Dual-stream adaptive data compression method and system based on electrochemical state awareness

Abstract

This invention discloses a dual-stream adaptive data compression method and system based on electrochemical state sensing, belonging to the field of battery management data processing. The method collects and preprocesses time-series data from BMS sensors; constructs an adaptive state and prediction module to calculate battery SOC / SOH and identify operating modes, dynamically selecting a prediction model to calculate residuals; applies differentiated error constraints through a fidelity intelligent control module to quantize residuals; matches the optimal entropy encoder based on a state-dependent codebook switching module; and completes compression by encapsulating the metadata stream and residual data stream using a dual-stream encoding master control module. This invention achieves dynamic adaptive processing through electrochemical state sensing, balancing ultra-high compression ratios with high fidelity of key features, reusing existing BMS computing resources to reduce computational power consumption, adapting to edge device constraints, and improving decoding robustness through a dual-stream architecture. It is applicable to various scenarios such as large-scale energy storage and new energy vehicles, and has broad application value.

Description

Technical Field

[0001] This invention belongs to the field of battery management data processing, specifically relating to a dual-stream adaptive data compression method and system based on electrochemical state sensing. Background Technology

[0002] With the widespread application of electrochemical energy storage technology in grid-scale energy storage systems (ESS), battery health management (PHM) throughout the entire battery lifecycle has become crucial. To achieve accurate state of charge (SOC) estimation, state of health (SOH) monitoring, and thermal runaway early warning, modern BMS typically collects operational data from individual battery cells at high sampling rates (e.g., 1Hz to 100Hz). This high-frequency sampling results in massive data generation. For a large-scale energy storage power station, the daily data volume can easily reach the terabyte (TB) level. Such a massive data volume poses a significant challenge to the storage costs of cloud servers. Therefore, efficient compression of BMS data has become an industry necessity. BMS data compression typically involves the integration of multiple technologies, including electrochemical analysis, signal processing, data compression algorithms, and embedded system design. Furthermore, the BMS data compression process encompasses several core challenges, including handling non-stationary entropy, balancing fidelity and compression ratio, and meeting real-time requirements under edge resource constraints. Each of the above challenges requires careful consideration and design in conjunction with the actual operation scenario of the BMS, and strict assurance that the final compression solution achieves a balance between accuracy, efficiency and practicality.

[0003] Currently, existing BMS data compression generally employs two approaches: general lossless compression and globally lossy compression. General lossless compression methods are typically designed for stationary data with uniform entropy distribution. However, when dealing with non-stationary sequences that rapidly switch between multiple modes in a BMS, they face the challenge of drastic entropy fluctuations. The data characteristics differ significantly at each stage, and relying on a fixed compression strategy results in poor adaptability, low compression ratios, or increased distortion during dynamic phases. Globally lossy compression methods compress the entire dataset using a uniform error threshold (such as global RMSE), typically based on global rules such as polynomial fitting and uniform quantization. Compression based on global rules can lead to the smoothing or distortion of key local electrochemical features. Furthermore, BMS data compression needs to balance high fidelity and high compression ratio, but existing globally lossy compression methods have rigid error control modes and lack local adaptability. Therefore, a more efficient compression method that can adapt to data patterns and ensure the integrity of local features is needed. Summary of the Invention

[0004] To address these issues, this invention provides a dual-stream adaptive data compression method and system based on electrochemical state awareness, which solves the technical problems in existing BMS data compression, such as poor adaptability of general lossless compression, easy loss of key electrochemical features in global lossy compression, difficulty in achieving both high compression ratio and high fidelity, and adaptation to the computing power constraints of embedded devices.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dual-stream adaptive data compression method based on electrochemical state sensing, comprising the following steps:

[0006] Collect the sensor time-series data stream of the battery management system. The sensor time-series data stream includes voltage, current and temperature data. Perform noise reduction filtering, time synchronization alignment, outlier handling and circular buffer storage operations on the sensor time-series data stream to obtain the preprocessed data source.

[0007] An adaptive state and prediction module is constructed based on the preprocessed data source. The state of charge and health of the battery are calculated by the state estimation algorithm of the adaptive state and prediction module. The real-time electrochemical operation mode of the battery is identified by combining the current, voltage change rate and filter output. The optimal prediction model is dynamically selected according to the identified operation mode to generate the prediction value, and the prediction residual between the original data and the prediction value is calculated.

[0008] A fidelity intelligent control module is constructed. Based on the real-time electrochemical operation mode identified by the adaptive state and prediction module, and combined with the fidelity requirements of electrochemical diagnosis for specified features, a differentiated local error constraint is dynamically applied, and the prediction residual is adaptively quantized to obtain a quantized residual that meets the fidelity requirements.

[0009] A state-dependent codebook switching module is constructed to analyze the statistical characteristics of the quantization residuals corresponding to the real-time electrochemical operation mode, and to dynamically match and call the appropriate entropy encoder for encoding processing of the quantization residuals.

[0010] A dual-stream encoding execution control module is constructed to encapsulate the encoded result into a synchronous metadata stream and a residual data stream. The metadata stream includes the context information required for decoding, and the residual data stream includes the compressed residual bit stream, thus completing data compression.

[0011] As a preferred embodiment of the dual-stream adaptive data compression method based on electrochemical state sensing, the denoising filtering for the voltage signal employs a sliding window-based median filtering combined with wavelet thresholding. The threshold value for wavelet thresholding is... The formula is:

[0012]

[0013] In the formula, The standard deviation of noise. The signal length;

[0014] The size of the circular buffer is:

[0015]

[0016] In the formula, Sampling frequency, This represents the maximum expected delay.

[0017] As a preferred embodiment of the dual-current adaptive data compression method based on electrochemical state sensing, the adaptive state and prediction module runs an adaptive nonlinear Kalman filter algorithm to calculate the battery's state of charge (SOC) and state of health (SOH), and identifies the battery's real-time electrochemical operating mode based on current, voltage change rate, and filter output. The real-time electrochemical operation mode The classification logic is as follows:

[0018]

[0019] In the formula, For current threshold, For the sliding window variance, For the target voltage, For voltage threshold, For a moment The current, For a moment voltage, For a moment The electrochemical operating mode, The current stability variance threshold. This represents the time interval of the sliding window.

[0020] As a preferred scheme of the dual-stream adaptive data compression method based on electrochemical state sensing, the selection rule for the optimal prediction model in the process of dynamically selecting the optimal prediction model to generate predicted values ​​according to the identified operating mode is as follows:

[0021] When real-time electrochemical operation mode In static mode, zero-order hold prediction is used, and the predicted value is... In the formula, For a moment The predicted value, For a moment The original data;

[0022] When real-time electrochemical operation mode In constant current CC / constant voltage CV mode, second-order difference linear extrapolation is used to predict the value. In the formula, For a moment The original data;

[0023] When real-time electrochemical operation mode In the pulse mode, an AUKF-based model is used for prediction, and the predicted value is... In the formula, For time-based Raw data and time Current The AUKF prior estimate;

[0024] Predicted residuals In the formula, For a moment The original data.

