradient descent to minimize , and finally the optimized autoencoder network is obtained. Based on the optimized low-dimensional feature vector , construct the capacity decay prediction layer. The prediction layer consists of 3 fully connected layers, and the output formula of each fully connected layer is: ; in, and are the weight matrix and bias term respectively, In order to reduce the risk of overfitting, a Dropout layer is inserted between the fully connected layers to randomly block the neuron connections with a certain probability. The capacity decay prediction layer outputs the battery health status prediction results, such as the battery capacity decay rate .

[0026] In one example, distributed training is performed in each monitoring area based on the battery health prediction model, and the model fusion weight is determined according to the data quality and prediction accuracy to obtain a global battery monitoring model set, including: Initialize the parameters of the battery health prediction model, distribute the initialized model to each monitoring area, and assign an independent model training task to each monitoring area to obtain a regional training plan; Based on the regional training plan, the training data of each monitoring area is scored for data quality. The data quality score includes weighted calculation of three dimensions: data integrity, data consistency, and data timeliness, and the regional data weight is obtained. The training process of each monitoring area is calculated in parallel, the model parameters are updated using the stochastic gradient descent method, and the prediction accuracy during the training process is monitored in real time to obtain the regional model performance indicators; The model fusion weight matrix is ​​constructed according to the regional data weight and regional model performance index, and the model parameters of each monitoring area are weighted averaged to obtain the fusion model parameters; The fusion model parameters are globally verified to obtain parameter verification results, and the model structure and parameter distribution are optimized and adjusted based on the parameter verification results to obtain a global battery monitoring model set.

[0027] In this example, the parameters of the battery health prediction model are initialized. Assume that the parameter set of the model is , initialization parameters Use uniform distribution or Gaussian distribution for random assignment. For example, when using Gaussian distribution, the initialization formula is: ; in, and are the mean and standard deviation respectively, and are set empirically as and The initialized model is distributed to each monitoring area. Assuming monitoring areas, then the initial value of the model parameter set for each area is ,in Indicates the region number. After the model is distributed, an independent model training task is assigned to each monitoring region. This task depends on the training data collected in the region. Based on the regional training plan, the data quality of each monitoring region is scored. The data quality score is divided into three dimensions: data integrity , Data consistency and data timeliness Among them, data integrity Indicates the coverage of collected data, and the calculation formula is: ; is the number of valid data points, is the total number of data points. Data consistency By calculating the smoothness of the data in time, we get: ; in, is the first values, is the total number of data points. Data timeliness It reflects the real-time nature of the data and is defined as the data upload time interval Time interval with target Deviation: ; Regional data weight Calculated based on the above scores using a weighted formula: ; in, , , is the weight coefficient, satisfying The model training process of each monitoring area is calculated in parallel. The model parameters are updated using the stochastic gradient descent method, and the gradient update formula is: ; in, is the learning rate, is the loss function, which depends on the regional training data . Real-time monitoring of prediction accuracy during training , and its calculation formula is: ; in, is the number of samples predicted correctly, is the total number of samples. The prediction accuracy is used as a model performance indicator to provide a basis for subsequent model fusion. After the training is completed, the regional data weight and model performance indicators , construct the model fusion weight matrix Assume that the regional model parameters are , fusion model parameters The weighted average formula is: ; This fusion method can comprehensively consider data quality and training effect, and improve the reliability and adaptability of the global model. Fusion model parameters Validation is performed globally. Through the validation dataset Calculating validation loss and prediction accuracy , the calculation formula of validation loss is: ; in, is the loss function of a single sample, is the true value, is the model prediction value, is the number of validation data samples. If the validation loss is high or the prediction accuracy is lower than the set threshold, the structure and parameter distribution of the fusion model are optimized, such as increasing the number of model layers or adjusting the learning rate. The optimized global battery monitoring model set is obtained through the above process. .

