A lithium battery digital twin construction method and device based on real-time data

Abstract

The application relates to the technical field of battery management, and discloses a lithium battery digital twin construction method and device based on real-time data. The method comprises the following steps: acquiring a multi-dimensional data stream of a target lithium battery in real time; performing time sequence alignment processing on the multi-dimensional data stream to obtain a simultaneous time sequence multi-dimensional data stream; constructing a dynamic data bucket, wherein the dynamic data bucket comprises a current data bucket and a closed data bucket, and the current data bucket is a data bucket that is receiving and caching real-time data and has not been marked as a closed state; performing adaptive amplitude limiting filtering processing and feature point identification processing on the latest multi-dimensional data in the simultaneous time sequence multi-dimensional data stream based on the current data bucket to update the current data bucket; and updating a current lithium battery digital twin model based on the data in the updated current data bucket to obtain an updated lithium battery digital twin model. The method realizes millisecond-level synchronization and dynamic evolution of a digital twin and a physical battery state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery management, and particularly relates to a lithium battery digital twin construction method and device based on real-time data. BACKGROUND

[0002] Lithium-ion batteries have become the core power source in the fields of electric vehicles, large-scale energy storage systems, and portable electronic devices due to their high energy density and long cycle life. However, the performance degradation and safety risks of batteries during use are key bottlenecks that hinder their further development. With the improvement of sensor technology and edge / cloud computing capabilities of the battery management system (BMS), it is possible to obtain higher frequency and finer granularity of battery operation data (such as single-cell voltage, current, and temperature), which provides a data basis for building more accurate battery management models.

[0003] Existing BMS generally uses state estimation algorithms based on equivalent circuit models or empirical formulas. These models usually rely on low-frequency sampling data at the second or minute level, making it difficult to capture the rapid dynamic processes occurring inside the battery at the millisecond to microsecond level. In recent years, digital twin technology has been introduced into the field of battery management, aiming to build a virtual mapping of the physical entity of the battery, and to achieve state monitoring, prediction, and optimization. However, the parameter updating or state synchronization of the digital twin model usually relies on low-frequency data or periodic calibration, which cannot achieve true "real-time" synchronization with the physical battery state, and cannot reflect the instantaneous changes inside the battery. The high-frequency raw data stream from the BMS is not fully utilized or effectively processed, resulting in the loss or smoothing of a large amount of high-frequency details containing key dynamic information. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a lithium battery digital twin construction method and device based on real-time data to solve the above technical problems.

[0005] The technical solution of the present application to solve the above technical problems is as follows: a lithium battery digital twin construction method based on real-time data, comprising: acquiring a multi-dimensional data stream for a target lithium battery in real time; performing time alignment processing on the multi-dimensional data stream to obtain a simultaneous time multi-dimensional data stream; constructing a dynamic data bucket, the dynamic data bucket comprising a current data bucket and a closed data bucket, the current data bucket being a data bucket that is receiving and caching real-time data and has not been marked as closed; for the latest multi-dimensional data generated in the simultaneous time multi-dimensional data stream, performing adaptive amplitude limiting filtering processing and feature point recognition processing on the latest multi-dimensional data based on the current data bucket to update the current data bucket; updating the current lithium battery digital twin model based on the data in the updated current data bucket to obtain an updated lithium battery digital twin model.

[0006] The present application has the beneficial effects that: the present application directly collects and processes raw data streams of voltage, current, temperature and the like at the battery monomer / module level, and fully excavates the internal dynamic information of the battery at the millisecond level. By constructing a dynamic data bucket mechanism, combined with time sequence alignment, adaptive filtering and feature point recognition processing, real-time cleaning and feature extraction of high-frequency data streams of lithium batteries are realized, effectively solving the problem of insufficient fidelity caused by data delay, noise interference and feature loss in traditional digital twin models. The method can continuously inject high-frequency real-time data streams into the lithium battery digital twin, realizing millisecond-level synchronization and dynamic evolution of the digital twin and the physical battery state.

[0007] On the basis of the above technical scheme, the present application can also be improved as follows.

[0008] Further, the real-time acquisition of the multi-dimensional data stream for the target lithium battery includes: based on a feature name and feature value independent transmission mechanism and a sharding aggregation mechanism, the multi-dimensional data stream is acquired in real time.

[0009] Further, the time sequence alignment processing according to the multi-dimensional data stream to obtain a simultaneous time sequence multi-dimensional data stream includes: through a nearest neighbor resampling algorithm, the timestamps of the multi-dimensional data stream are resampled, the timestamp sequences of the data of each dimension are aligned to a unified target time sequence, and the simultaneous time sequence multi-dimensional data stream is formed.