[0025] As a preferred embodiment of the electrochemical state-sensing-based dual-stream adaptive data compression method, the local error constraint The setting logic is as follows:

[0026] Preset standard error limits and critical error limits The dynamic error limit is defined when the following conditions are met. ,otherwise :

[0027]

[0028] In the formula, This is a critical window condition for DVA. For the state of charge calculated by AUKF, For a predefined SOC range that includes the DVA peak;

[0029] The step size for adaptive quantization is:

[0030]

[0031] The quantization residual is:

[0032]

[0033] In the formula, For a moment The original residuals, This is the rounding function;

[0034] Reconstructing residuals And satisfy In the formula, For a moment The reconstruction residual.

[0035] As a preferred embodiment of the electrochemical state-sensing-based dual-stream adaptive data compression method, the switching rule of the entropy encoder is as follows:

[0036] When real-time electrochemical operation mode In static mode, run-length encoding (RLE) is selected, and the encoding logic is to count the number of consecutive zeros in the quantized residuals. If a non-zero value is encountered Then output The encoding formula is:

[0037]

[0038] In the formula, The encoding result of the residual data stream, For a moment to Quantized residual sequence;

[0039] When real-time electrochemical operation mode When using constant current (CC) / constant voltage (CV) mode, choose either Adaptive Huffman Coding (AHE) or Golomb-Rice Coding. The Golomb mapping formula is:

[0040]

[0041] In the formula, For quotient, For Rice parameters, The remainder;

[0042] When real-time electrochemical operation mode In pulse mode, select either context-adaptive binary arithmetic coding (CABAC) or Exp-Golomb coding.

[0043] As a preferred embodiment of the dual-stream adaptive data compression method based on electrochemical state awareness, the metadata stream includes the context information required for decoding, adopts Lempel-Ziv or fixed-length encoding, and the data packet structure includes a Header and a BlockPayload.

[0044] The header includes a timestamp. Initial value and In the formula, This is the start timestamp of the data. The original data at the initial moment, This refers to the original data from the time step preceding the initial time step;

[0045] The block payload includes each state block. Len and In the formula, Here, Len is the status label, and Len is the number of data points that this status lasts. Index for the error limits used;

[0046] The residual data stream includes the compressed residual bit stream, and the two streams are synchronously stored and transmitted through index files and data files.

[0047] The present invention also provides a dual-stream adaptive data compression system based on electrochemical state sensing, comprising:

[0048] The data preprocessing module is used to collect the sensor time-series data stream of the battery management system. The sensor time-series data stream includes voltage, current and temperature data. The module performs noise reduction filtering, time synchronization alignment, outlier handling and circular buffer storage operations on the sensor time-series data stream to obtain the preprocessed data source.

[0049] The adaptive state and prediction module is used to run the state estimation algorithm to solve the battery's state of charge and health state, and to identify the real-time electrochemical operation mode of the battery by combining current, voltage change rate and filter output. Then, based on the identified operation mode, the optimal prediction model is dynamically selected to generate predicted values, and the prediction residual between the original data and the predicted values ​​is calculated.

[0050] The fidelity intelligent control module is used to dynamically apply differentiated local error constraints based on the real-time electrochemical operation mode identified by the adaptive state and prediction module, combined with the fidelity requirements of electrochemical diagnosis for specified features, and to perform adaptive quantization processing on the prediction residual to obtain a quantized residual that meets the fidelity requirements.

[0051] The state-dependent codebook switching module is used to analyze the statistical characteristics of the quantization residual corresponding to the real-time electrochemical operation mode, dynamically match the quantization residual, and call the appropriate entropy encoder for encoding processing.

[0052] The dual-stream encoding execution control module is used to encapsulate the encoded result into a synchronous metadata stream and a residual data stream. The metadata stream includes the context information required for decoding, and the residual data stream includes the compressed residual bit stream, thus completing the data compression.

[0053] As a preferred embodiment of the electrochemical state-sensing-based dual-stream adaptive data compression system, the data preprocessing module includes:

[0054] The denoising filter for the voltage signal employs a sliding window-based median filter combined with wavelet thresholding. The threshold value for wavelet thresholding is... The formula is:

[0055]

[0056] In the formula, The standard deviation of noise. The signal length;

[0057] The size of the circular buffer is:

[0058]

[0059] In the formula, Sampling frequency, This represents the maximum expected delay.

[0060] As a preferred embodiment of a dual-stream adaptive data compression system based on electrochemical state sensing, the adaptive state and prediction module uses an adaptive nonlinear Kalman filter algorithm to calculate the battery's state of charge (SOC) and state of health (SOH), and identifies the battery's real-time electrochemical operating mode based on current, voltage change rate, and filter output. The real-time electrochemical operation mode The classification logic is as follows:

[0061]

[0062] In the formula, For current threshold, For the sliding window variance, For the target voltage, For voltage threshold, For a moment The current, For a moment voltage, For a moment The electrochemical operating mode, The current stability variance threshold. This represents the time interval of the sliding window.

[0063] As a preferred embodiment of the dual-stream adaptive data compression system based on electrochemical state sensing, the selection rule for the optimal prediction model in the adaptive state and prediction module is as follows:

[0064] When real-time electrochemical operation mode In static mode, zero-order hold prediction is used, and the predicted value is... In the formula, For a moment The predicted value, For a moment The original data;

[0065] When real-time electrochemical operation mode In constant current CC / constant voltage CV mode, second-order difference linear extrapolation is used to predict the value. In the formula, For a moment The original data;

[0066] When real-time electrochemical operation mode In the pulse mode, an AUKF-based model is used for prediction, and the predicted value is... In the formula, For time-based Raw data and time Current The AUKF prior estimate;

[0067] Predicted residuals In the formula, For a moment The original data.

[0068] As a preferred embodiment of a dual-stream adaptive data compression system based on electrochemical state sensing, the local error constraint in the fidelity intelligent control module... The setting logic is as follows:

[0069] Preset standard error limits and critical error limits The dynamic error limit is defined when the following conditions are met. ,otherwise :

[0070]

[0071] In the formula, This is a critical window condition for DVA. For the state of charge calculated by AUKF, For a predefined SOC range that includes the DVA peak;

[0072] The step size for adaptive quantization is:

[0073]

[0074] The quantization residual is:

[0075]

[0076] In the formula, For a moment The original residuals, This is the rounding function;

[0077] Reconstructing residuals And satisfy In the formula, For a moment The reconstruction residual.

[0078] As a preferred embodiment of the electrochemical state-aware dual-stream adaptive data compression system, the switching rule of the entropy encoder in the state-dependent codebook switching module is as follows:

[0079] When real-time electrochemical operation mode In static mode, run-length encoding (RLE) is selected, and the encoding logic is to count the number of consecutive zeros in the quantized residuals. If a non-zero value is encountered Then output The encoding formula is:

[0080]

[0081] In the formula, The encoding result of the residual data stream, For a moment to Quantized residual sequence;

[0082] When real-time electrochemical operation mode When using constant current (CC) / constant voltage (CV) mode, choose either Adaptive Huffman Coding (AHE) or Golomb-Rice Coding. The Golomb mapping formula is:

[0083]

[0084] In the formula, For quotient, For Rice parameters, The remainder;

[0085] When real-time electrochemical operation mode In pulse mode, select either context-adaptive binary arithmetic coding (CABAC) or Exp-Golomb coding.

