Power battery energy efficiency optimization method and system based on big data fusion

Abstract

The application provides a power battery energy efficiency optimization method and system based on big data fusion, comprising the following steps: S1, constructing a three-dimensional multi-modal data acquisition architecture, and synchronously acquiring electrical, thermodynamic, mechanical stress and electrochemical impedance spectrum data streams; S2, establishing an edge lightweight feature extraction model, adaptively pre-processing and feature engineering the multi-modal data, and generating an edge feature vector; S3, establishing a cloud deep energy efficiency state fusion model, calculating risk propagation weights based on a graph attention mechanism, and realizing multi-scale space-time feature fusion; S4, establishing a multi-level energy efficiency state quantitative evaluation system, and calculating a health state index, an energy efficiency attenuation probability and a remaining service life; and S5, constructing a cloud-edge collaborative closed-loop feedback test mechanism, dynamically adjusting a data acquisition frequency and an optimization strategy, forming a virtual-real mapping optimization closed loop, and realizing continuous improvement of energy efficiency.

Description

Technical Field

[0001] This invention relates to the field of data analysis, and in particular to a method and system for optimizing the energy efficiency of power batteries based on big data fusion. Background Technology

[0002] As the electric vehicle industry continues to expand, the energy efficiency of power batteries has become a key bottleneck restricting the overall vehicle range and user experience. Currently, due to issues such as low energy utilization, rapid range degradation, and inefficient energy management, range anxiety among electric vehicle users has not been fundamentally alleviated. Existing energy efficiency assessment methods are mostly limited to offline laboratory testing, relying on periodic static capacity testing and human experience judgment, making it difficult to capture the dynamic evolution of energy efficiency degradation trends during actual vehicle operation. Specifically, existing technologies for optimizing the energy efficiency of power batteries have the following shortcomings: First, evaluation methods often focus on electrical performance indicators while neglecting the comprehensive impact of thermo-mechanical-chemical coupling on battery energy conversion efficiency, making it difficult to fully identify the multidimensional factors leading to energy efficiency loss. Second, there are significant differences between standard laboratory testing conditions and complex real-world vehicle operating conditions, causing energy efficiency assessment results to deviate from actual operating energy consumption, resulting in assessment bias and inaccurate energy efficiency management. Third, response mechanisms are relatively lagging; traditional periodic maintenance and offline analysis modes cannot provide minute-level real-time responses and optimization adjustments to dynamic changes such as energy efficiency degradation and abnormal energy consumption during real-world vehicle operation. Fourth, a closed-loop energy efficiency data system covering the entire lifecycle from production to service to retirement has not yet been established, lacking the ability to track and continuously optimize the battery energy efficiency evolution trajectory over a long period, making it difficult to achieve optimal energy efficiency management throughout the entire lifecycle. Therefore, it is urgent to construct a multi-dimensional energy efficiency situational awareness and optimization system that integrates cloud computing power and edge real-time capabilities to achieve accurate assessment, dynamic optimization, and continuous improvement of power battery energy efficiency throughout its entire lifecycle. Summary of the Invention

[0003] This invention aims to solve the technical problems existing in the energy efficiency management technology of power batteries, such as single data dimension, insufficient evaluation accuracy and delayed optimization response. It innovatively proposes a power battery energy efficiency optimization method and system based on big data fusion.

[0004] To achieve the above-mentioned objectives of this invention, this invention provides a method for optimizing the energy efficiency of power batteries based on big data fusion, comprising:

[0005] S1, construct a three-dimensional multimodal data acquisition architecture to simultaneously acquire electrical, thermodynamic, mechanical stress and electrochemical impedance spectroscopy data streams;

[0006] S2, Establish a lightweight feature extraction model for the edge, perform adaptive preprocessing and feature engineering on multimodal data, and generate edge feature vectors;

[0007] S3 establishes a cloud-based deep energy efficiency state fusion model, calculates risk propagation weights based on graph attention mechanism, and achieves multi-scale spatiotemporal feature fusion;

[0008] S4. Establish a multi-level energy efficiency status quantitative assessment system to calculate the health status index, energy efficiency degradation probability and remaining service life.

[0009] S5 constructs a cloud-edge collaborative closed-loop feedback testing mechanism, dynamically adjusts the data collection frequency and optimization strategy, and forms a virtual-real mapping optimization closed loop.