[0010] Further, the adaptive limiting filter processing and feature point recognition processing are performed on the latest multi-dimensional data in the simultaneous time sequence multi-dimensional data stream based on the current data bucket, so as to update the current data bucket, including: after the latest multi-dimensional data is generated, the current standard deviation is calculated according to the data in the current data bucket; the absolute difference between the latest multi-dimensional data and the previous sampling node data is calculated, the previous sampling node data is the multi-dimensional data obtained by the previous sampling node corresponding to the sampling node of the latest multi-dimensional data; the absolute difference is compared with the threshold value with the current standard deviation as the threshold value; when the absolute difference is less than or equal to the threshold value, the latest multi-dimensional data is added to the current data bucket as new data; when the absolute difference is greater than the threshold value, the previous sampling node data is added to the current data bucket as new data.

[0011] Further, the adaptive amplitude limiting filtering processing and the feature point identification processing are performed on the latest multidimensional data generated in the simultaneous sequence multidimensional data stream based on the current data bucket to update the current data bucket, and the method further comprises: calculating a new standard deviation according to the data in the current data bucket to which the new data is added; calculating a new bucket size based on the new standard deviation, a pre-stored historical maximum standard deviation and a pre-stored historical minimum standard deviation, and in combination with a bucket size of a previous time node and a preset scaling coefficient; and wherein the new bucket size is constrained between a bucket size initial value and a preset maximum value.

[0012] Further, the adaptive amplitude limiting filtering processing and the feature point identification processing are performed on the latest multidimensional data generated in the simultaneous sequence multidimensional data stream based on the current data bucket to update the current data bucket, and the method further comprises: comparing the new standard deviation with the pre-stored historical maximum standard deviation; when the new standard deviation is greater than the pre-stored historical maximum standard deviation, identifying and marking the new data as a feature point; after identifying that the new data is a feature point, adding a standard deviation of a last stable data bucket before the occurrence of the feature point to a pre-stored historical statistical sequence, and recalculating the pre-stored historical maximum standard deviation and the pre-stored historical minimum standard deviation based on the updated historical statistical sequence.

[0013] Further, the method further comprises: monitoring a visual rendering performance index of the lithium battery digital twin model in real time, the visual rendering performance index comprising a page frame rate, a rendering time and a layout drawing time; triggering a GPU accelerated rendering mechanism when it is detected that the page frame rate continuously is less than a first threshold value, the rendering time is greater than a second threshold value and the layout drawing time is greater than a third threshold value; and continuously monitoring and evaluating a rendering effect after the GPU accelerated rendering is enabled, so as to dynamically adjust a GPU acceleration strategy according to an evaluation result obtained.

[0014] Further, the method further comprises: monitoring a processing delay generated from data in the updated current data bucket to the updated lithium battery digital twin model; calculating a scaling coefficient evaluation value according to the processing delay and a target real-time threshold value; comparing the processing delay with the target real-time threshold value and comparing the scaling coefficient evaluation value with a preset coefficient threshold value; and starting a parameter updating mechanism when the processing delay continuously is greater than the target real-time threshold value or the scaling coefficient evaluation value is greater than the preset coefficient threshold value.

[0015] To solve the above technical problems, the application further provides a lithium battery digital twin construction device based on real-time data, comprising:

[0016] a data acquisition module configured to acquire a multi-dimensional data stream in real time for a target lithium battery;

[0017] an alignment processing module configured to perform timing alignment processing on the multi-dimensional data stream to obtain a simultaneous timing multi-dimensional data stream;

[0018] a bucket division module configured to construct dynamic data buckets, the dynamic data buckets including a current data bucket and a closed data bucket, the current data bucket being a data bucket that is receiving and caching real-time data and has not been marked as closed;

[0019] a data processing module configured to, for newly generated multi-dimensional data in the simultaneous timing multi-dimensional data stream, perform adaptive clipping filtering processing and feature point recognition processing on the newly generated multi-dimensional data based on the current data bucket, to update the current data bucket;

[0020] a model updating module configured to update a current lithium battery digital twin model based on data in the updated current data bucket, to obtain an updated lithium battery digital twin model. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 a flowchart of a lithium battery digital twin construction method based on real-time data according to an embodiment of the present application;

[0022] Figure 2 a flowchart of the first embodiment of the present application;

[0023] Figure 3 a GPU acceleration control flowchart according to the present application;

[0024] Figure 4 a schematic diagram of a lithium battery digital twin construction device based on real-time data according to the present application. DETAILED DESCRIPTION

[0025] The principles and features of the present application are described below, and the examples are only used to explain the present application and not to limit the scope of the present application.

[0026] Embodiment One

[0027] As shown in Figure 1 , the present embodiment provides a lithium battery digital twin construction method based on real-time data, comprising:

[0028] S101, acquiring a multi-dimensional data stream in real time for a target lithium battery.

