Multi-working-condition sensor method and system suitable for power battery

Abstract

The invention relates to the field of power battery data processing methods, in particular to a multi-working-condition sensor method and system suitable for a power battery, and the method comprises the steps: synchronously reading the voltage, current and temperature data of the power battery, and setting the sampling frequency and range; calculating a temperature gradient, calculating an equivalent multiplying power, and determining a working condition type; performing composite drift compensation by applying a temperature gradient and an aging drift term, executing wavelet noise reduction, and updating filter parameters; calculating the median of each sensor group and the standard deviation of all the sensors, combining the confidence components of the three parts, and calculating the confidence of each sensor through normalization processing; the weight of each sensor is calculated, a weighted average value is calculated, a second derivative is estimated, dynamic smooth constraint is applied to execute adaptive weighted fusion, and anomaly detection and isolation are carried out; issuing high-precision data, generating a health report, and triggering an early warning mechanism; checking calibration conditions, executing zero point / gain calibration, and updating compensation parameters. According to the invention, the abnormity of the power battery can be dynamically and accurately found in real time.

Description

Technical Field

[0001] This invention relates to the field of power battery data processing methods, and more specifically to a multi-condition sensor method and system applicable to power batteries. Background Technology

[0002] As the "heart" of new energy vehicles, the power battery provides energy storage and power output for their operation. The power battery is very important for new energy vehicles, so it is crucial to accurately collect multiple parameters of the power battery using different sensors for the analysis of the operating conditions of new energy vehicles.

[0003] However, in the actual operation of new energy vehicles, the driving conditions are very complex, and the data collected by sensors must meet certain accuracy requirements, and some even require collection at extreme accuracy. Currently, static parameter models at extreme accuracy cannot adapt to changes in the nonlinear characteristics of the battery, leading to the following problems: Accuracy degradation issue: Traditional sensors suffer severe accuracy degradation under complex operating conditions (low temperature, high magnification), with errors exceeding 15% at low temperatures and exceeding 12% at high magnification; Insufficient dynamic adaptability: Existing algorithms cannot dynamically adapt to the combined effects of temperature, magnification, and aging; Insufficient real-time performance: The latency of multi-sensor data fusion exceeds 200ms, which is difficult to meet the requirements of real-time control; Lifecycle accuracy decline: Lacking a full lifecycle self-calibration mechanism, accuracy declines by more than 30% after one year of use. Summary of the Invention

[0004] The present invention aims to provide a multi-condition sensor method and system suitable for power batteries to solve the problems of accuracy decay, insufficient dynamic adaptability and insufficient real-time performance.

[0005] According to one aspect of the present invention, a multi-condition sensor method suitable for power batteries is provided, comprising the following steps: Step 1: Synchronously read the voltage, current, and temperature data of the power battery, and set the sampling frequency and range; Step 2: Calculate the temperature gradient based on the temperature data, calculate the equivalent ratio based on the current data, and determine the operating condition type based on the gradient temperature, equivalent ratio, and internal resistance change rate. Step 3: Apply temperature gradient and aging drift term for composite drift compensation, perform wavelet denoising, and update filter parameters; Step 4: Calculate the median of each sensor group and the standard deviation of all sensors, then combine the three confidence components, normalize them to calculate the confidence of each sensor; calculate the weight of each sensor, calculate the weighted average, estimate the second derivative, apply dynamic smoothing constraints to perform adaptive weighted fusion, and perform anomaly detection and isolation of the power battery. Step 5: Publish high-precision data, generate a health report, and trigger the early warning mechanism; Step 6: Check calibration conditions, perform zero-point / gain calibration, and update compensation parameters.