[0086] As a preferred embodiment of the electrochemical state-aware dual-stream adaptive data compression system, the dual-stream encoding execution control module includes:

[0087] The metadata stream includes the context information required for decoding, using Lempel-Ziv or fixed-length encoding, and the data packet structure includes a Header and a Block Payload.

[0088] The header includes a timestamp. Initial value and In the formula, This is the start timestamp of the data. The original data at the initial moment, This refers to the original data from the time step preceding the initial time step;

[0089] The block payload includes each state block. Len and In the formula, Here, Len is the status label, and Len is the number of data points that this status lasts. Index for the error limits used;

[0090] The residual data stream includes the compressed residual bit stream, and the two streams are synchronously stored and transmitted through index files and data files.

[0091] The present invention has the following advantages:

[0092] First, this invention uses a state-dependent codebook switching mechanism to dynamically match the optimal entropy encoder according to different operating modes such as battery rest, constant current, and pulse, breaking through the bottleneck of general compression algorithms and achieving a compression ratio far higher than that of the traditional LZW / SDT algorithm.

[0093] Secondly, this invention utilizes state-adjusted dynamic error limit technology to intelligently identify key electrochemical windows such as the DVA feature region and automatically switch to microvolt-level ultra-low error coding, avoiding the loss of key diagnostic features and ensuring the accuracy of battery state of health (SOH) assessment.

[0094] Third, this invention deeply integrates state estimation and data compression, directly reuses the calculation results of the BMS built-in AUKF algorithm, and does not require running an additional complex feature extraction model, which greatly reduces the computational load and power consumption of the embedded MCU.

[0095] Fourth, by eliminating high-frequency redundant data and residual quantization, this invention significantly reduces data volume, lowers the Flash storage occupancy rate at the edge, reduces 4G / 5G network transmission traffic costs, and saves operating costs for large-scale energy storage power stations.

[0096] Fifth, the present invention adopts a dual-stream separation transmission design, with the metadata stream independently carrying key context information. Even if some data packets are lost, the decoding end can still rely on the metadata to reconstruct the basic data framework, avoiding the risk of "one mistake, the whole system is messed up" in traditional single-stream compression.

[0097] Sixth, this invention endows edge devices with the ability to "understand" data. By performing physical sensing preprocessing and compression at the edge, the cloud receives high-quality data that has been cleaned and retains its complete features, greatly reducing the computing power pressure on cloud servers for data cleaning and feature extraction, and improving the overall response speed of the digital twin system.

[0098] Seventh, this invention is specifically optimized for embedded environments and employs pipelined processing logic. Compared to compression methods based on deep learning (RNN / Transformer), SC-DSE does not require massive inference matrix operations and can ensure that all compression tasks are completed within a 50ms control cycle without affecting the real-time protection function of the BMS.

[0099] Eighth, this invention has the ability to self-adjust as the battery ages. As the battery's state of equilibrium (SOH) declines, its internal resistance and capacity change. The reused state estimation automatically updates the internal parameters, thereby driving the compression model to automatically adapt to the voltage curve characteristics after aging, without the need for manual recalibration or updating of the algorithm model.

[0100] Ninth, this invention supports multi-level precision decoding. The cloud can quickly browse the operation overview through the metadata stream alone, or reconstruct high-precision waveforms by combining the residual data stream, thereby improving the customization capabilities of data analysis and visualization.

[0101] Tenth, this invention is based on a universal battery physics model. Whether it is a ternary lithium (NMC), lithium iron phosphate (LFP), or sodium-ion battery, as long as the basic state classification threshold is adjusted, the workload of repeated development for different battery chemical systems is greatly reduced. Attached Figure Description

[0102] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0103] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0104] Figure 1 This is a schematic diagram of the process of the dual-stream adaptive data compression method based on electrochemical state sensing provided in an embodiment of the present invention;

[0105] Figure 2 This is a technical framework diagram of the dual-stream adaptive data compression method based on electrochemical state sensing provided in the embodiments of the present invention;

[0106] Figure 3This is a schematic diagram of the dual-stream adaptive data compression system architecture based on electrochemical state awareness provided in an embodiment of the present invention. Detailed Implementation

[0107] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0108] Example 1

[0109] See Figure 1 and Figure 2 Embodiment 1 of the present invention provides a dual-stream adaptive data compression method based on electrochemical state sensing, comprising the following steps:

[0110] S1. Acquire the sensor time-series data stream from the battery management system (BMS). This data stream includes voltage, current, and temperature data. Perform denoising filtering, time synchronization alignment, outlier handling, and circular buffer storage operations on the sensor time-series data stream to obtain a preprocessed data source. The raw sensor data acquired by the BMS contains high-frequency noise, timestamp deviations, and abnormal fluctuations. Directly using this data for compression would lead to decreased compression efficiency and compromised fidelity. Denoising filtering removes useless noise while retaining key electrochemical characteristics; time synchronization alignment ensures a one-to-one correspondence between voltage, current, and temperature data in the time dimension; outlier handling avoids interference from extreme data on modeling; and circular buffer storage provides a low-latency, highly consistent data cache for real-time calculations.

[0111] S2. An adaptive state and prediction module is constructed based on the preprocessed data source. This module uses a state estimation algorithm to calculate the battery's state of charge (SOC) and health status. It then identifies the battery's real-time electrochemical operating mode by combining current, voltage change rate, and filter output. Based on the identified operating mode, the optimal prediction model is dynamically selected to generate predicted values, and the prediction residual between the original data and the predicted values ​​is calculated. The statistical characteristics of battery data differ significantly under different operating modes (static, constant current, constant voltage, pulse), and a fixed prediction model cannot adapt to all scenarios. The state estimation algorithm can calculate the internal states of SOC and SOH, providing underlying data support for operating mode identification. Through multi-dimensional indicators such as current, voltage change rate, and filter output, accurate classification of the battery's real-time operating mode can be achieved. Dynamically selecting a matching prediction model for different modes minimizes prediction bias, concentrating the prediction residual within a smaller range and reducing the amount of data for subsequent encoding.

[0112] S3. Construct a fidelity intelligent control module. Based on the real-time electrochemical operating mode identified by the adaptive state and prediction module, and combined with the fidelity requirements of electrochemical diagnosis for specified features, dynamically apply differentiated local error constraints, and adaptively quantize the predicted residuals to obtain quantized residuals that meet the fidelity requirements. BMS data compression needs to balance high compression ratios with the fidelity of key features, and fidelity requirements vary in different scenarios. The core of the fidelity intelligent control module is to establish a mapping relationship of "mode-requirement-error constraint". In critical electrochemical windows such as the DVA feature region, battery health diagnosis has extremely high requirements for data accuracy, and ultra-low error constraints need to be applied to avoid the loss of key features; in non-critical intervals, error limits can be relaxed to pursue the ultimate compression ratio. Through adaptive quantization, the residuals are mapped to discrete quantized values, ensuring that the error after decoding does not exceed the preset threshold, and reducing data redundancy through quantization.