[0029] S102, performing timing alignment processing on the multi-dimensional data stream to obtain a simultaneous timing multi-dimensional data stream.

[0030] S103, construct a dynamic data bucket, the dynamic data bucket includes a current data bucket and a closed data bucket, the current data bucket is a data bucket that is receiving and caching real-time data and has not been marked as a closed state.

[0031] S104, for the generated latest multidimensional data in the simultaneous sequence multidimensional data stream, respectively, the adaptive clipping filtering processing and the feature point identification processing are carried out on the latest multidimensional data based on the current data bucket, so as to update the current data bucket.

[0032] S105, based on the data in the updated current data bucket, update the current lithium battery digital twin model to obtain an updated lithium battery digital twin model.

[0033] The application directly collects and processes raw data streams such as voltage, current and temperature of battery monomers / module level, and fully excavates the internal dynamic information of the battery on the millisecond scale. By constructing a dynamic data bucket mechanism, combined with time sequence alignment, adaptive filtering and feature point identification processing, real-time cleaning and feature extraction of high-frequency data stream of lithium battery are realized, effectively solving the problem of insufficient fidelity caused by data delay, noise interference and feature loss of traditional digital twin model. The method can continuously inject high-frequency real-time data stream into the lithium battery digital twin body, realize millisecond-level synchronization and dynamic evolution of the digital twin body and the physical battery state.

[0034] Optionally, in the embodiment, the real-time acquisition of the multidimensional data stream of the target lithium battery comprises: based on the feature name and feature value independent transmission mechanism and the sharding aggregation mechanism, the multidimensional data stream is acquired in real time.

[0035] The target lithium battery is a battery monomer or a battery module. By adopting the feature name and feature value independent transmission mechanism, the high-frequency multidimensional data stream is accessed, that is, Figure 2 The data access process can realize 300%-400% improvement in throughput of 300-dimensional feature data.

[0036] Specifically, the data access process mainly includes two processes of source data encryption and data decryption. The feature name (such as voltage, current, temperature) and the feature value (such as 0.1V, 1A, 25℃) are transmitted respectively. Usually, the feature name is transmitted only when needed (such as when the connection is established, the field is changed). The feature name data transmission adopts change transmission, and a string Str (i.e. source data encryption) is transmitted after encryption. The field data is cached at the receiving end, the feature name is decrypted by aes_decrypt, and is cached locally in the server. When there is new MQTT data, the feature name is updated, as shown in Figure 2

[0037] The feature name decryption can be represented as: ​

[0038] str1_str2_str3_str4_... = aes_decrypt(Str);

[0039] Wherein, str1, str2, str3, str4 are the decrypted feature name.

[0040] The high-frequency transmission is only the data value, that is, the feature value. Also transmitted is an encrypted string Val. After decryption of the feature value, the string is split.

[0041] The feature value decryption can be expressed as:

[0042] val1_val2_val3_val4_... = aes_decrypt(Val);

[0043] Wherein, val1, val2, val3, val4 are the decrypted feature values.

[0044] After decryption of the feature value, the data decryption process is completed. Then the feature value is correspondingly spliced with the cached feature name to obtain complete JSON data.

[0045] At the data transmission level, data of different types and different collection frequencies are grouped and packaged to achieve load balancing of network transmission, that is, a sharding aggregation mechanism is adopted. According to the frequency of feature value collection, in this embodiment, vehicle cell data is taken as an example. 120 voltage values of the cell are taken as a group vol_part, 120 temperature values of the cell are taken as a group temp_part, the current is taken as a group curr_part, and other calculated values are taken as another group other_part, so that the number of fields in each group is basically the same, and the efficient use of data is maximized.

[0046] Optionally, in the embodiment, the time sequence alignment processing according to the multi-dimensional data stream to obtain the simultaneous time sequence multi-dimensional data stream comprises: resampling the time stamps of the multi-dimensional data stream by a nearest neighbor resampling algorithm, aligning the time stamp sequences of the data of each dimension to a unified target time sequence, and forming the simultaneous time sequence multi-dimensional data stream.

[0047] After decryption of the data group, the nearest neighbor resampling algorithm is used to perform time sequence alignment on the data, aligning the data of each dimension to a unified time reference, and outputting a multi-dimensional data matrix with strictly synchronized time sequences.

[0048] The nearest neighbor resampling algorithm is specifically:

[0049] Construct the target alignment time sequence: .

[0050] Nearest neighbor matching is performed: each target time point , find the time node of voltage, temperature, current and other characteristics , whose corresponding value is . It can be expressed as:

[0051] ;

[0052] where, ∈ , is the time series of variables such as voltage, temperature, current, etc.