[0010] In a preferred embodiment of the above technical solution, S1 includes:

[0011] A three-dimensional, multimodal data acquisition architecture is constructed, and a distributed sensor network is deployed within the battery pack to simultaneously acquire electrical parameters, thermodynamic parameters, mechanical stress parameters, and electrochemical impedance spectroscopy data, establishing a multi-source heterogeneous data stream. The electrical parameters include: single-cell voltage data. Bus current data Insulation impedance data Where i represents the cell number, i=1,2,…,N, N is the total number of cells, and t is the sampling time; the thermodynamic parameters include: cell surface temperature data. Module inlet and outlet coolant temperature difference data Ambient temperature The mechanical stress parameters include: battery casing strain. acceleration Where j represents the strain gauge placement location number, and k represents the axial marker of the accelerometer; the electrochemical impedance spectroscopy data is the real impedance obtained through a distributed EIS sensor. With imaginary impedance The subscript re represents the real part, the subscript im represents the imaginary part, and f represents the sweep frequency point.

[0012] In a preferred embodiment of the above technical solution, step S2 includes:

[0013] A lightweight feature extraction model for edge detection is established, and adaptive preprocessing and feature engineering are performed on the multimodal data collected by S1 to generate edge feature vectors. First, the various types of sensor data are divided into sliding window segments, with the window length denoted as . The sliding step size is denoted as The data matrix within the time window is obtained. Where M is the total number of sensor channels; then, adaptive weighted normalization is used to eliminate dimensional differences, and the normalized data... The calculation formula is: ;in, This represents the raw data of the m-th sensor channel at the l-th sampling point, where m = 1, 2, ..., M, l = 1, 2, ..., ; The average of historical data for the m-th channel; The standard deviation of the historical data for the m-th channel; As the numerical stability factor, take Magnitude; The channel weight coefficient, calculated based on information entropy, characterizes the sensitivity of this physical quantity to energy efficiency status. Its calculation formula is as follows: ;in, Let m be the information entropy of the data in channel m. The probability that the data from this channel falls into the c-th discrete interval, where C is the total number of discrete intervals, is given by the denominator. of To iterate through the index, the index terms of all channels including m itself are summed.

[0014] In a preferred embodiment of the above technical solution, step S3 includes:

[0015] Establish a cloud-based deep energy efficiency state fusion model and receive feature vectors from the edge via a 5G communication link. And combined with the full lifecycle historical data stored in the cloud. To conduct deep situational awareness; the aforementioned full lifecycle historical data Includes: formation stage capacity Number of loops Cumulative fast delivery volume Historical abuse records A multi-scale spatiotemporal graph convolutional network is constructed to model the electrical connections between individual cells within the battery pack as a graph structure G=(V,E), where V is the set of individual battery nodes, |V|=N, and E is the set of edges representing the connections; node feature matrix. Composed of single-unit level features extracted by S2, For node feature dimensions.

[0016] In a preferred embodiment of the above technical solution, step S4 includes:

[0017] A multi-level energy efficiency status quantitative assessment system is established, and the health status index SOH(t) and energy efficiency degradation probability are calculated based on the deep learning output of S3. and remaining useful life RUL(t); the health status index is calculated using the following formula: in, The current available capacity is estimated based on Kalman filtering. The nominal capacity; the energy efficiency degradation probability. Calculated by fusing multi-dimensional anomaly scores: ;in, The electrical anomaly score is calculated based on voltage dispersion and current fluctuation rate. The heat loss anomaly fraction is calculated based on temperature field non-uniformity and temperature rise rate. The fractional mechanical abuse is calculated based on strain amplitude and vibration spectrum characteristics; The electrochemical decay fraction is calculated based on impedance spectral shift. For each dimension's weight coefficients, satisfying .

[0018] In a preferred embodiment of the above technical solution, step S5 includes:

[0019] A closed-loop feedback testing mechanism for cloud-edge collaboration is constructed, with the cloud generating optimization strategy adjustment instructions based on the S4 evaluation results. The data is then distributed to the edge for execution; the optimization strategy includes: adjusting the data acquisition frequency. Switching feature extraction algorithm modes Trigger the active performance evaluation procedure ;when At that time, among them This is the risk threshold.