[0028] In one example, the collaborative monitoring data stream of the power battery is obtained, and the collaborative monitoring data stream is analyzed in multiple dimensions based on the global battery monitoring model set to obtain the battery health level assessment result, including: The computing load of the global battery monitoring model set is divided into time windows, and the computing resource usage of each monitoring area is statistically analyzed based on the time windows to obtain the resource occupancy state matrix; Perform singular value decomposition on the resource occupancy state matrix to obtain eigenvectors, and prioritize the computing tasks based on the eigenvectors to obtain a task scheduling sequence; Based on the task scheduling sequence, a computing resource balanced allocation model is established to dynamically allocate the processor occupancy, memory usage and network bandwidth of each monitoring area to obtain a load balancing solution; Extract fault features from the operating status data of each monitoring device, classify and identify the device status by building a multi-layer perceptron network, and obtain the device health index; Design a backup switching strategy for monitoring equipment based on equipment health indicators, and optimize the data flow path during the switching process based on the load balancing solution to obtain a device fault tolerance mechanism; The equipment fault tolerance mechanism is integrated with the load balancing solution to uniformly schedule the collection, transmission and processing of monitoring data to obtain a coordinated monitoring data flow. Based on the global battery monitoring model set, a multi-dimensional analysis of the collaborative monitoring data stream is performed to obtain the battery health level assessment results.

[0029] In this example, the computational load of the global battery monitoring model set is divided into time windows, and the time window size is set Divide the entire time series into For each computing task in each time window, record the processor occupancy rate of each monitoring area. , memory usage and network bandwidth usage , forming a resource occupancy state matrix: ; This matrix records the dynamic distribution of each resource dimension within the time window. In order to extract the core characteristics of resource occupancy, the matrix Perform singular value decomposition, and the decomposition formula is: ; in, is the resource feature vector matrix, is a diagonal matrix of singular values, is the eigenvector matrix of the task distribution. The eigenvector corresponding to the largest singular value , determine the priority of each computing task within the time window and form a task scheduling sequence ,in Indicates the highest priority In the task scheduling sequence Based on this, a computing resource balanced allocation model is established. Assuming that the monitoring area has nodes, and the processor usage of each node is , memory usage is , the network bandwidth is The resource balancing goal is to minimize the deviation of each node's resources. The objective function is: ; in, , , are the global averages of processor, memory and bandwidth respectively. By optimizing the objective function, the resource allocation weights are adjusted dynamically to obtain a load balancing solution. At the same time, the fault characteristics are extracted from the operating status data of each monitoring device. The operating parameters of the acquisition device, such as temperature , vibration intensity and current , and construct the feature vector The multi-layer perceptron network is used to classify and judge the equipment status. The multi-layer perceptron includes an input layer, a hidden layer, and an output layer. Its output is the equipment health index. , the classification function is: ; in, is the weight matrix, is the bias term. By training the multi-layer perceptron network, the equipment status is divided into three categories: normal, sub-healthy and faulty. , design a backup switching strategy. (threshold), the backup device is switched and the data flow path is optimized in combination with the load balancing solution. Assume that the current path is , by reallocating bandwidth resources , the optimized path Minimize the total transmission delay: ; in, is the distance between the path nodes. After integrating the device fault tolerance mechanism with the load balancing solution, the collection, transmission and processing of monitoring data are uniformly scheduled to generate a collaborative monitoring data stream. The data flow combines resource optimization allocation and equipment fault tolerance processing to adapt to complex operating environments and improve the reliability of data transmission. , using the global battery monitoring model set to perform multi-dimensional analysis, evaluate the battery health status from the time dimension, space dimension and performance dimension. For example, from the time dimension to analyze the battery capacity attenuation trend, the fitting model is: ; in, is the capacity, is the initial capacity, is the attenuation rate. From the spatial dimension, the performance differences of batteries in different regions are evaluated, and the performance indicators are: ; in, is the energy output, The battery health level is analyzed from the performance dimension, and the comprehensive health score is calculated based on multi-dimensional data: ; in, is the weight factor, comprehensive score As the basis for battery health level assessment.