[0113] S4. Construct a state-dependent codebook switching module to analyze the statistical characteristics of the quantization residuals corresponding to the real-time electrochemical operating modes, dynamically match the quantization residuals, and call the appropriate entropy encoder for encoding processing. The quantization residuals under different operating modes have different statistical characteristics. For example, in the static mode, the residuals are mostly zero; in the constant current / constant voltage mode, the residuals exhibit a low variance distribution; and in the pulse mode, the residuals exhibit a high entropy distribution. A single entropy encoder cannot adapt to all statistical characteristics. The state-dependent codebook switching module analyzes the residual statistical characteristics in real time and matches the optimal entropy encoder for different scenarios: RLE encoding is used for residuals with a large number of zero values; adaptive Huffman encoding is used for low-variance small integer residuals; and CABAC encoding is used for high-entropy residuals. This makes the encoding process approach the signal entropy limit, maximizing compression efficiency while avoiding the efficiency bottleneck of general encoding algorithms in specific scenarios.

[0114] S5. Construct a dual-stream encoding execution control module to encapsulate the encoded result into a synchronous metadata stream and a residual data stream. The metadata stream includes the context information required for decoding, and the residual data stream includes the compressed residual bit stream, completing data compression. Traditional single-stream encoding carries the risk of "one error, the whole system is messed up," and it is difficult to balance decoding reliability and compression efficiency. The metadata stream carries the context information necessary for decoding, such as initial values, status labels, and error limits, and uses lossless or low compression ratio encoding to ensure reliability. Even if part of the residual data stream is lost, the decoding end can still reconstruct the basic data framework based on the metadata. The residual data stream focuses on carrying the compressed residual bit stream and uses efficient encoding to pursue a high compression ratio. Dual-stream synchronous storage and transmission not only ensures the robustness of decoding but also minimizes data volume, meeting the dual needs of edge storage and remote transmission.

[0115] In this embodiment, in step S1, the denoising filter for the voltage signal employs a median filter based on a sliding window combined with wavelet thresholding. The threshold value for wavelet thresholding is... The formula is:

[0116]

[0117] In the formula, The standard deviation of noise. The signal length is given. Noise in voltage signals mainly includes high-frequency random noise and impulse noise. A single denoising method struggles to balance denoising effectiveness with feature preservation. Median filtering effectively suppresses impulse noise while protecting signal edge features. Wavelet thresholding denoising decomposes the signal into wavelet coefficients of different scales and applies a threshold to noisy wavelet coefficients (coefficients smaller than the threshold are set to zero, and coefficients larger than the threshold are contracted), thus eliminating high-frequency noise. The universal threshold formula is an optimal threshold derived from statistical theory, balancing denoising effectiveness and signal fidelity while minimizing mean square error, ensuring that key electrochemical characteristics such as the DVA peak in the voltage signal are not compromised.

[0118] The size of the circular buffer is:

[0119]

[0120] In the formula, Sampling frequency, The maximum expected latency. A circular buffer is a first-in, first-out (FIFO) cache structure suitable for real-time data processing scenarios. Its size design must meet the requirement of caching all sampled data within the maximum expected latency: sampling frequency. Determines the amount of data per unit of time, and the maximum expected latency. This is the longest data waiting time allowed by the system; the product of these two values ​​is the minimum required buffer capacity. This design ensures that subsequent modules such as state estimation and pattern recognition can obtain continuous and complete time-series data, avoiding data loss. It also reduces memory usage through a fixed capacity limit, adapting to the resource constraints of embedded BMS, while supporting cyclic data overwriting to guarantee the real-time performance of cached data.

[0121] In this embodiment, in step S2, the adaptive nonlinear Kalman filter algorithm is run through the adaptive state and prediction module to calculate the battery's state of charge (SOC) and state of health (SOH), and the real-time electrochemical operating mode of the battery is identified based on the current, voltage change rate, and filter output. The real-time electrochemical operation mode The classification logic is as follows:

[0122]

[0123] In the formula, For current threshold, For the sliding window variance, For the target voltage, For voltage threshold, For a moment The current, For a moment voltage, For a moment The electrochemical operating mode, The current stability variance threshold. This represents the time interval of the sliding window.

[0124] Specifically, Adaptive Nonlinear Kalman Filtering (AUKF) is an optimal estimation method for nonlinear systems. Through unscented transformation, it processes nonlinear state and observation equations, solving for internal states such as battery SOC and SOH, and updating the noise covariance matrix in real time. The operating mode classification logic is designed based on the physical operating laws of the battery. The static mode is characterized by current approaching zero; the constant current mode is characterized by an absolute current value greater than a threshold and small fluctuations (sliding window variance less than the stability threshold); the constant voltage mode requires that "voltage is close to the target value" and "the previous state was constant current or constant voltage" (to avoid misjudgment); the pulse mode covers dynamic scenarios beyond the above three modes. The combination of multi-dimensional indicators ensures the accuracy of mode recognition, providing a reliable basis for subsequent prediction model selection, error constraint setting, and encoder switching.

[0125] In this embodiment, during step S2, when dynamically selecting the optimal prediction model to generate predicted values ​​based on the identified operating mode, the selection rule for the optimal prediction model is as follows:

[0126] When real-time electrochemical operation mode In static mode, zero-order hold prediction is used, and the predicted value is... In the formula, For a moment The predicted value, For a moment The original data;

[0127] When real-time electrochemical operation mode In constant current CC / constant voltage CV mode, second-order difference linear extrapolation is used to predict the value. In the formula, For a moment The original data;

[0128] When real-time electrochemical operation mode In the pulse mode, an AUKF-based model is used for prediction, and the predicted value is... In the formula, For time-based Raw data and time Current The AUKF prior estimate.

[0129] Specifically, in static mode, the battery has no charging or discharging current, and the voltage signal tends to be stable with minimal changes. Zero-order hold prediction directly uses historical data as the predicted value for the current moment. This model has a simple structure and low computational cost, minimizing the computational power consumption of embedded devices while ensuring prediction accuracy. Furthermore, the prediction residual is close to zero, creating conditions for efficient coding. In constant current / constant voltage mode, the battery voltage exhibits an approximately linear trend over time. Second-order differential linear extrapolation, based on the assumption that the rate of change of data between adjacent moments remains stable, derives the predicted value for the current moment from the original data of the previous two moments. This model can capture linear change patterns with small prediction errors; simultaneously, the model only relies on historical time-series data, without introducing additional complex parameters, resulting in high computational efficiency and adaptability to real-time compression requirements. In pulse mode, the battery current changes rapidly, and the voltage exhibits a nonlinear transient response, making accurate prediction difficult with simple linear models. The AUKF-based model prediction utilizes the prior estimation results of the AUKF algorithm. This estimation integrates the physical characteristics of the battery equivalent circuit model and real-time current input information, and can capture the nonlinear dynamic changes of voltage. Compared with the linear prediction model, it significantly improves the prediction accuracy, ensures that the residual in the pulse mode remains within a small range, and avoids the reduction in compression efficiency due to excessive prediction deviation.