[0020] In a preferred embodiment of the above technical solution, step S5 includes:

[0021] Optimization strategy adjustment instructions For a three-dimensional decision vector, its mathematical expression is: ,in Let t be the data acquisition frequency at time t. This represents the feature extraction algorithm mode at time t. The trigger flag for the active performance evaluation program at time t;

[0022] First, the comprehensive testing demand index for cloud computing. The formula for quantifying the current requirement for test accuracy is as follows: ;

[0023] in, This represents the risk probability after exponential smoothing. ,in The probability of energy efficiency degradation calculated for S4. ∈[0,1] is the smoothing coefficient, which is set to 0.7, and Δt is the sampling time interval; =SOH(t) / 100% represents the normalized health status, and SOH(t) is the health status index calculated by S4; The remaining useful life calculated for S4; Nominal initial life; The demand weighting coefficient satisfies Furthermore, it is configured according to the differences in battery chemistry systems.

[0024] In a preferred embodiment of the above technical solution, step S5 includes:

[0025] Cloud-based evaluation of edge resource constraint functions The available computing and communication resources at the edge are represented by the following formula: ;

[0026] in This represents the remaining bandwidth of the 5G link at the current moment. Maximum bandwidth; Percentage of idle computing power of edge processors; Backup power supply for the edge gateway; This is the minimum operating power threshold;

[0027] Based on the above and Calculate the three components of the optimization strategy respectively:

[0028] (1) Data acquisition frequency Adaptive computation:

[0029] First, calculate the combined rate of change of the multimodal signals. : In the formula: This represents the normalized data for the m-th channel in S2. S2 represents the channel weighting coefficients calculated for S2; M represents the total number of sensor channels; d / dt is approximated using the difference method and is expressed as... ;

[0030] Then, calculate the minimum frequency requirement based on the risk profile. :

[0031] The calculation formula is ;

[0032] In the formula: coefficient 2 satisfies the Nyquist minimum sampling requirement; These are the sampling coefficients;

[0033] Establish a frequency-resource tradeoff function :

[0034] The calculation formula is ;

[0035] In the formula: Base frequency; This is the hardware's maximum frequency limit; To prevent division by zero, a numerical stability factor is used; the first term is the optimization benefit term, and the second term is the resource penalty term;

[0036] Solving the optimization problem yields : ;

[0037] Its analytical solution is: ;

[0038] When the calculation result is lower At that time, take Higher than At that time, take .

[0039] In a preferred embodiment of the above technical solution, step S5 includes:

[0040] Feature extraction algorithm mode Decision calculation:

[0041] First, calculate the complexity index of the current operating condition. :

[0042] ;

[0043] in This represents the standard deviation of the current voltage range. This represents the historical average voltage range. The graph attention coefficients calculated for S3; The mean of the attention coefficient; This is the operating condition identifier; it is 1.5 for fast charging, 1.0 for slow charging, and 0.8 for standby.

[0044] Then, calculate the edge computing load margin. : ;

[0045] In the formula: This represents the actual time spent on feature extraction in the previous cycle. This represents the duration of the current sampling period.

[0046] Establish dynamic decision thresholds and : ;

[0047] ;

[0048] in =0.3 is the base threshold; =0.7 is the base threshold;

[0049] Final decision for:

[0050] ;

[0051] in =1 indicates lightweight statistical mode. =2 indicates a deep time-series mode. =3 indicates the full graph neural network mode;

[0052] (3) Active performance evaluation procedure Trigger determination:

[0053] First, calculate the risk accumulation index. :

[0054] ;

[0055] in =300s is the risk observation window; The attenuation coefficient is used; the integral is approximated by discrete summation: , where K= / Δτ;

[0056] Then, calculate the risk mutation rate. : ;

[0057] Establish aging stage factors :

[0058] ;

[0059] Adjust the trigger threshold: , ,in =0.6× As the cumulative risk threshold, =0.85 is the high-risk threshold;

[0060] The final trigger determination is

[0061] ;

[0062] In the formula: This is an indicator function that takes the value 1 if the condition is met, and 0 otherwise. This is the threshold for the risk mutation rate.

[0063] The present invention also discloses a computer system, comprising:

[0064] processor;

[0065] Memory used to store processor-executable instructions;

[0066] The processor is configured to implement the power battery energy efficiency optimization method based on big data fusion as described in any one of claims 1 to 9 when executing the executable instructions.