[0053] For example, for the target time point t i =1000ms. In the voltage source data, the value at t=999ms is 3.70V and the value at t=1002ms is 3.69V. Since |1000-999|=1 is less than |1000-1002|=2, the value of 3.70V at t=999ms is selected as the voltage value of t i . In the temperature source data, the value at t=1001ms is 26.0℃ and the value at t=995ms is 25.9℃. Since |1000-1001|=1 is the smallest, the value of 26.0℃ at t=1001ms is selected as the temperature value of t i .

[0054] After the above processing, the final result is a multi-dimensional data matrix (simultaneous multi-dimensional time series data stream), where each row corresponds to a unified target time point, each column corresponds to a sensor dimension, and all data are strictly synchronized in time series.

[0055] The input simultaneous multi-dimensional time series data stream is divided into continuous dynamic data buckets , see Figure 2 Data bucketing, each contains several consecutive time points:

[0056] ;

[0057] where, is the bucket data, is the start time, is the time size of the bucket, is the time node, is the multi-dimensional data. The initial value can be selected as [1, 2, 5, 10] seconds, which is consistent with the frequency of traditional vehicle cloud data transmission.

[0058] For bucketing, Figure 2The process of the bucket data processing includes amplitude limiting filtering, feature point identification evaluation and amplitude threshold optimization. The process is described in detail as follows.

[0059] Optionally, in an embodiment, the adaptive amplitude limiting filtering processing and the feature point identification processing are performed on the latest multi-dimensional data in the current data bucket based on the latest multi-dimensional data in the time-series multi-dimensional data stream, to update the current data bucket, including: after the latest multi-dimensional data is generated, a current standard deviation is calculated according to data in the current data bucket; an absolute difference between the latest multi-dimensional data and previous sampling node data is calculated, the previous sampling node data being multi-dimensional data obtained by a previous sampling node of a sampling node corresponding to the latest multi-dimensional data; the absolute difference is compared with the threshold value with the current standard deviation as the threshold value; when the absolute difference is less than or equal to the threshold value, the latest multi-dimensional data is added to the current data bucket as new data; and when the absolute difference is greater than the threshold value, the previous sampling node data is added to the current data bucket as new data.

[0060] Specifically, the standard deviation calculation formula is:

[0061] ;

[0062] wherein n is the number of time nodes in the bucket, is the value of each time node in the bucket, is the mean value of the values in the bucket.

[0063] The current standard deviation is calculated by the above formula, and the threshold value is obtained, to realize dynamic adjustment of the gradient threshold value for determining the data stable section / mutation section.

[0064] According to the following judgment formula, the data stream amplitude limiting filtering processing is performed:

[0065] If , then ;

[0066] If , then ;

[0067] wherein is the current sampling node data, is the previous sampling node data, is the threshold value, is the output data.

[0068] Optionally, in the embodiment, the adaptive clipping filtering processing and the feature point recognition processing are respectively performed on the latest multidimensional data in the current data bucket based on the latest multidimensional data in the simultaneous sequence multidimensional data stream, so as to update the current data bucket, and the method further comprises: calculating a new standard deviation according to the data in the current data bucket to which the new data is added; calculating a new bucket size based on the new standard deviation, a pre-stored historical maximum standard deviation and a pre-stored historical minimum standard deviation, and in combination with a bucket size of a previous time node and a preset scaling coefficient; and wherein the new bucket size is constrained between a bucket size initial value and a preset maximum value.

[0069] For adaptive adjustment of the bucket size, the following formula can be referred to:

[0070] ;

[0071] wherein, is an initial value when t=0, i.e. [1, 2, 5, 10] seconds; is a preset scaling coefficient; , , are a historical maximum standard deviation, a historical minimum standard deviation and a new standard deviation, respectively; is a bucket size of a previous time node; is a fixed value, which is set to 2 min, and lower than 2 min will result in a reduced display effect of digital twinning.

[0072] Specifically, the bucket size initial value of the data bucket is set to a preset 1, 2, 5 or 10 seconds. In the running process, the bucket size of the current data bucket is adaptively adjusted in real time. The bucket size is calculated according to the formula:

[0073] ;

[0074] If the obtained result is greater than and less than , the obtained result is taken as the new bucket size; if the obtained result is less than or equal to , the new bucket size is determined to be ; if the obtained result is greater than or equal to , the new bucket size is determined to be .

[0075] Optionally, in the embodiment, based on the current data bucket, adaptive amplitude limiting filtering processing and feature point identification processing are respectively performed on the latest multidimensional data generated in the simultaneous sequence multidimensional data stream to update the current data bucket, and the method further comprises: comparing the new standard deviation with the pre-stored historical maximum standard deviation; when the new standard deviation is greater than the pre-stored historical maximum standard deviation, identifying and marking the new data as a feature point; after identifying that the new data is a feature point, adding the standard deviation of the last stable data bucket before the occurrence of the feature point to the pre-stored historical statistical sequence, and recalculating the pre-stored historical maximum standard deviation and the pre-stored historical minimum standard deviation based on the updated historical statistical sequence.