[0030] In one example, a multi-dimensional analysis of the collaborative monitoring data stream is performed based on the global battery monitoring model set to obtain a battery health level assessment result, including: The collaborative monitoring data stream is input into multiple encoder modules in the global battery monitoring model set, and the voltage, current, and temperature data are mapped to a reduced dimension through four convolutional layers and ReLU activation functions to obtain the feature mapping matrix of each encoder module; The feature mapping matrix of each encoder module is enhanced through the residual connection module, and the correlation analysis of battery performance parameters is performed using three residual units and a multi-head attention layer to obtain multiple enhanced feature vectors; Inputting multiple enhanced feature vectors into multiple decoder modules in the global battery monitoring model set, reconstructing data through four deconvolution layers and Tanh activation function to obtain multiple reconstructed data sequences; The mean square error is calculated for multiple reconstructed data sequences and collaborative monitoring data streams, and the reconstruction error is normalized through a batch normalization layer to obtain multiple anomaly detection indicators; The enhanced feature vector is input into multiple capacity decay prediction layers in the global battery monitoring model set, and the battery capacity decay trend is modeled through three fully connected layers and Sigmoid activation function. The parameters are optimized through the Dropout layer between the fully connected layers to obtain multiple decay prediction values. An evaluation matrix is ​​established based on multiple anomaly detection indicators and multiple attenuation prediction values. The battery health status is comprehensively scored through weighted fusion operations to obtain the battery health level evaluation result.

[0031] In this example, the collaborative monitoring data flow is input to multiple encoder modules in a global battery monitoring model set. Contains voltage , Current and temperature Three types of core data are mapped through four convolutional layers for dimensionality reduction. The calculation formula for each convolutional layer is: ; in, is the input data tensor, is the weight matrix of the convolution kernel, is the bias term, Conv Represents the convolution operation and ReLU is an activation function used to introduce nonlinearity. The output of the convolutional layer is stacked to form a feature mapping matrix ,in Indicates the encoder module number, matrix It is a dimensionality-reduced expression of battery performance data. Feature mapping matrix It is input into the residual connection module for feature enhancement. The residual connection module contains 3 residual units, each of which consists of two convolutional layers and a short-circuit connection. The calculation formula is: ; in, is the feature after residual connection, is the convolution kernel weight, is the bias term. Short-circuit connections ensure efficient propagation of gradients and alleviate the gradient vanishing problem of deep networks. Multi-head attention layers are inserted between residual units to dynamically allocate weights by calculating the correlation matrix between features. The formula is: ; in, They are query matrix, key matrix and value matrix respectively. is the dimension of the key matrix, and the Softmax function is used for normalization. The multi-head attention layer outputs the enhanced feature vector , each Represents the enhanced features of an encoder module. The enhanced feature vector Input multiple decoder modules in the global battery monitoring model set. The decoder module consists of 4 deconvolution layers, which restore low-dimensional features to high-dimensional space through deconvolution operations. The output formula of each deconvolution layer is: ; in, and are the weight matrix and bias term of the deconvolution kernel, Tanh Used to limit the output data range. The deconvolution process generates a reconstructed data sequence , represents the collaborative data reconstructed by each decoder module. In order to evaluate the reconstruction quality, the reconstructed data sequence With input data Calculate mean square error , the formula is: ; in, is the number of data points, and are the reconstructed and original data respectively. The error is normalized through the batch normalization layer to obtain multiple anomaly detection indicators At the same time, the feature vector Input multiple capacity decay prediction layers in the global battery monitoring model set. The prediction layer consists of 3 fully connected layers, and the output formula of each layer is: ; in, is the weight matrix, is the bias term, Sigmoid The output value is limited to between 0 and 1. To prevent overfitting, a Dropout layer is added between the fully connected layers to optimize the parameters by randomly blocking the connections of neurons. The output of the prediction layer is multiple capacity decay prediction values. , each Indicates the battery capacity attenuation trend predicted for a specific module. Based on multiple anomaly detection indicators and capacity fade prediction , establish an evaluation matrix , whose elements Indicates Module No. health indicators. Through the weighted fusion formula: ; in, is the weight factor, It is a comprehensive score that serves as the final basis for the battery health level.