[0006] Furthermore, in step 2, the calculation process for the temperature gradient is as follows: A three-dimensional coordinate system mapping the sensor position is established based on the temperature data. The central difference method is used to calculate the temperature partial derivatives in each direction. Based on the temperature partial derivatives, a three-dimensional temperature gradient vector is synthesized, which is expressed as: ; in, x is the rate of temperature change in the direction. The sensor spacing is 1-2 cm, with a typical value of 1.5 cm. The temperature value at location (i,j,k) is expressed in °C. The temperature gradient (°C / cm) has the following range: Normal operating conditions: 0-2°C / cm; Warning state: 2-5°C / cm; Dangerous state: >5°C / cm. The calculation process for the efficiency ratio is as follows: Calculation based on current data The moving root mean square is used to obtain the current SOH estimate, and the aging compensation factor is applied to correct the equivalent ratio. The output is the equivalent ratio. , is represented as: ; in, To obtain the real-time current through sampling; Rated capacity (Ah) is a parameter inherent to the battery. This is a time window, with a value range of 10-60 seconds and a default of 30 seconds. For a healthy state, the value ranges from 80% to 100%. This is the aging sensitivity coefficient, with a value range of 0.15-0.25 and a typical value of 0.2. The process for determining the operating condition type is as follows: The eigenvector F is calculated based on the gradient temperature, equivalent ratio, and rate of change of internal resistance, and is expressed as follows: ; in, The change rate is expressed as a multiple, in C / min, and ranges from -10 to 10. The rate of change of internal resistance (range 0.1-0.5); The pre-trained classification model outputs the current dominant operating condition, and the expression for outputting the current dominant operating condition is: ; Among them, number of working conditions Operating condition categories include low temperature low expansion, low temperature high expansion, normal temperature low expansion, normal temperature high expansion, high temperature low expansion, and high temperature high expansion. For classification weights.

[0007] Furthermore, in step 3, the composite drift compensation process involves: measuring the current temperature T and the rate of change dT / dt; calculating the temperature drift; and obtaining the state of equilibrium (SOH) and current. Calculate aging drift and output the compensated voltage. The composite drift compensation is expressed as: ; The temperature drift term is represented as follows: ; The aging drift term is represented as: ; These are the temperature polynomial coefficients; c is the temperature change rate coefficient; c is the aging sensitivity coefficient. The characteristic current is SOH; the healthy state is SOH. The multi-scale wavelet denoising process is as follows: The signal is decomposed into 5-level wavelet decomposition, the noise variance at each scale is estimated, adaptive weights are calculated, and the reconstructed signal is represented as follows: ; The formula for optimizing wavelet coefficients is: ; in, For wavelet coefficients at scale j; For noise variance estimation; For scale weights, satisfying j represents the decomposition level, j = 1-5.

[0008] Furthermore, in step 4, the formula for calculating the sensor confidence level is: ; in, This is the calibration accuracy factor; Here, is the recent variance, and is the sliding window variance; This is the median of the sensor group; The standard deviation for all sensors; Confidence level; The formula for adaptive weighted fusion is expressed as: ; Among them, smoothing factor The value range is 0.05-0.15, expressed as: ; The normalized rate of change of current is expressed as: ; This is the average value from the sensor.

[0009] Furthermore, in step 4, the abnormal detection of the power battery is represented as follows: ; in, The median; This represents the absolute deviation of the median. The maximum permissible rate of change (related to operating conditions) is 0.1 V / s at low magnification and 5 V / s at high magnification. This is the Pearson correlation coefficient.

[0010] Furthermore, in step 6, zero-point drift calibration is expressed as: ; The triggering condition for zero-point drift calibration is: continued ; ; .

[0011] Furthermore, in step 6, the gain error compensation process is as follows: connect the reference voltage source, measure the output voltage, calculate the gain error, and update the gain coefficient, expressed by the following formula: ; in, This is the reference voltage.

[0012] The triggering conditions for gain error compensation are: after the external reference source is available, the constant voltage charging stage is met, or the monthly scheduled execution is met. Attached Figure Description

[0013] Figure 1 This is a schematic block diagram of an embodiment of the multi-condition sensor method applicable to power batteries. Detailed Implementation

[0014] The following detailed description provides further details on specific implementation methods.

[0015] Example 1 Multi-condition sensor methods applicable to power batteries, such as Figure 1 As shown, it includes the following steps: Step 1: Synchronously read the voltage, current, and temperature data of the power battery, and set the sampling frequency and voltage range.

[0016] Step 2: Calculate the temperature gradient based on the temperature data, calculate the equivalent ratio based on the current data, and determine the operating condition type based on the gradient temperature, equivalent ratio, and internal resistance change rate.