[0067] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0068] This invention overcomes the limitations of traditional single-mode energy consumption monitoring by constructing a three-dimensional multimodal data acquisition architecture, achieving comprehensive perception of battery energy efficiency status under thermo-mechanical-electrochemical coupling. Through a collaborative computing architecture combining lightweight feature extraction at the edge and deep energy efficiency situational awareness in the cloud, it resolves the contradiction between real-time energy efficiency monitoring and analytical accuracy, with millisecond-level response at the edge and second-level deep analysis in the cloud complementing each other. By continuously accumulating historical data throughout the entire lifecycle and constructing digital twins, a complete data chain from production formation to decommissioning and recycling is established, enabling traceability and predictability of aging trajectories and providing a data foundation for accurate judgment of energy efficiency degradation trends. Through a multi-level energy efficiency status quantitative evaluation system, subjective experience judgments are transformed into objective probability outputs, significantly improving the accuracy and interpretability of energy efficiency degradation assessment and achieving refined management of battery system energy efficiency.

[0069] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0070] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0071] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation

[0072] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0073] like Figure 1 As shown, this invention discloses a method and system for optimizing the energy efficiency of power batteries based on big data fusion, comprising:

[0074] S1. Construct a three-dimensional, multimodal data acquisition architecture, deploy a distributed sensor network within the battery pack, and simultaneously acquire electrical parameters, thermodynamic parameters, mechanical stress parameters, and electrochemical impedance spectroscopy data to establish a multi-source heterogeneous data stream; the electrical parameters include: single-cell voltage data. Bus current data Insulation impedance data Where i represents the cell number, i=1,2,…,N, N is the total number of cells, and t is the sampling time; the thermodynamic parameters include: cell surface temperature data. Module inlet and outlet coolant temperature difference data Ambient temperature The mechanical stress parameters include: battery casing strain. acceleration Where j represents the strain gauge placement location number, and k represents the axial marker of the accelerometer; the electrochemical impedance spectroscopy data is the real impedance obtained through a distributed EIS sensor. With imaginary impedance In this context, the subscript re represents the real part of the letter description, the subscript im represents the imaginary part of the letter description, and f represents the sweep frequency point.

[0075] S2. Establish a lightweight feature extraction model for the edge, and perform adaptive preprocessing and feature engineering on the multimodal data collected in S1 to generate edge feature vectors. First, the various types of sensor data are divided into sliding window segments, with the window length denoted as . The sliding step size is denoted as The data matrix within the time window is obtained. Where M is the total number of sensor channels; then, adaptive weighted normalization is used to eliminate dimensional differences, and the normalized data... The calculation formula is: ;in, This represents the raw data of the m-th sensor channel at the l-th sampling point, where m = 1, 2, ..., M, l = 1, 2, ..., ; The average of historical data for the m-th channel; The standard deviation of the historical data for the m-th channel; As the numerical stability factor, take Magnitude; The channel weight coefficient, calculated based on information entropy, characterizes the sensitivity of this physical quantity to energy efficiency status. Its calculation formula is as follows: ;in, Let m be the information entropy of the data in channel m. The probability that the data from this channel falls into the c-th discrete interval, where C is the total number of discrete intervals, is given by the denominator. of To iterate through the index, the index terms of all channels including m itself are summed.

[0076] Furthermore, temporal statistical features, frequency domain energy features, and shallow temporal series features are extracted from the normalized data to construct an edge feature vector. Where D is the feature dimension and voltage range Temperature standard deviation impedance spectrum low-frequency slope Energy efficiency indicators;

[0077] S3 establishes a cloud-based deep energy efficiency state fusion model, receiving feature vectors from the edge via a 5G communication link. And combined with the full lifecycle historical data stored in the cloud. Conduct in-depth energy efficiency situational awareness; the aforementioned full lifecycle historical data Includes: formation stage capacity Number of loops Cumulative fast delivery volume Historical abuse records A multi-scale spatiotemporal graph convolutional network is constructed to model the electrical connections between individual cells within the battery pack as a graph structure G=(V,E), where V is the set of individual battery nodes, |V|=N, and E is the set of edges representing the connections; node feature matrix. Composed of single-unit level features extracted by S2, For node feature dimensions;

[0078] The risk propagation weights and attention coefficients between individuals are calculated using a graph attention mechanism. The calculation formula is:

[0079] ;in, Let be the initial feature vector of node i; The characteristic transformation matrix, For the internal dimensions of the attention mechanism; This represents the attention weight vector; [.||.] indicates a vector concatenation operation. Represents the set of first-order neighbor nodes of node i;