[0032] Among them, an evaluation matrix is ​​established according to multiple anomaly detection indicators and multiple attenuation prediction values, and the battery health status is comprehensively scored through weighted fusion operations to obtain the battery health level evaluation results, including: extracting features from historical battery data according to monitoring areas, operating conditions and environmental factors, and constructing a historical feature matrix containing battery performance parameters, operating status indicators and external environmental variables to obtain a training sample set; constructing a gradient boosting decision tree model based on the training sample set, predicting the changing trend of battery performance parameters, and calculating the confidence interval of each prediction result to obtain an initial uncertainty boundary; substituting the initial uncertainty boundary into the mixed integer programming model to establish an objective function and a set of constraints, wherein the objective function includes a prediction error term and a boundary size penalty term, and the constraints include historical data coverage constraints and boundary continuity constraints to obtain an optimization problem model; the optimization problem model is analyzed by using a gradient boosting decision tree model. The branch-and-bound algorithm is used to solve the problem and obtain the optimal uncertainty set boundary, which is dynamically adjusted with the prediction confidence and the distribution of historical data. Based on the optimal uncertainty set boundary, adaptive weight coefficients are constructed for multiple anomaly detection indicators. The weight coefficients are positively correlated with the prediction confidence and negatively correlated with the size of the uncertainty set to obtain a dynamic weight matrix. According to the dynamic weight matrix, weighted summation is performed on multiple anomaly detection indicators to obtain the abnormal state score, and weighted average is performed on multiple attenuation prediction values ​​to obtain the capacity attenuation score. The support vector regression model is used to perform nonlinear mapping between the abnormal state score and the capacity attenuation score, and a health status assessment function is established to obtain a comprehensive battery score. A hierarchical decision tree is constructed based on the comprehensive battery score to divide the battery status into four levels: normal, attention, warning and danger, and early warning information is generated in combination with the confidence of the uncertainty set to obtain the battery health level assessment result.

[0033] Reference Figure 2 This embodiment provides a remote monitoring and management device for a new energy vehicle power battery, including: Construction module 1 is used to construct a hierarchical monitoring network structure according to the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and hierarchically processed to obtain a hierarchical data transmission sequence of each monitoring area; Feature extraction module 2, used to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area using a feature-enhanced automatic encoder to obtain a battery health prediction model for each monitoring area; Model fusion module 3 is used to perform distributed training in each monitoring area based on the battery health prediction model, and determine the model fusion weight according to the data quality and prediction accuracy to obtain a global battery monitoring model set; The analysis module 4 is used to obtain the collaborative monitoring data stream of the power battery, and perform multi-dimensional analysis on the collaborative monitoring data stream based on the global battery monitoring model set to obtain the battery health level assessment result.

[0034] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0035] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0036] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0037] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0038] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0039] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0040] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A remote monitoring and management method for a new energy vehicle power battery, characterized in that: The following steps are involved: A hierarchical monitoring network structure is constructed based on the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and processed hierarchically to obtain a hierarchical data transmission sequence for each monitoring area. The feature-enhanced autoencoder is used to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area to obtain the battery health prediction model of each monitoring area. Based on the battery health prediction model, distributed training is performed in each monitoring area, and the model fusion weight is determined according to the data quality and prediction accuracy to obtain a global battery monitoring model set; The collaborative monitoring data stream of the power battery is obtained, and the collaborative monitoring data stream is analyzed in multiple dimensions based on the global battery monitoring model set to obtain a battery health level assessment result.

2. The remote monitoring and management method for new energy vehicle power batteries according to claim 1 is characterized in that: The hierarchical monitoring network structure is constructed according to the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and hierarchically processed to obtain a hierarchical data transmission sequence of each monitoring area, including: A data acquisition module and a data cache module are configured at each vehicle-mounted terminal layer, and a sampling time interval is set for the data acquisition module to obtain original sampling data of each monitoring area in the power battery; Dividing the original sampled data of each monitoring area in the power battery into regions according to the geographical location information to obtain divided data, and allocating the divided data to the corresponding regional monitoring center layer to obtain regional data grouping; An edge computing server and a data synchronization module are configured at each monitoring area layer, and the area data groups are pre-processed based on the edge computing server to obtain area pre-processed data; Transmitting the regional pre-processed data to a central management platform layer, and configuring a data backup module at the central management platform layer to obtain a data storage solution; According to the data storage scheme, a two-way communication link is configured for each level of data, and a control instruction issuing channel and a status information reporting channel are established to obtain a network communication scheme. At the same time, based on the network communication scheme, the original sampling data of each monitoring area in the power battery is cached for disconnected data and retransmitted after the network is restored to obtain an initial data set of each monitoring area in the power battery; The initial data sets of each monitoring area are processed in a hierarchical manner to obtain a hierarchical data transmission sequence.