[0076] Feature point identification is judged by standard deviation. When the standard deviation exceeds the sequence standard deviation value of the stable segment (charging phase, discharging phase) detected by the experiment, it means that the point is a mutation feature point, and the feature point should be retained. At the same time, the standard deviation value before the point is counted into the sequence of standard deviations for 、 recalculation.

[0077] A multi-physics field digital twin is constructed, and a high-fidelity core model of an electro-thermal-mechanical coupling mechanism is constructed, including: multi-physics field determination and twin model construction. The twin data of the twin model is determined according to the voltage, current, temperature, SOC (state of charge of the battery cell), SOP (state of power of the battery cell), SOH (state of health of the battery cell), internal short circuit, and consistency and life estimation of the battery pack. Two visualization representation methods are adopted: one is to map the voltage, current and temperature three real-time running parameters into the color depth of the battery cell model to represent the state of the battery cell; the other is to map the life deformation associated with the swelling force, structural stress and health state (SOH) into the physical deformation of the battery cell model. Through the process, the multi-physics field digital twin is constructed and the twin model is rendered.

[0078] The rendering effect of the model is monitored in real time to determine whether the rendering reaches the expectation. If yes, the rendering is continued; otherwise, GPU acceleration processing or the like is started to improve the rendering effect.

[0079] Optionally, in the embodiment, the method further comprises: monitoring the visualization rendering performance indicators of the lithium battery digital twin model in real time, the visualization rendering performance indicators including a page frame rate, a rendering time and a layout drawing time; when it is detected that the page frame rate continuously is less than a first threshold value, the rendering time is greater than a second threshold value, and the layout drawing time is greater than a third threshold value, a GPU acceleration rendering mechanism is triggered; after the GPU acceleration rendering is enabled, the rendering effect is continuously monitored and evaluated to dynamically adjust the GPU acceleration strategy according to the evaluation result.

[0080] Figure 3 AsFigure 3 The GPU acceleration mechanism is automatically triggered when the page frame rate is detected to be continuously below 50 frames per second (first threshold value) for 10 seconds, and the rendering time exceeds 1000 milliseconds (second threshold value), and the layout / drawing time exceeds 1000 milliseconds (third threshold value) since the page loading is completed.

[0081] After the GPU acceleration is enabled, the page rendering effect is continuously monitored, and real-time evaluation is performed through the established performance digital twin model, and performance evaluation is performed again. If the performance meets the requirements, the GPU acceleration is maintained. If the performance does not meet the preset performance threshold, the GPU acceleration strategy is dynamically adjusted, and the performance is re-evaluated until the performance meets the requirements. Among them, the GPU acceleration strategy includes optimizing the priority of the acceleration elements and updating the parameters of the twin model, so as to realize the continuous optimization of the page rendering performance and the accurate improvement of the acceleration effect.

[0082] Optionally, in the embodiment, it also includes: performing real-time optimization processing on the simultaneous time series multi-dimensional data stream, adopting a time window sliding and incremental calculation strategy to minimize data processing delay, and dynamically allocating processing priorities for different dimensions or different types of data based on the current running scene.

[0083] Specifically, a dynamic computing resource allocation and stream data processing optimization strategy is adopted. Through the time window sliding mechanism and incremental calculation, the delay of full data processing is reduced. The real-time optimization objective function is defined as:

[0084] ;

[0085] Among them, is the data incremental change amount in the time window t; and are weight coefficients; is the average processing delay, and the expression is:

[0086] ;

[0087] Among them, n is the number of sampling data points; is the update completion timestamp of the i th data point; is the arrival timestamp of the i th data point.

[0088] The data priority division is specifically: according to the influence degree of the key indicators (such as the electric-thermal-mechanical coupling error) of the twin model, a dynamic priority scoring mechanism is designed:

[0089] ;

[0090] Among them, is the priority score of the data ; The update frequency of the data is The update frequency sequence is The data is prioritized into high, medium and low priority. The prioritization is based on scenarios, for example, sometimes voltage is the feature of special concern, and sometimes current is the feature of special concern, which needs to be processed in priority.

[0091] The prioritization is based on scenarios, for example, sometimes voltage is the feature of special concern, and sometimes current is the feature of special concern, which needs to be processed in priority.

[0092] Optionally, in the embodiment, it further comprises: monitoring the processing delay generated from accessing the data in the updated current data bucket to the updated lithium battery digital twin model; calculating the scaling coefficient evaluation value according to the processing delay and the target real-time threshold; comparing the processing delay with the target real-time threshold, and comparing the scaling coefficient evaluation value with the preset coefficient threshold; when the processing delay is continuously greater than the target real-time threshold or the scaling coefficient evaluation value is greater than the preset coefficient threshold, starting the parameter updating mechanism.