[0130] Among them, the predicted residual In the formula, For a moment The original data is used for prediction. The prediction residual is the difference between the original data and the predicted value. Its function is to remove redundant information (i.e., the part that the prediction model can interpret) from the original data, retaining only the unpredictable fluctuation components. The numerical range of the residual is much smaller than that of the original data, and after prediction, the residual exhibits a concentrated distribution characteristic. Through quantization and entropy coding, the data volume can be significantly compressed. At the same time, the magnitude of the residual directly reflects the adaptation effect of the prediction model, providing an indirect reference for fidelity control and encoder selection.

[0131] In this embodiment, in step S3, the local error constraint The setting logic is as follows:

[0132] Preset standard error limits and critical error limits The dynamic error limit is defined when the following conditions are met. ,otherwise :

[0133]

[0134] In the formula, This is a critical window condition for DVA. For the state of charge calculated by AUKF, This is a predefined SOC range that includes the DVA peak.

[0135] Specifically, the peak value of the differential voltage analysis (DVA) is a core characteristic for evaluating the state of harmonics (SOH) of a battery, and its corresponding state of charge (SOC) range is a critical electrochemical window, requiring ultra-high data fidelity. The logic for setting local error constraints is based on scenario-demand matching: when the battery is in constant current mode and the SOC falls within the critical DVA range, the critical error limit is triggered. This ensures that the decoded data can accurately reproduce the DVA peak value; in other scenarios, standard error limits are used. The compression ratio is maximized within an acceptable error range to achieve a dynamic balance between fidelity and compression ratio.

[0136] The step size for adaptive quantization is:

[0137]

[0138] The quantization residual is:

[0139]

[0140] In the formula, For a moment The original residuals, This is the rounding function.

[0141] Specifically, the quantization step size ensures that the decoded error does not exceed a preset error limit. Since the quantization process maps continuous residuals to discrete integers, the quantization step size... At that time, the maximum error of the quantization value corresponding to any continuous residual does not exceed (i.e., the residual falls on) When within the interval, the quantization error is ≤ The quantization process strictly meets the preset fidelity requirements. Simultaneously, the quantization step size adaptively adjusts according to the dynamic error limit, using a small step size in critical windows to ensure accuracy and a large step size in non-critical windows to reduce the number of quantized values, balancing accuracy and compression efficiency. The quantization residual is the discrete value obtained after normalization and rounding of the original residual. It transforms the continuous residual signal into a discrete integer sequence, providing a suitable input for entropy coding. By dividing the original residual by the quantization step size, the residual is normalized, and then rounded to obtain the integer quantized value. This process ensures the controllability of quantization error and reduces the range of data values ​​through discretization, enabling subsequent entropy coding to compress data more efficiently and reduce storage and transmission costs.

[0142] Among them, reconstruction residual And satisfy In the formula, For a moment The reconstruction residual.

[0143] Specifically, the reconstructed residual is the result of inverse quantization of the quantized residual, used to approximate the original residual during the decoding stage. The inverse quantization process is achieved by multiplying the quantized residual by the quantization step size. Since the error range is strictly controlled during quantization, the difference between the reconstructed residual and the original residual does not exceed a preset error limit. This ensures that the decoded data meets the accuracy requirements of electrochemical diagnosis; at the same time, the calculation logic for reconstructing the residual is simple, and the decoding end does not require complex calculations, adapting to the fast processing needs of both the edge end and the decoding end.

[0144] In this embodiment, in step S4, the switching rule of the entropy encoder is:

[0145] When real-time electrochemical operation mode In static mode, run-length encoding (RLE) is selected, and the encoding logic is to count the number of consecutive zeros in the quantized residuals. If a non-zero value is encountered Then output The encoding formula is:

[0146]

[0147] In the formula, The encoding result of the residual data stream, For a moment to The quantized residual sequence is in a static mode where the vast majority of the quantized residuals are 0, with only a small number of non-zero values, exhibiting the characteristics of "long zero sequences and sparse non-zero values". Run-length encoding (RLE) uses the number of consecutive zeros and combinations of non-zero values ​​to represent the data sequence, eliminating the need to repeatedly store a large number of zero values ​​and significantly reducing data volume. For example, the sequence "0,0,0,3,0,0,5" can be encoded as "<3,3>、<2,5>", demonstrating extremely high encoding efficiency and perfectly adapting to the residual data characteristics in static mode.

[0148] When real-time electrochemical operation mode When using constant current (CC) / constant voltage (CV) mode, choose either Adaptive Huffman Coding (AHE) or Golomb-Rice Coding. The Golomb mapping formula is:

[0149]

[0150] In the formula, For quotient, For Rice parameters, The remainder is the quotient. In constant current / constant voltage mode, the quantized residuals follow a Laplace distribution, concentrated on small integers near 0, and the data exhibits the characteristics of "dense small values ​​and sparse large values". Adaptive Huffman coding (AHE) dynamically constructs a Huffman tree, assigning short codes to frequently occurring small values ​​and long codes to infrequently occurring large values, approximating the entropy limit of the data; Golomb-Rice coding decomposes the quantized residuals into a quotient q and a remainder r, using variable-length codes for the quotient and fixed-length codes for the remainder, adapting to the dense distribution of small integers. Both coding methods can efficiently compress the residual data in this mode, with low computational complexity, making them suitable for real-time processing.

[0151] When real-time electrochemical operation mode In impulse mode, either context-adaptive binary arithmetic coding (CABAC) or Exp-Golomb coding is selected. In impulse mode, the quantization residuals are dispersed, containing many large values, resulting in high data entropy, making efficient compression difficult with simple entropy coding. Context-adaptive binary arithmetic coding (CABAC) dynamically adjusts coding parameters using context information, adapting to changes in the statistical characteristics of the data, achieving coding efficiency close to the entropy limit, making it particularly suitable for high-entropy data. Exp-Golomb coding uses exponentially variable-length coding for large-value data, effectively reducing the coding length of large-value data. Both coding methods can efficiently handle high-entropy residuals in impulse mode, ensuring compression efficiency.

[0152] In this embodiment, in step S5, the metadata stream includes the context information required for decoding, uses Lempel-Ziv or fixed-length encoding, and the data packet structure includes a Header and a Block Payload.

[0153] The header includes a timestamp. Initial value and In the formula, This is the start timestamp of the data. The original data at the initial moment, This refers to the original data from the time step preceding the initial time step;

[0154] The block payload includes each state block. Len and In the formula, Here, Len is the status label, and Len is the number of data points that this status lasts. Index for the error limits used;

[0155] The residual data stream includes the compressed residual bit stream, and the two streams are synchronously stored and transmitted through index files and data files.

[0156] Specifically, the function of the metadata stream is to provide the necessary context information for decoding. Its reliability directly determines whether decoding can proceed normally, therefore, low compression ratio or lossless encoding is required. Lempel-Ziv encoding can achieve a certain degree of compression while ensuring losslessness, saving storage and transmission resources; fixed-length encoding has simple logic and fast decoding speed, adapting to the fast decoding needs of edge devices. The data packet structure is divided into Header and Block Payload. The Header stores global context information (such as start timestamp and initial data), and the Block Payload stores segmented state information (such as the mode, length, and error limits of each state block). The hierarchical structure facilitates fast indexing during decoding and can adapt to dynamic parameter adjustments in state switching scenarios.