3. The remote monitoring and management method for new energy vehicle power batteries according to claim 2 is characterized in that: The hierarchical processing of the initial data sets of the monitoring areas to obtain a hierarchical data transmission sequence includes: Identify the data types of the initial data sets of each monitoring area, divide the voltage, current, and power data into key parameter groups, and divide the temperature, position, and environment data into auxiliary parameter groups; The sampling frequency of the voltage and current data in the key parameter group is set, and the temperature and position data in the auxiliary parameter group are downsampled to obtain a multi-level sampling scheme; Based on the multi-level sampling scheme, real-time data bandwidth monitoring and network delay calculation are performed to obtain network load indicators; Based on the network load indicator, a high priority transmission channel is set for the key parameter group, and a low priority transmission channel is set for the auxiliary parameter group, to obtain a transmission channel allocation scheme; The data is differentially encoded and compressed according to the transmission channel allocation scheme to obtain a data compression sequence, and the transmission success rate and packet loss rate of the data compression sequence are monitored in real time and the transmission parameters are dynamically adjusted to obtain a hierarchical data transmission sequence.

4. The remote monitoring and management method for new energy vehicle power batteries according to claim 3 is characterized in that: The feature-enhanced automatic encoder extracts and reconstructs the battery performance features of the hierarchical data transmission sequence of each monitoring area to verify the battery health prediction model of each monitoring area, including: Performing time segmentation and normalization processing on the hierarchical data transmission sequence according to the charge and discharge cycle to obtain a standardized data sequence; Constructing a feature-enhanced autoencoder network according to the standardized data sequence, the autoencoder network comprising an encoder module and a decoder module, the encoder module comprising 4 convolutional layers, each convolutional layer using a ReLU activation function and a batch normalization layer, the decoder module comprising 4 deconvolutional layers, each deconvolutional layer using a Tanh activation function and a batch normalization layer; A residual connection module is set between the convolution layer and the deconvolution layer, wherein the residual connection module includes three residual units, each residual unit includes two convolution layers and a short-circuit connection, and a multi-head attention layer is added between the residual units to obtain a feature enhancement network structure; Input the standardized data sequence into the feature enhancement network structure, perform dimensionality reduction mapping through the encoder module to obtain a low-dimensional feature vector, and perform feature reconstruction through the decoder module to obtain a reconstructed data sequence; Calculating a reconstruction error between the reconstructed data sequence and the standardized data sequence, and optimizing and updating the parameters of the feature enhancement network structure based on the reconstruction error to obtain an optimized autoencoder network; A capacity decay prediction layer is constructed based on the low-dimensional feature vector, and the capacity decay prediction layer includes three fully connected layers. Each fully connected layer uses a Sigmoid activation function, and a Dropout layer is added between the fully connected layers to obtain a battery health prediction model for each monitoring area.

5. The remote monitoring and management method for new energy vehicle power batteries according to claim 4 is characterized in that: The distributed training is performed in each monitoring area based on the battery health prediction model, and the model fusion weight is determined according to the data quality and prediction accuracy to obtain a global battery monitoring model set, including: Initializing parameters of the battery health prediction model, distributing the initialized model to each monitoring area, and assigning an independent model training task to each monitoring area to obtain a regional training plan; Based on the regional training scheme, the training data of each monitoring area is scored for data quality, wherein the data quality score includes weighted calculation of three dimensions: data integrity, data consistency, and data timeliness, to obtain a regional data weight; The training process of each monitoring area is calculated in parallel, the model parameters are updated using the stochastic gradient descent method, and the prediction accuracy during the training process is monitored in real time to obtain the regional model performance indicators; A model fusion weight matrix is ​​constructed according to the regional data weight and the regional model performance index, and the model parameters of each monitoring area are weighted averaged and aggregated to obtain fusion model parameters; The fusion model parameters are globally verified to obtain parameter verification results, and the model structure and parameter distribution are optimized and adjusted based on the parameter verification results to obtain a global battery monitoring model set.