[0093] The parameter verification of the data processing model is performed by evaluating the scaling coefficient during the bucketing according to the time length and resource consumption of the entire quantity flow process of data access to data rendering completion.

[0094] ;

[0095] Wherein, is the time delay from data input to digital twin model to visual rendering completion; is the target real-time threshold, which can be set to 500ms; is the weight coefficient, which balances the model accuracy and rendering efficiency. When is continuously greater than the threshold or >1, the parameter updating is started.

[0096] According to the process of the present application, in the protection of useful feature point recognition, through testing, the evaluation method of confusion matrix is used for reference to calculate the recognition effect of feature points, as shown in Table 1.

[0097] Table 1 Confusion matrix of feature point recognition performance

[0098]

[0099] The accuracy rate is calculated according to the following formula:

[0100] .

[0101] ​Wherein, TP is also called True Positive or true positive; TN is also called True Negative or true negative; FP is also called False Positive or false positive; FN is also called False Negative or false negative.

[0102] The method is verified on the data of 37 new energy vehicles for 2 years, and the protection accuracy of the data feature points reaches 99.4%, which well constructs the digital twin of useful features in the data.

[0103] Embodiment two

[0104] As Figure 4 shown, the embodiment provides a lithium battery digital twin construction device 200 based on real-time data, comprising:

[0105] The data acquisition module 201 is configured to acquire a multi-dimensional data stream for a target lithium battery in real time.

[0106] The alignment processing module 202 is configured to perform time alignment processing on the multi-dimensional data stream to obtain a simultaneous time multi-dimensional data stream.

[0107] The bucket division module 203 is configured to construct a dynamic data bucket, wherein the dynamic data bucket comprises a current data bucket and a closed data bucket, and the current data bucket is a data bucket that is receiving and caching real-time data and has not been marked as a closed state.

[0108] The data processing module 204 is configured to perform adaptive amplitude limiting filtering processing and feature point identification processing on the latest multi-dimensional data in the simultaneous time multi-dimensional data stream based on the current data bucket, so as to update the current data bucket.

[0109] The model updating module 205 is configured to update the current lithium battery digital twin model based on the data in the updated current data bucket to obtain an updated lithium battery digital twin model.

[0110] Optionally, in the embodiment, the data acquisition module 201 comprises:

[0111] The data acquisition unit is configured to acquire the multi-dimensional data stream in real time based on a feature name and feature value independent transmission mechanism and a sharding aggregation mechanism.

[0112] Optionally, in the embodiment, the alignment processing module 202 comprises:

[0113] An alignment processing unit is configured to align timestamp sequences of data in each dimension to a uniform target time sequence by a nearest neighbor resampling algorithm, and form the simultaneous time sequence multi-dimensional data stream.

[0114] Optionally, in an embodiment, the data processing module 204 comprises:

[0115] A current standard deviation calculation unit is configured to calculate a current standard deviation according to data in the current data bucket after the latest multi-dimensional data is generated;

[0116] An absolute difference calculation unit is configured to calculate an absolute difference between the latest multi-dimensional data and previous sampling node data, the previous sampling node data being multi-dimensional data obtained by a previous sampling node of a sampling node corresponding to the latest multi-dimensional data;

[0117] A first comparison unit is configured to compare the absolute difference with the current standard deviation as a threshold value;

[0118] A first data merging unit is configured to add the latest multi-dimensional data as new data to the current data bucket when the absolute difference is less than or equal to the threshold value;

[0119] A second data merging unit is configured to add the previous sampling node data as new data to the current data bucket when the absolute difference is greater than the threshold value.

[0120] Optionally, in an embodiment, the data processing module 204 further comprises:

[0121] A new standard deviation calculation unit is configured to calculate a new standard deviation according to data in the current data bucket to which the new data is added;

[0122] A bucket size updating unit is configured to calculate a new bucket size based on the new standard deviation, a pre-stored historical maximum standard deviation and a pre-stored historical minimum standard deviation, and in combination with a bucket size of a previous time node and a preset scaling coefficient;

[0123] The new bucket size is constrained between a bucket size initial value and a preset maximum value.

[0124] Optionally, in an embodiment, the data processing module 204 further comprises:

[0125] A second comparison unit is configured to compare the new standard deviation with the pre-stored historical maximum standard deviation;

[0126] A feature point identification unit is configured to identify and mark the new data as a feature point when the new standard deviation is greater than the pre-stored historical maximum standard deviation;

[0127] a sequence updating unit configured to, after identifying that the new data is a feature point, add a standard deviation of a last stable data bucket before the feature point to a pre-stored historical statistical sequence, and recompute a pre-stored historical maximum standard deviation and a pre-stored historical minimum standard deviation based on the updated historical statistical sequence.