[0017] The temperature gradient is calculated as follows: First, read data from N temperature sensors distributed on the battery surface, where N≥6. Establish a three-dimensional coordinate system (x,y,z) mapping the sensor positions. Calculate the partial temperature derivatives in each direction using the central difference method. Based on these partial temperature derivatives, synthesize a three-dimensional temperature gradient vector, expressed as: ; in, x is the rate of temperature change in the direction. The sensor spacing is 1-2 cm, with a typical value of 1.5 cm. The temperature value at location (i,j,k) is expressed in °C. The temperature gradient (°C / cm) has the following range: Normal operating conditions: 0-2°C / cm; Warning state: 2-5°C / cm; Dangerous state: >5°C / cm.

[0018] The calculation process for the equivalent ratio is as follows: Read the current data acquired at a sampling rate of 1kHz and calculate... The moving root mean square (RMS) is used to obtain the current SOH estimate, which is derived from the battery BMS. An aging compensation factor is applied to correct the equivalent rate, and the output is the equivalent rate. for: ; in, To obtain the real-time current through sampling; Rated capacity (Ah) is a parameter inherent to the battery. This is a time window, with a value range of 10-60 seconds and a default of 30 seconds. For a healthy state, the value ranges from 80% to 100%. This is the aging sensitivity coefficient, with a value range of 0.15-0.25 and a typical value of 0.2.

[0019] The aging compensation factor is a dynamic correction term used to correct the impact of battery aging on the equivalent rate. Its specific expression is: [1+α(1-SOH)]. The core of this aging compensation factor is to dynamically adjust the equivalent rate based on the degree of battery aging: When the battery is brand new (1-SOH=0), the "basic rate calculation term" is directly used as the equivalent rate (without aging correction). After battery aging, the "basic rate calculation term" is slightly amplified to correct the "actual effective rate deviation" caused by aging. The aging compensation factor allows the equivalent rate to more accurately match the actual operating intensity of the battery under its current aging state: Battery aging (reduced SOH) causes a change in the "actual effective rate corresponding to the same real-time current." Through the dynamic adjustment of this factor, the calculation deviation caused by aging can be eliminated, ensuring that the equivalent rate truly reflects the battery's current actual load (rather than just the theoretical value based on a new battery).

[0020] The process for determining the operating condition type is as follows: The eigenvector F is calculated based on the gradient temperature, equivalent ratio, and rate of change of internal resistance, and is expressed as follows: ; in, The change rate is expressed as a multiple, in C / min, and ranges from -10 to 10. The internal resistance change rate (range 0.1-0.5).

[0021] The current dominant operating condition is output through a pre-trained classification model, which can be an existing SVM or neural network. The training process is existing and will not be described in detail here. The expression for outputting the current dominant operating condition is: ; Among them, number of working conditions Operating condition categories include low temperature low expansion, low temperature high expansion, normal temperature low expansion, normal temperature high expansion, high temperature low expansion, and high temperature high expansion. These are the classification weights, which are obtained through offline training.

[0022] Step 3: Based on the equivalent multiplier from Step 2 (equivalent multiplier range 0.01C-15C), calculate the sigmoid function output, set the optimized ADC sampling frequency, and use the standard sigmoid formula to map the equivalent multiplier to a 0-1 interval value σ. Match the sampling rate according to the sigmoid output interval, specifically: When σ < 0.3, a low sampling rate of 100-500Hz is matched; When 0.3 < σ < 0.7, the sampling rate is 500-5kHz in the matching range. When σ>0.7, a high sampling rate of 5-20kHz is matched.

[0023] Configure DMA transfer parameters. The core of DMA parameter configuration is to accurately match the optimized ADC sampling frequency to ensure that the voltage / current data acquired by the ADC is transferred from the ADC register to the MCU memory without loss and with low latency. The parameters are dynamically adjusted according to the sampling frequency. There is no unified formula; only key items need to be adapted according to the sampling frequency. The process of adapting key items is as follows: (1) The core configuration rule of the transmission mode is: fixed selection of the cyclic mode. Example scenario: applicable to all low / medium / high rate conditions, adapting to the continuous sampling requirements of power batteries, without the need for repeated manual configuration of transmission trigger.

[0024] (2) The core configuration rule based on the buffer size is: calculate the buffer size according to the formula, buffer size = optimized ADC sampling frequency × number of bytes per sample (the number of bytes per sample for a 16-bit ADC is fixed at 2 bytes). Example scenarios: under low magnification, the sampling rate is 100Hz: the buffer size is 100×2=200 bytes; under medium magnification, the sampling rate is 500Hz: the buffer size is 500×2=1000 bytes; under high magnification, the sampling rate is 5kHz: the buffer size is 5000×2=10000 bytes.