6. The remote monitoring and management method for new energy vehicle power batteries according to claim 5 is characterized in that: The acquiring of the collaborative monitoring data stream of the power battery and performing multi-dimensional analysis on the collaborative monitoring data stream based on the global battery monitoring model set to obtain a battery health level assessment result includes: Dividing the computing load of the global battery monitoring model set into time windows, and performing statistical analysis on the computing resource usage of each monitoring area based on the time windows to obtain a resource occupancy state matrix; Performing a singular value decomposition operation on the resource occupancy state matrix to obtain a eigenvector, and prioritizing the computing tasks based on the eigenvector to obtain a task scheduling sequence; Based on the task scheduling sequence, a computing resource balanced allocation model is established to dynamically allocate the processor occupancy rate, memory usage rate and network bandwidth of each monitoring area to obtain a load balancing solution; Extract fault features from the operating status data of each monitoring device, classify and identify the device status by building a multi-layer perceptron network, and obtain the device health index; Design a backup switching strategy for the monitoring device according to the device health index, and optimize the path of data flow during the switching process based on the load balancing solution to obtain a device fault tolerance mechanism; The device fault tolerance mechanism is integrated with the load balancing solution to uniformly schedule the collection, transmission and processing of monitoring data to obtain a collaborative monitoring data stream; Based on the global battery monitoring model set, a multi-dimensional analysis is performed on the collaborative monitoring data stream to obtain a battery health level assessment result.

7. The remote monitoring and management method for new energy vehicle power batteries according to claim 6 is characterized in that: The multi-dimensional analysis of the collaborative monitoring data stream based on the global battery monitoring model set to obtain a battery health level assessment result includes: Input the collaborative monitoring data stream into multiple encoder modules in the global battery monitoring model set, perform dimension reduction mapping on the voltage, current, and temperature data through four convolutional layers and ReLU activation functions, and obtain a feature mapping matrix for each encoder module; The feature mapping matrix of each encoder module is enhanced through the residual connection module, and the correlation analysis of battery performance parameters is performed using three residual units and a multi-head attention layer to obtain multiple enhanced feature vectors; Inputting the multiple enhanced feature vectors into multiple decoder modules in the global battery monitoring model set, reconstructing the data through four deconvolution layers and a Tanh activation function, and obtaining multiple reconstructed data sequences; Calculating mean square errors of the multiple reconstructed data sequences and the collaborative monitoring data stream, and normalizing the reconstruction errors through a batch normalization layer to obtain multiple anomaly detection indicators; Input the enhanced feature vector into multiple capacity decay prediction layers in the global battery monitoring model set, model the battery capacity decay trend through three fully connected layers and Sigmoid activation function, and perform parameter optimization through Dropout layers between the fully connected layers to obtain multiple decay prediction values; An evaluation matrix is ​​established based on the multiple abnormality detection indicators and the multiple attenuation prediction values, and the battery health status is comprehensively scored through weighted fusion operation to obtain a battery health level evaluation result.

8. A remote monitoring and management device for a new energy vehicle power battery, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: A construction module is used to construct a hierarchical monitoring network structure according to the vehicle terminal layer, the regional monitoring center layer and the central management platform layer. At the same time, the initial data set of each monitoring area in the power battery is collected and hierarchically processed to obtain a hierarchical data transmission sequence of each monitoring area; A feature extraction module is used to extract and reconstruct the battery performance features of the hierarchical data transmission sequence of each monitoring area using a feature-enhanced automatic encoder to obtain a battery health prediction model for each monitoring area; A model fusion module, used to perform distributed training in each monitoring area based on the battery health prediction model, and determine the model fusion weight according to data quality and prediction accuracy to obtain a global battery monitoring model set; The analysis module is used to obtain the collaborative monitoring data stream of the power battery, and perform multi-dimensional analysis on the collaborative monitoring data stream based on the global battery monitoring model set to obtain a battery health level assessment result.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images