[0128] Optionally, in an embodiment, the device further comprises:

[0129] a rendering monitoring module configured to monitor a visual rendering performance index of the lithium battery digital twin model in real time, the visual rendering performance index comprising a page frame rate, a rendering time and a layout drawing time;

[0130] an acceleration module configured to trigger a GPU accelerated rendering mechanism when it is detected that the page frame rate continuously is less than a first threshold value, the rendering time is greater than a second threshold value and the layout drawing time is greater than a third threshold value;

[0131] a strategy adjustment module configured to, after the GPU accelerated rendering is enabled, continuously monitor and evaluate a rendering effect, and dynamically adjust a GPU acceleration strategy according to an evaluation result obtained.

[0132] Optionally, in an embodiment, the device further comprises:

[0133] a time delay monitoring module configured to monitor a processing time delay from accessing data in the updated current data bucket to generating the updated lithium battery digital twin model;

[0134] a coefficient evaluation module configured to calculate a scaling coefficient evaluation value according to the processing time delay and a target real-time threshold value;

[0135] a third comparison module configured to compare the processing time delay with the target real-time threshold value, and compare the scaling coefficient evaluation value with a preset coefficient threshold value;

[0136] a parameter updating module configured to, when the processing time delay continuously is greater than the target real-time threshold value or the scaling coefficient evaluation value is greater than the preset coefficient threshold value, start a parameter updating mechanism.

[0137] In some embodiments, the real-time data-based lithium battery digital twin construction apparatus 200 of the present application can be implemented in a combination of software and hardware. For example, the real-time data-based lithium battery digital twin construction apparatus 200 of the present application can be a hardware decoding processor programmed to perform the real-time data-based lithium battery digital twin construction method of the present application. For example, the hardware decoding processor can be one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic elements.

[0138] The modules described in the embodiments of the present application can be implemented in software or hardware. In some cases, the names of the modules do not limit the modules themselves.

[0139] The above description is merely preferred embodiments of the present application and a description of the principles of the technology used. Those skilled in the art should understand that the disclosed scope of the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the disclosed concept. For example, the above features can be replaced with similar features disclosed in the present application (but not limited to) to form technical solutions.

[0140] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and represent a specific order or sequence. In appropriate cases, the order of similar objects can be interchanged, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described.

[0141] Those skilled in the art know that the present application can be implemented as a system, a method or a computer program product, therefore, the present application can be embodied in the form of a complete hardware, a complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, which is generally referred to as "circuit", "module" or "system". In addition, in some embodiments, the present application can also be implemented as a computer program product in one or more computer readable media, which contains computer readable program codes.

[0142] Although the embodiments of the present application have been shown and described above, it should be understood by those skilled in the art that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for constructing a lithium battery digital twin based on real-time data, characterized in that, include: Real-time acquisition of multi-dimensional data streams for the target lithium battery; Perform time-series alignment processing on the multidimensional data stream to obtain a simultaneous multidimensional data stream; Construct dynamic data buckets, which include current data buckets and closed data buckets. The current data bucket is the data bucket that is receiving and caching real-time data and has not yet been marked as closed. For the latest multidimensional data generated in the simultaneous multidimensional data stream, adaptive amplitude limiting filtering and feature point recognition processing are performed on the latest multidimensional data based on the current data bucket to update the current data bucket; Based on the updated data in the current data bucket, the current lithium battery digital twin model is updated to obtain the updated lithium battery digital twin model. Specifically, for the latest multidimensional data generated in the simultaneous multidimensional data stream, adaptive amplitude limiting filtering and feature point recognition processing are performed on the latest multidimensional data based on the current data bucket to update the current data bucket, including: After generating the latest multidimensional data, calculate the current standard deviation based on the data in the current data bucket; Calculate the absolute difference between the latest multidimensional data and the data of the previous sampling node, where the data of the previous sampling node is the multidimensional data obtained from the sampling node preceding the sampling node corresponding to the latest multidimensional data; Using the current standard deviation as a threshold, the absolute difference is compared with the threshold. When the absolute difference is less than or equal to the threshold, the latest multidimensional data is added as new data to the current data bucket; When the absolute difference is greater than the threshold, the data from the previous sampling node is added as new data to the current data bucket. The step of updating the current data bucket by performing adaptive amplitude limiting filtering and feature point recognition processing on the latest multidimensional data generated in the simultaneous multidimensional data stream based on the current data bucket, further includes: Calculate the new standard deviation based on the data in the current data bucket containing the newly added data; Based on the new standard deviation, the pre-stored historical maximum standard deviation and the pre-stored historical minimum standard deviation, and combined with the bucket size of the previous time node and the preset scaling factor, the new bucket size is calculated. The new bucket size is constrained between an initial bucket size value and a preset maximum value; The step of updating the current data bucket by performing adaptive amplitude limiting filtering and feature point recognition processing on the latest multidimensional data generated in the simultaneous multidimensional data stream based on the current data bucket, further includes: Compare the new standard deviation with the pre-stored historical maximum standard deviation; When the new standard deviation is greater than the pre-stored historical maximum standard deviation, the newly added data is identified and marked as a feature point; After identifying the newly added data as a feature point, the standard deviation of the last stable data bucket before the feature point occurred is added to the pre-stored historical statistical sequence, and the pre-stored historical maximum standard deviation and the pre-stored historical minimum standard deviation are recalculated based on the updated historical statistical sequence.