[0157] Among them, timestamp Used for time alignment of decoded data to ensure that the reconstructed data is consistent with the time dimension of the original data, providing a time reference for time series analysis and electrochemical diagnostics; initial values and This is a necessary initial input for the decoding phase to execute the prediction model. Without this initial value, the decoder cannot reconstruct the predicted value, and therefore cannot rebuild the original data. Together, they ensure the integrity and accuracy of the decoding. Status Label Used to inform the decoder of the battery operating mode corresponding to the current state block, so that the decoder can switch the matching prediction model and entropy decoder; Len is used to identify the data length of each state block, which facilitates the decoder to segment the data stream and achieve segmented decoding; Error Bounds Index Corresponding to the local error constraints during encoding, the decoder determines the inverse quantization step size, ensuring that the error of the reconstructed residual is controllable. The combination of these three elements enables the decoder to accurately reproduce the processing logic during encoding, guaranteeing the fidelity of the decoded data.

[0158] The residual data stream is the compressed data carrier, storing only the entropy-encoded residual bit stream to maximize the compression ratio. The dual-stream synchronization mechanism is implemented through an "index file (metadata stream) and a data file (residual data stream)". The metadata stream serves as an index, recording information such as the position, length, and parameters of each state block. During decoding, the metadata stream is read first to obtain the context, and then the corresponding segment of the residual bit stream is extracted from the residual data stream according to the index for decoding, ensuring accurate matching between the two data streams. Synchronous storage and transmission not only avoid the poor fault tolerance problem of single-stream encoding, but also optimize data management through separate storage, improving the overall flexibility of the system.

[0159] The application scenarios of this invention are as follows:

[0160] Scenario 1: Grid-scale large-scale energy storage power station (ESS):

[0161] Grid-scale energy storage power stations require the connection of hundreds or thousands of individual battery cells, collecting voltage, current, and temperature data at high sampling rates of 1Hz-100Hz, generating terabytes of data daily. This invention reduces cloud storage costs and 4G / 5G transmission traffic fees through ultra-high compression ratios, while ensuring high fidelity of key characteristics such as peak DVA. It supports battery lifecycle health management (PHM), accurate SOC / SOH estimation, and thermal runaway early warning, adapting to the long-term stable operation requirements of power stations. The high robustness of the dual-stream architecture can cope with data transmission packet loss problems in the complex network environment of power stations, and low-latency processing at the edge does not affect the real-time protection function of the BMS.

[0162] Scenario 2: Power Battery Management for New Energy Vehicles

[0163] New energy vehicle power batteries frequently switch between multiple modes, including driving, charging, and stationary operation. The Battery Management System (BMS) needs to process high-frequency data in real time and transmit it to the vehicle terminal or cloud platform. This invention can dynamically adapt to different operating modes, such as CC / CV mode during charging, pulse mode during driving, and stationary mode after parking. While saving onboard Flash storage resources, it retains key battery health diagnostic data, assisting cloud-based analysis of battery degradation patterns and optimization of charging strategies. Its low computational consumption characteristic adapts to the resource constraints of onboard embedded MCUs, and its full lifecycle adaptive capability can cope with changes in data characteristics caused by battery aging.

[0164] Scenario 3: Distributed Energy Storage System (Residential / Commercial / Industrial Energy Storage):

[0165] Distributed energy storage systems, such as residential photovoltaic energy storage and industrial / commercial peak-valley arbitrage energy storage, typically employ an architecture of local storage at the edge and remote monitoring in the cloud. However, these systems are limited by installation space and maintenance costs, and the computing power and storage resources of edge devices are also limited. This invention reduces local storage usage and remote transmission bandwidth requirements through efficient data compression, allowing maintenance personnel to quickly obtain battery operation overviews (by parsing only the metadata stream) or view high-precision data (combined with residual data streams) via the cloud. Its broad electrochemical compatibility allows it to adapt to different battery types, such as ternary lithium and lithium iron phosphate, reducing the workload of system adaptation development.

[0166] Scenario 4: Battery Testing and R&D Scenarios

[0167] During the battery R&D and production testing phase, long-term, high-frequency sampling tests are required for individual battery cells or modules, such as cyclic charge-discharge tests and high / low temperature environment tests, generating massive amounts of time-series data for analyzing battery performance and optimizing battery design. This invention can compress data in real time during testing, significantly reducing the storage pressure on testing equipment. Simultaneously, by ensuring the fidelity of key features at the microvolt level, it ensures that the test data accurately reproduces the battery's voltage plateau, polarization response, and other core characteristics, providing reliable data support for battery material improvement and structural optimization. The multi-level precision decoding function can meet the diverse needs of R&D personnel for quickly browsing data trends or deeply analyzing detailed features.

[0168] It should be noted that the method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0169] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0170] Example 2

[0171] See Figure 3 Embodiment 2 of the present invention also provides a dual-stream adaptive data compression system based on electrochemical state sensing, comprising:

[0172] The data preprocessing module 100 is used to collect the sensor time-series data stream of the battery management system. The sensor time-series data stream includes voltage, current and temperature data. The module performs noise reduction filtering, time synchronization alignment, outlier handling and circular buffer storage operations on the sensor time-series data stream to obtain the preprocessed data source.

[0173] The adaptive state and prediction module 200 is used to calculate the state of charge and health of the battery by the operating state estimation algorithm, and to identify the real-time electrochemical operating mode of the battery by combining the current, voltage change rate and filter output. Then, based on the identified operating mode, the optimal prediction model is dynamically selected to generate the prediction value, so as to calculate the prediction residual between the original data and the prediction value.

[0174] The fidelity intelligent control module 300 is used to dynamically apply differentiated local error constraints based on the real-time electrochemical operation mode identified by the adaptive state and prediction module, combined with the fidelity requirements of electrochemical diagnosis for specified features, and to perform adaptive quantization processing on the prediction residual to obtain a quantized residual that meets the fidelity requirements.

[0175] The state-dependent codebook switching module 400 is used to analyze the statistical characteristics of the quantization residual corresponding to the real-time electrochemical operation mode, dynamically match the quantization residual, and call the appropriate entropy encoder for encoding processing.

[0176] The dual-stream encoding execution control module 500 is used to encapsulate the encoded result into a synchronous metadata stream and a residual data stream. The metadata stream includes the context information required for decoding, and the residual data stream includes the compressed residual bit stream, thus completing the data compression.

[0177] In this embodiment, the data preprocessing module 100 includes:

[0178] The denoising filter for the voltage signal employs a sliding window-based median filter combined with wavelet thresholding. The threshold value for wavelet thresholding is... The formula is:

[0179]

[0180] In the formula, The standard deviation of noise. The signal length;

[0181] The size of the circular buffer is:

[0182]

[0183] In the formula, Sampling frequency, This represents the maximum expected delay.