[0080] S4 establishes a multi-level energy efficiency status quantitative assessment system, and calculates the health status index SOH(t) and energy efficiency degradation probability based on the deep learning output of S3. and remaining useful life RUL(t); the health status index is calculated using the following formula: in, The current available capacity is estimated based on Kalman filtering. The nominal capacity; the energy efficiency degradation probability. Calculated by fusing multi-dimensional anomaly scores: ;in, The electrical anomaly score is calculated based on voltage dispersion and current fluctuation rate. The heat loss anomaly fraction is calculated based on temperature field non-uniformity and temperature rise rate. The fractional mechanical abuse is calculated based on strain amplitude and vibration spectrum characteristics; The electrochemical decay fraction is calculated based on impedance spectral shift. For each dimension's weight coefficients, satisfying Furthermore, based on the dynamic adjustment of the battery chemistry system, for ternary lithium batteries... =0.35, which is higher than the 0.25 for lithium iron phosphate batteries;

[0081] The prediction of the remaining useful life RUL(t) is based on an improved double-exponential model that takes into account the fast-charging acceleration factor: ;in, The end-of-life threshold capacity is typically set at 80%. ; Activation energy, unit: J R is the ideal gas constant, 8.314 J. ; The historical average operating temperature; λ represents the historical average fast charging rate; λ is the fast charging acceleration coefficient, experimentally calibrated to 0.015. ; This is the aging rate constant;

[0082] S5 establishes a closed-loop feedback optimization mechanism for cloud-edge collaboration, whereby the cloud generates optimization strategy adjustment instructions based on the evaluation results of S4. The data is then distributed to the edge for execution; the optimization strategy includes: adjusting the data acquisition frequency. Switching feature extraction algorithm modes Trigger the active performance evaluation procedure ;when At that time, among them The risk threshold is set to 0.75. At the edge, the system immediately switches to high-frequency monitoring mode, increasing the sampling frequency from 1Hz to 10Hz, and initiates a rapid impedance spectroscopy scan, covering a frequency range of... Extended from [1Hz, 1kHz] to [0.1Hz, 10kHz] to capture interface impedance variations;

[0083] At the same time, a digital twin is established in the cloud. Real-time synchronization of physical battery state; prediction of optimal energy efficiency boundary under different operating conditions through finite element-equivalent circuit coupled simulation; simulation error feedback coefficient. Used to correct the model parameters in S4, forming an optimized closed loop for virtual-real mapping.

[0084] The optimization strategy adjustment instruction For a three-dimensional decision vector, its mathematical expression is: ,in Let t be the data acquisition frequency at time t. This represents the feature extraction algorithm mode at time t. The trigger flag for the active performance evaluation program at time t;

[0085] First, the comprehensive testing demand index for cloud computing. The formula for quantifying the current requirement for test accuracy is as follows: ;

[0086] in, This represents the risk probability after exponential smoothing. ,in The probability of energy efficiency degradation calculated for S4. ∈[0,1] is the smoothing coefficient, which is set to 0.7, and Δt is the sampling time interval; =SOH(t) / 100% represents the normalized health status, and SOH(t) is the health status index calculated by S4; The remaining useful life calculated for S4; Nominal initial life; The demand weighting coefficient satisfies Furthermore, based on the differentiated settings of the battery chemistry system: for the ternary lithium system, take... =0.5, =0.3, =0.2, for the lithium iron phosphate system, take =0.3, =0.4, =0.3;

[0087] Meanwhile, the cloud evaluates the edge resource constraint function. The available computing and communication resources at the edge are represented by the following formula: ;

[0088] in This represents the remaining bandwidth of the 5G link at the current moment. Maximum bandwidth; Percentage of idle computing power of edge processors; Backup power supply for the edge gateway; This is the minimum operating power threshold;

[0089] Based on the above and Calculate the three components of the optimization strategy respectively:

[0090] (1) Data acquisition frequency Adaptive computation:

[0091] First, calculate the combined rate of change of the multimodal signals. : In the formula: This represents the normalized data for the m-th channel in S2. S2 represents the channel weighting coefficients calculated for S2; M represents the total number of sensor channels; d / dt is approximated using the difference method and is expressed as... ;

[0092] Then, calculate the minimum frequency requirement based on the risk profile. :

[0093] The calculation formula is ;

[0094] In the formula: coefficient 2 satisfies the Nyquist minimum sampling requirement; The sampling coefficient is set to 1.5.