2. The method of claim 1, wherein, The real-time acquisition of multi-dimensional data streams for the target lithium battery includes: acquiring the multi-dimensional data streams in real time based on an independent transmission mechanism of feature names and feature values ​​and a segmentation and aggregation mechanism.

3. The method of claim 1, wherein, The step of performing time-series alignment processing on the multidimensional data stream to obtain a simultaneous multidimensional data stream includes: resampling the timestamps of the multidimensional data stream using a nearest neighbor resampling algorithm, aligning the timestamp sequences of data in each dimension to a unified target time series, and forming the simultaneous multidimensional data stream.

4. The method of claim 1, wherein, Also includes: Real-time monitoring of the visualization rendering performance metrics of the lithium battery digital twin model, including page frame rate, rendering time, and layout drawing time; When it is detected that the page frame rate is consistently lower than the first threshold, the rendering time is greater than the second threshold, and the layout drawing time is greater than the third threshold, the GPU accelerated rendering mechanism is triggered. After GPU-accelerated rendering is enabled, the rendering effect is continuously monitored and evaluated, and the GPU acceleration strategy is dynamically adjusted based on the evaluation results.

5. The method of claim 1, wherein, Also includes: Monitor the processing latency from the updated data in the current data bucket to the generation of the updated lithium battery digital twin model; Calculate the scaling factor evaluation value based on the processing latency and the target real-time threshold; The processing latency is compared with the target real-time threshold, and the scaling factor evaluation value is compared with a preset coefficient threshold. When the processing delay is continuously greater than the target real-time threshold or the scaling factor evaluation value is greater than the preset coefficient threshold, the parameter update mechanism is activated.

6. A lithium battery digital twin construction device based on real-time data, characterized in that, include: The data acquisition module is used to acquire multi-dimensional data streams for the target lithium battery in real time. An alignment processing module is used to perform time-series alignment processing on the multidimensional data stream to obtain a simultaneous multidimensional data stream. The bucket partitioning module is used to construct dynamic data buckets, which include current data buckets and closed data buckets. The current data bucket is the data bucket that is receiving and caching real-time data and has not yet been marked as closed. The data processing module is used to perform adaptive amplitude limiting filtering and feature point recognition processing on the latest multidimensional data generated in the simultaneous multidimensional data stream based on the current data bucket, so as to update the current data bucket. The model update module is used to update the current lithium battery digital twin model based on the updated data in the current data bucket, so as to obtain the updated lithium battery digital twin model. The data processing module includes: The current standard deviation calculation unit is used to calculate the current standard deviation based on the data in the current data bucket after generating the latest multidimensional data. The absolute difference calculation unit is used to calculate the absolute difference between the latest multidimensional data and the data of the previous sampling node, wherein the data of the previous sampling node is the multidimensional data obtained from the sampling node preceding the sampling node corresponding to the latest multidimensional data. The first comparison unit is used to compare the absolute difference with the threshold, using the current standard deviation as the threshold. The first data merging unit is used to add the latest multidimensional data as new data to the current data bucket when the absolute difference is less than or equal to the threshold. The second data merging unit is used to add the data of the previous sampling node as new data to the current data bucket when the absolute difference is greater than the threshold. The data processing module also includes: A new standard deviation calculation unit is used to calculate the new standard deviation based on the data in the current data bucket containing the newly added data; The bucket size update unit is used to calculate the new bucket size based on the new standard deviation, the pre-stored historical maximum standard deviation and the pre-stored historical minimum standard deviation, and in combination with the bucket size of the previous time node and the preset scaling factor. The new bucket size is constrained between an initial bucket size value and a preset maximum value; The data processing module also includes: The second comparison unit is used to compare the new standard deviation with the pre-stored historical maximum standard deviation. The feature point recognition unit is used to identify and mark the newly added data as feature points when the new standard deviation is greater than the pre-stored historical maximum standard deviation; The sequence update unit is used to add the standard deviation of the last stationary data bucket before the occurrence of the feature point to the pre-stored historical statistical sequence after the newly added data is identified as a feature point, and to recalculate the pre-stored historical maximum standard deviation and the pre-stored historical minimum standard deviation based on the updated historical statistical sequence.

Images