[0184] In this embodiment, the adaptive state and prediction module 200 runs an adaptive nonlinear Kalman filter algorithm to calculate the battery's state of charge (SOC) and state of health (SOH), and identifies the battery's real-time electrochemical operating mode based on current, voltage change rate, and filter output. The real-time electrochemical operation mode The classification logic is as follows:

[0185]

[0186] In the formula, For current threshold, For the sliding window variance, For the target voltage, For voltage threshold, For a moment The current, For a moment voltage, For a moment The electrochemical operating mode, The current stability variance threshold. This represents the time interval of the sliding window.

[0187] In this embodiment, the selection rule for the optimal prediction model in the adaptive state and prediction module 200 is as follows:

[0188] When real-time electrochemical operation mode In static mode, zero-order hold prediction is used, and the predicted value is... In the formula, For a moment The predicted value, For a moment The original data;

[0189] When real-time electrochemical operation mode In constant current CC / constant voltage CV mode, second-order difference linear extrapolation is used to predict the value. In the formula, For a moment The original data;

[0190] When real-time electrochemical operation mode In the pulse mode, an AUKF-based model is used for prediction, and the predicted value is... In the formula, For time-based Raw data and time Current The AUKF prior estimate;

[0191] Predicted residuals In the formula, For a moment The original data.

[0192] In this embodiment, the local error constraint in the fidelity intelligent control module 300 The setting logic is as follows:

[0193] Preset standard error limits and critical error limits The dynamic error limit is defined when the following conditions are met. ,otherwise :

[0194]

[0195] In the formula, This is a critical window condition for DVA. For the state of charge calculated by AUKF, For a predefined SOC range that includes the DVA peak;

[0196] The step size for adaptive quantization is:

[0197]

[0198] The quantization residual is:

[0199]

[0200] In the formula, For a moment The original residuals, This is the rounding function;

[0201] Reconstructing residuals And satisfy In the formula, For a moment The reconstruction residual.

[0202] In this embodiment, the switching rule of the entropy encoder in the state-dependent codebook switching module 400 is as follows:

[0203] When real-time electrochemical operation mode In static mode, run-length encoding (RLE) is selected, and the encoding logic is to count the number of consecutive zeros in the quantized residuals. If a non-zero value is encountered Then output The encoding formula is:

[0204]

[0205] In the formula, The encoding result of the residual data stream, For a moment to Quantized residual sequence;

[0206] When real-time electrochemical operation mode When using constant current (CC) / constant voltage (CV) mode, choose either Adaptive Huffman Coding (AHE) or Golomb-Rice Coding. The Golomb mapping formula is:

[0207]

[0208] In the formula, For quotient, For Rice parameters, The remainder;

[0209] When real-time electrochemical operation mode In pulse mode, select either context-adaptive binary arithmetic coding (CABAC) or Exp-Golomb coding.

[0210] In this embodiment, the dual-stream encoding execution control module 500 includes:

[0211] The metadata stream includes the context information required for decoding, using Lempel-Ziv or fixed-length encoding, and the data packet structure includes a Header and a Block Payload.

[0212] The header includes a timestamp. Initial value and In the formula, This is the start timestamp of the data. The original data at the initial moment, This refers to the original data from the time step preceding the initial time step;

[0213] The block payload includes each state block. Len and In the formula, Here, Len is the status label, and Len is the number of data points that this status lasts. Index for the error limits used;

[0214] The residual data stream includes the compressed residual bit stream, and the two streams are synchronously stored and transmitted through index files and data files.

[0215] It should be noted that the information interaction and execution process between the modules of the above-mentioned device are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0216] Example 3

[0217] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code of an electrochemical state-aware dual-stream adaptive data compression method. The program code includes instructions for executing the electrochemical state-aware dual-stream adaptive data compression method of Embodiment 1 or any possible implementation thereof.

[0218] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0219] Example 4

[0220] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0221] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can execute the electrochemical state-aware dual-stream adaptive data compression method of Embodiment 1 or any possible implementation thereof by calling the program instructions.

[0222] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0223] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0224] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0225] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A dual-stream adaptive data compression method based on electrochemical state sensing, characterized in that, Includes the following steps: Collect the sensor time-series data stream of the battery management system. The sensor time-series data stream includes voltage, current and temperature data. Perform noise reduction filtering, time synchronization alignment, outlier handling and circular buffer storage operations on the sensor time-series data stream to obtain the preprocessed data source. An adaptive state and prediction module is constructed based on the preprocessed data source. The state of charge and health of the battery are calculated by the state estimation algorithm of the adaptive state and prediction module. The real-time electrochemical operation mode of the battery is identified by combining the current, voltage change rate and filter output. The optimal prediction model is dynamically selected according to the identified operation mode to generate the prediction value, and the prediction residual between the original data and the prediction value is calculated. A fidelity intelligent control module is constructed. Based on the real-time electrochemical operation mode identified by the adaptive state and prediction module, and combined with the fidelity requirements of electrochemical diagnosis for specified features, a differentiated local error constraint is dynamically applied, and the prediction residual is adaptively quantized to obtain a quantized residual that meets the fidelity requirements. A state-dependent codebook switching module is constructed to analyze the statistical characteristics of the quantization residuals corresponding to the real-time electrochemical operation mode, and to dynamically match and call the appropriate entropy encoder for encoding processing of the quantization residuals. A dual-stream encoding execution control module is constructed to encapsulate the encoded result into a synchronous metadata stream and a residual data stream. The metadata stream includes the context information required for decoding, and the residual data stream includes the compressed residual bit stream, thus completing data compression.

2. The dual-stream adaptive data compression method based on electrochemical state sensing according to claim 1, characterized in that, The denoising filter for the voltage signal employs a sliding window-based median filter combined with wavelet thresholding. The threshold value for wavelet thresholding is... The formula is: In the formula, The standard deviation of noise. The signal length; The size of the circular buffer is: In the formula, Sampling frequency, This represents the maximum expected delay.

3. The dual-stream adaptive data compression method based on electrochemical state sensing according to claim 1, characterized in that, The adaptive state and prediction module uses an adaptive nonlinear Kalman filter algorithm to calculate the battery's state of charge (SOC) and state of health (SOH), and identifies the battery's real-time electrochemical operating mode based on current, voltage change rate, and filter output. The real-time electrochemical operation mode The classification logic is as follows: In the formula, For current threshold, For the sliding window variance, For the target voltage, For voltage threshold, For a moment The current, For a moment voltage, For a moment The electrochemical operating mode, The current stability variance threshold. This represents the time interval of the sliding window.

4. The dual-stream adaptive data compression method based on electrochemical state sensing according to claim 3, characterized in that, In the process of dynamically selecting the optimal prediction model to generate predicted values ​​based on the identified operating mode, the selection rule for the optimal prediction model is as follows: When real-time electrochemical operation mode In static mode, zero-order hold prediction is used, and the predicted value is... In the formula, For a moment The predicted value, For a moment The original data; When real-time electrochemical operation mode In constant current CC / constant voltage CV mode, second-order difference linear extrapolation is used to predict the value. In the formula, For a moment The original data; When real-time electrochemical operation mode In the pulse mode, an AUKF-based model is used for prediction, and the predicted value is... In the formula, For time-based Raw data and time Current The AUKF prior estimate; Predicted residuals In the formula, For a moment The original data.