[0095] Establish a frequency-resource tradeoff function :

[0096] The calculation formula is ;

[0097] In the formula: =1Hz is the base frequency; =100Hz is the hardware limit frequency; To prevent division by zero, a numerical stability factor is used; the first term is the optimization benefit term, and the second term is the resource penalty term;

[0098] Solving the optimization problem yields : ;

[0099] Its analytical solution is: ;

[0100] When the calculation result is lower At that time, take Higher than At that time, take ;

[0101] (2) Feature extraction algorithm mode Decision calculation:

[0102] First, calculate the complexity index of the current operating condition. :

[0103] ;

[0104] in This represents the standard deviation of the current voltage range. This represents the historical average voltage range. The graph attention coefficients calculated for S3; The mean of the attention coefficient; This is the operating condition identifier; it is 1.5 for fast charging, 1.0 for slow charging, and 0.8 for standby.

[0105] Then, calculate the edge computing load margin. : ;

[0106] In the formula: This represents the actual time spent on feature extraction in the previous cycle. This represents the duration of the current sampling period.

[0107] Establish dynamic decision thresholds and : ;

[0108] ;

[0109] in =0.3 is the base threshold; =0.7 is the base threshold;

[0110] Final decision for:

[0111] ;

[0112] in =1 indicates lightweight statistical mode. =2 indicates a deep time-series mode. =3 indicates the full graph neural network mode;

[0113] (3) Active performance evaluation procedure Trigger determination:

[0114] First, calculate the risk accumulation index. :

[0115] ;

[0116] in =300s is the risk observation window; The attenuation coefficient is used; the integral is approximated by discrete summation: , where K= / Δτ;

[0117] Then, calculate the risk mutation rate. : ;

[0118] Establish aging stage factors :

[0119] ;

[0120] Adjust the trigger threshold: , ,in =0.6× As the cumulative risk threshold, =0.85 is the high-risk threshold;

[0121] The final trigger determination is

[0122] ;

[0123] In the formula: This is an indicator function that takes the value 1 if the condition is met, and 0 otherwise. This is the threshold for the risk mutation rate;

[0124] when Immediately initiate an active performance evaluation procedure, including rapid impedance spectroscopy scanning, analysis of the composition of abnormal gas generated by heat loss, or module disassembly and testing; when and When the value is 0.75, the edge end immediately switches to high-frequency monitoring mode, increasing the sampling frequency from 1Hz to 10Hz, and initiates a fast impedance spectroscopy scan, with the scan frequency range... Extended from [1Hz, 1kHz] to [0.1Hz, 10kHz];

[0125] Finally, a digital twin is built in the cloud. Real-time synchronization of physical battery state; prediction of optimal energy efficiency boundary under different operating conditions through finite element-equivalent circuit coupled simulation; simulation error feedback coefficient. Used to correct model parameters in S4 And the demand weighting coefficient in S5 This forms an optimized closed loop of virtual-real mapping; after the optimization strategy is executed at the edge, the actual resource consumption and optimization benefits are fed back to the cloud, and the weight coefficients are updated using the gradient descent method. Where η=0.01 is the learning rate and e(t) is the feedback error, ensuring that the optimization strategy continues to be optimized as the battery characteristics evolve.

[0126] A distributed sensor network is deployed within the battery pack, including: 96 individual cell voltage sampling points, 24 surface temperature sensors (one for every four cells), 4 strain gauges (arranged at the four corners of the housing), 2 triaxial accelerometers (arranged on the module end plate), and 8 EIS detection points (covering the high and low voltage sampling units). The base sampling frequency is set to fbase = 1 Hz. During vehicle fast charging, when the current Ibus > 200 A is detected, a high-frequency acquisition mode is automatically triggered, increasing the frequency to fhigh = 10 Hz.

[0127] During a 350 kW super-fast charging test, the following typical data were collected: In the single-cell voltage sequence Ui(t), the voltage of the 37th cell, U37, was 3.215 V, while the average voltage Uˉ was 3.412 V, a voltage deviation of −197 mV; the corresponding location temperature, Ts,37, was 42.5°C, exceeding the average temperature Tˉ = 35.8°C by 6.7 K; the mechanical strain ε3 was 285 × 10⁻⁶. −6 Exceeding the historical baseline value by 180×10 −6 .

[0128] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for optimizing the energy efficiency of power batteries based on big data fusion, characterized in that, include: S1, construct a three-dimensional multimodal data acquisition architecture to simultaneously acquire electrical, thermodynamic, mechanical stress and electrochemical impedance spectroscopy data streams; S2, Establish a lightweight feature extraction model for the edge, perform adaptive preprocessing and feature engineering on multimodal data, and generate edge feature vectors; S3 establishes a cloud-based deep energy efficiency state fusion model, calculates risk propagation weights based on graph attention mechanism, and achieves multi-scale spatiotemporal feature fusion; S4. Establish a multi-level energy efficiency status quantitative assessment system to calculate the health status index, energy efficiency degradation probability and remaining service life. S5 constructs a cloud-edge collaborative closed-loop feedback testing mechanism, dynamically adjusts the data collection frequency and optimization strategy, and forms a virtual-real mapping optimization closed loop.

2. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S1 includes: A three-dimensional, multimodal data acquisition architecture is constructed, and a distributed sensor network is deployed within the battery pack to simultaneously acquire electrical parameters, thermodynamic parameters, mechanical stress parameters, and electrochemical impedance spectroscopy data, establishing a multi-source heterogeneous data stream. The electrical parameters include: single-cell voltage data. Bus current data Insulation impedance data Where i represents the cell number, i=1,2,…,N, N is the total number of cells, and t is the sampling time; the thermodynamic parameters include: cell surface temperature data. Module inlet and outlet coolant temperature difference data Ambient temperature The mechanical stress parameters include: battery casing strain. acceleration Where j represents the strain gauge placement location number, and k represents the axial marker of the accelerometer; the electrochemical impedance spectroscopy data is the real impedance obtained through a distributed EIS sensor. With imaginary impedance The subscript re represents the real part of the letter description, the subscript im represents the imaginary part of the letter description, and f represents the sweep frequency point.

3. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S2 includes: A lightweight feature extraction model for edge detection is established, and adaptive preprocessing and feature engineering are performed on the multimodal data collected by S1 to generate edge feature vectors. First, the various types of sensor data are divided into sliding window segments, with the window length denoted as . The sliding step size is denoted as The data matrix within the time window is obtained. Where M is the total number of sensor channels; then, adaptive weighted normalization is used to eliminate dimensional differences, and the normalized data... The calculation formula is: ;in, This represents the raw data of the m-th sensor channel at the l-th sampling point, where m = 1, 2, ..., M, l = 1, 2, ..., ; The average of historical data for the m-th channel; The standard deviation of the historical data for the m-th channel; As the numerical stability factor, take Magnitude; The channel weight coefficient, calculated based on information entropy, characterizes the sensitivity of this physical quantity to energy efficiency status. Its calculation formula is as follows: ;in, Let m be the information entropy of the data in channel m. The probability that the data from this channel falls into the c-th discrete interval, where C is the total number of discrete intervals, is given by the denominator. of To iterate through the index, the index terms of all channels including m itself are summed.

4. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S3 includes: Establish a cloud-based deep energy efficiency state fusion model and receive feature vectors from the edge via a 5G communication link. And combined with the full lifecycle historical data stored in the cloud. To conduct deep situational awareness; the aforementioned full lifecycle historical data Includes: formation stage capacity Number of loops Cumulative fast delivery volume Historical abuse records A multi-scale spatiotemporal graph convolutional network is constructed to model the electrical connections between individual cells within the battery pack as a graph structure G=(V,E), where V is the set of individual battery nodes, |V|=N, and E is the set of edges representing the connections; node feature matrix. Composed of single-unit level features extracted by S2, For node feature dimensions.

5. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S4 includes: A multi-level energy efficiency status quantitative assessment system is established, and the health status index SOH(t) and energy efficiency degradation probability are calculated based on the deep learning output of S3. and remaining useful life RUL(t); the health status index is calculated using the following formula: in, The current available capacity is estimated based on Kalman filtering. The nominal capacity; the energy efficiency degradation probability. Calculated by fusing multi-dimensional anomaly scores: ;in, The electrical anomaly score is calculated based on voltage dispersion and current fluctuation rate. The heat loss anomaly fraction is calculated based on temperature field non-uniformity and temperature rise rate; The fractional mechanical abuse is calculated based on strain amplitude and vibration spectrum characteristics; The electrochemical decay fraction is calculated based on impedance spectral shift. For each dimension's weight coefficients, satisfying .

6. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S5 includes: A closed-loop feedback testing mechanism for cloud-edge collaboration is constructed, with the cloud generating optimization strategy adjustment instructions based on the S4 evaluation results. The data is then distributed to the edge for execution; the optimization strategy includes: adjusting the data acquisition frequency. Switching feature extraction algorithm modes Trigger the active performance evaluation procedure ;when At that time, among them This is the risk threshold.

7. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S5 includes: Optimization strategy adjustment instructions For a three-dimensional decision vector, its mathematical expression is: ,in Let t be the data acquisition frequency at time t. This represents the feature extraction algorithm mode at time t. The trigger flag for the active performance evaluation program at time t; First, the comprehensive testing demand index for cloud computing. The formula for quantifying the current requirement for test accuracy is as follows: ; in, This represents the risk probability after exponential smoothing. ,in The probability of energy efficiency degradation calculated for S4. ∈[0,1] is the smoothing coefficient, which is set to 0.7, and Δt is the sampling time interval; =SOH(t) / 100% represents the normalized health status, and SOH(t) is the health status index calculated by S4; The remaining useful life calculated for S4; Nominal initial life; The demand weighting coefficient satisfies Furthermore, it is configured according to the differences in battery chemistry systems.

8. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S5 includes: Cloud-based evaluation of edge resource constraint functions The available computing and communication resources at the edge are represented by the following formula: ; in This represents the remaining bandwidth of the 5G link at the current moment. Maximum bandwidth; Percentage of idle computing power of edge processors; Backup power supply for the edge gateway; This is the minimum operating power threshold; Based on the above and Calculate the three components of the optimization strategy respectively: (1) Data acquisition frequency Adaptive computation: First, calculate the combined rate of change of the multimodal signals. : In the formula: This represents the normalized data for the m-th channel in S2. S2 represents the channel weighting coefficients calculated for S2; M represents the total number of sensor channels; d / dt is approximated using the difference method and is expressed as... ; Then, calculate the minimum frequency requirement based on the risk profile. : The calculation formula is ; In the formula: coefficient 2 satisfies the Nyquist minimum sampling requirement; These are the sampling coefficients; Establish a frequency-resource tradeoff function : The calculation formula is ; In the formula: Base frequency; This is the hardware's maximum frequency limit; To prevent division by zero, a numerical stability factor is used; the first term is the optimization benefit term, and the second term is the resource penalty term; Solving the optimization problem yields : ; Its analytical solution is: ; When the calculation result is lower At that time, take Higher than At that time, take .

9. The method for optimizing the energy efficiency of power batteries based on big data fusion according to claim 1, characterized in that, S5 includes: Feature extraction algorithm mode Decision calculation: First, calculate the complexity index of the current operating condition. : ； in This represents the standard deviation of the current voltage range. This represents the historical average voltage range. The graph attention coefficients calculated for S3; The mean of the attention coefficient; This is the operating condition identifier; it is 1.5 for fast charging, 1.0 for slow charging, and 0.8 for standby. Then, calculate the edge computing load margin. : ; In the formula: This represents the actual time spent on feature extraction in the previous cycle. This represents the duration of the current sampling period. Establish dynamic decision thresholds and : ; ； in =0.3 is the basic threshold; =0.7 is the base threshold; Final decision for: ； in =1 indicates lightweight statistical mode. =2 indicates a deep time-series mode. =3 indicates the full graph neural network mode; (3) Active performance evaluation procedure Trigger determination: First, calculate the risk accumulation index. : ； in =300s is the risk observation window; The attenuation coefficient is used; the integral is approximated by discrete summation: , where K= / Δτ; Then, calculate the risk mutation rate. : ; Establish aging stage factors : ； Adjust the trigger threshold: , ,in =0.6× As the cumulative risk threshold, =0.85 is the high-risk threshold; The final trigger determination is ； In the formula: This is an indicator function that takes the value 1 if the condition is met, and 0 otherwise. This is the threshold for the risk mutation rate.

10. A computer system, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the power battery energy efficiency optimization method based on big data fusion as described in any one of claims 1 to 9 when executing the executable instructions.

Images