5. The dual-stream adaptive data compression method based on electrochemical state sensing according to claim 1, characterized in that, The local error constraint The setting logic is as follows: Preset standard error limits and critical error limits The dynamic error limit is defined when the following conditions are met. ,otherwise : In the formula, This is a critical window condition for DVA. For the state of charge calculated by AUKF, For a predefined SOC range that includes the DVA peak; The step size for adaptive quantization is: The quantization residual is: In the formula, For a moment The original residuals, This is the rounding function; Reconstructing residuals And satisfy In the formula, For a moment The reconstruction residual.

6. The dual-stream adaptive data compression method based on electrochemical state sensing according to claim 1, characterized in that, The switching rule for the entropy encoder is as follows: When real-time electrochemical operation mode In static mode, run-length encoding (RLE) is selected, and the encoding logic is to count the number of consecutive zeros in the quantized residuals. If a non-zero value is encountered Then output The encoding formula is: In the formula, The encoding result of the residual data stream, For a moment to Quantized residual sequence; When real-time electrochemical operation mode When using constant current (CC) / constant voltage (CV) mode, choose either Adaptive Huffman Coding (AHE) or Golomb-Rice Coding. The Golomb mapping formula is: In the formula, For quotient, For Rice parameters, The remainder; When real-time electrochemical operation mode In pulse mode, select either context-adaptive binary arithmetic coding (CABAC) or Exp-Golomb coding.

7. The dual-stream adaptive data compression method based on electrochemical state sensing according to claim 1, characterized in that, The metadata stream includes the context information required for decoding, using Lempel-Ziv or fixed-length encoding, and the data packet structure includes a Header and a Block Payload. The header includes a timestamp. Initial value and In the formula, This is the start timestamp of the data. The original data at the initial moment, This refers to the original data from the time step preceding the initial time step; The block payload includes each state block. Len and In the formula, Here, Len is the status label, and Len is the number of data points that this status lasts. Index for the error limits used; The residual data stream includes the compressed residual bit stream, and the two streams are synchronously stored and transmitted through index files and data files.

8. A dual-stream adaptive data compression system based on electrochemical state sensing, characterized in that, include: The data preprocessing module is used to collect the sensor time-series data stream of the battery management system. The sensor time-series data stream includes voltage, current and temperature data. The module performs noise reduction filtering, time synchronization alignment, outlier handling and circular buffer storage operations on the sensor time-series data stream to obtain the preprocessed data source. The adaptive state and prediction module is used to run the state estimation algorithm to solve the battery's state of charge and health state, and to identify the real-time electrochemical operation mode of the battery by combining current, voltage change rate and filter output. Then, based on the identified operation mode, the optimal prediction model is dynamically selected to generate predicted values, and the prediction residual between the original data and the predicted values ​​is calculated. The fidelity intelligent control module is used to dynamically apply differentiated local error constraints based on the real-time electrochemical operation mode identified by the adaptive state and prediction module, combined with the fidelity requirements of electrochemical diagnosis for specified features, and to perform adaptive quantization processing on the prediction residual to obtain a quantized residual that meets the fidelity requirements. The state-dependent codebook switching module is used to analyze the statistical characteristics of the quantization residual corresponding to the real-time electrochemical operation mode, dynamically match the quantization residual, and call the appropriate entropy encoder for encoding processing. The dual-stream encoding execution control module is used to encapsulate the encoded result into a synchronous metadata stream and a residual data stream. The metadata stream includes the context information required for decoding, and the residual data stream includes the compressed residual bit stream, thus completing the data compression.

9. The dual-stream adaptive data compression system based on electrochemical state sensing according to claim 8, characterized in that, In the data preprocessing module: The denoising filter for the voltage signal employs a sliding window-based median filter combined with wavelet thresholding. The threshold value for wavelet thresholding is... The formula is: In the formula, The standard deviation of noise. The signal length; The size of the circular buffer is: In the formula, Sampling frequency, Maximum expected delay; The adaptive state and prediction module uses an adaptive nonlinear Kalman filter algorithm to calculate the battery's state of charge (SOC) and state of health (SOH), and identifies the battery's real-time electrochemical operating mode based on current, voltage change rate, and filter output. The real-time electrochemical operation mode The classification logic is as follows: In the formula, For current threshold, For the sliding window variance, For the target voltage, For voltage threshold, For a moment The current, For a moment voltage, For a moment The electrochemical operating mode, The current stability variance threshold. The time interval of the sliding window; In the adaptive state and prediction module, the selection rule for the optimal prediction model is as follows: When real-time electrochemical operation mode In static mode, zero-order hold prediction is used, and the predicted value is... In the formula, For a moment The predicted value, For a moment The original data; When real-time electrochemical operation mode In constant current CC / constant voltage CV mode, second-order difference linear extrapolation is used to predict the value. In the formula, For a moment The original data; When real-time electrochemical operation mode In the pulse mode, an AUKF-based model is used for prediction, and the predicted value is... In the formula, For time-based Raw data and time Current The AUKF prior estimate; Predicted residuals In the formula, For a moment The original data.

10. The dual-stream adaptive data compression system based on electrochemical state sensing according to claim 9, characterized in that, In the fidelity intelligent control module, the local error constraint The setting logic is as follows: Preset standard error limits and critical error limits The dynamic error limit is defined when the following conditions are met. ,otherwise : In the formula, This is a critical window condition for DVA. For the state of charge calculated by AUKF, For a predefined SOC range that includes the DVA peak; The step size for adaptive quantization is: The quantization residual is: In the formula, For a moment The original residuals, This is the rounding function; Reconstructing residuals And satisfy In the formula, For a moment The reconstruction residual; In the state-dependent codebook switching module, the switching rule for the entropy encoder is as follows: When real-time electrochemical operation mode In static mode, run-length encoding (RLE) is selected, and the encoding logic is to count the number of consecutive zeros in the quantized residuals. If a non-zero value is encountered Then output The encoding formula is: In the formula, The encoding result of the residual data stream, For a moment to Quantized residual sequence; When real-time electrochemical operation mode When using constant current (CC) / constant voltage (CV) mode, choose either Adaptive Huffman Coding (AHE) or Golomb-Rice Coding. The Golomb mapping formula is: In the formula, For quotient, For Rice parameters, The remainder; When real-time electrochemical operation mode In pulse mode, select either context-adaptive binary arithmetic coding (CABAC) or Exp-Golomb coding. In the dual-stream encoding execution control module: The metadata stream includes the context information required for decoding, using Lempel-Ziv or fixed-length encoding, and the data packet structure includes a Header and a Block Payload. The header includes a timestamp. Initial value and In the formula, This is the start timestamp of the data. The original data at the initial moment, This refers to the original data from the time step preceding the initial time step; The block payload includes each state block. Len and In the formula, Here, Len is the status label, and Len is the number of data points that this status lasts. Index for the error limits used; The residual data stream includes the compressed residual bit stream, and the two streams are synchronously stored and transmitted through index files and data files.

Images