[0025] (3) The core configuration rule of the trigger source is: bind the sampling completion interrupt (EOC) of the ADC. Example scenario: all gears are triggered by the ADC completing one sampling, ensuring that the sampling action and data transmission are synchronized, and avoiding data delay or loss.

[0026] (4) The core configuration rule based on transmission priority is: the higher the sampling frequency, the higher the DMA transmission priority setting. For example, in the case of low-rate operation of 100-500Hz: set to low priority; in the case of medium-rate operation of 500-5kHz: set to medium priority; in the case of high-rate operation of 5-20kHz: set to high priority to avoid high-frequency sampling data transmission being blocked.

[0027] (5) The core configuration rule for address configuration is: the source address is fixed to the ADC data register (such as the ADC_DR register); the destination address is set to the MCU memory (such as SRAM) and incremental mode is enabled; Example scenario: all operating conditions follow this configuration, the source address always points to the storage register of the ADC data acquisition (fixed), and the destination address is automatically offset by 2 bytes after each data transmission (matching the number of bytes in a single sample) to achieve sequential storage of continuous data.

[0028] In summary, DMA parameters do not have complex calculation formulas. The core principle is that "the sampling frequency determines the buffer size and priority": the higher the sampling frequency, the larger the buffer needs to be and the higher the priority. Combined with the cyclic mode and ADC sampling trigger, it ensures real-time data transmission under different operating conditions and avoids overflow or blocking.

[0029] The formula for optimizing the sampling frequency is expressed as: ; Among them, equivalent ratio The temperature range is 0.01°C to 15°C, and the sampling rate is... The scope is: Low operating conditions, sampling rate The range is 100-500Hz Medium operating conditions, sampling rate The range is 500-5kHz. High operating conditions, sampling rate The range is 5-20kHz.

[0030] The optimized sampling frequency replaces the traditional fixed sampling rate throughout the entire process, dynamically switching according to the real-time operating conditions of the power battery. The triggering condition is strongly bound to the equivalent rate (after normalization by the Sigmoid function): Under low-rate operating conditions (i.e., equivalent rate 0.01C~0.5C, Sigmoid output <0.3): use 100~500Hz to adapt to low-speed driving / shallow charge and discharge and reduce data redundancy; In medium-rate operation (i.e., equivalent rate 0.5C~3C, Sigmoid output 0.3~0.7): use 500~5kHz, suitable for regular driving / normal charging, balancing accuracy and real-time performance; Under high-rate conditions (i.e., equivalent rate 3C~15C, Sigmoid output >0.7): use 5~20kHz to adapt to rapid acceleration / fast charging, capture rapid changes in voltage / current, and avoid data distortion.

[0031] By calculating the output of the sigmoid function, the ADC sampling frequency can be dynamically set and optimized, and DMA transfer parameters can be configured to achieve adaptive matching of "operating conditions (equivalent magnification) - sampling rate", thus solving the problems of "low magnification redundancy and high magnification insufficient accuracy" at fixed sampling rates.

[0032] Monitor the voltage fluctuation over the last 10ms and calculate the peak voltage, which is expressed as: Select the smallest suitable range setting, which is: [0-2V, 0-5V, ±1V, ±2V, ±5V]. Configure the PGA gain, following the principle of: high gain for small range settings and low gain for large range settings. This ensures that the peak voltage (V_peak) is amplified to match the full scale of the ADC (the default full scale of the ADC reference is 5V). This improves the sampling accuracy of small voltage signals and avoids voltage overflow distortion after amplification. Dynamically optimize the voltage range in step 1. The dynamic voltage range is expressed as: ; in, The maximum permissible range is 5V.

[0033] The process of selecting the smallest suitable range is as follows: First, determine the polarity of the peak voltage (V_peak): is it only positive voltage, only negative voltage, or a bipolar voltage containing both positive and negative values? Then, filter for ranges that "completely cover the peak voltage": positive voltage must meet the requirement of "range upper limit ≥ V_peak", negative voltage must meet the requirement of "range absolute value ≥ |V_peak|", and bipolar voltage must meet the requirement of "both positive and negative absolute values ​​≥ |V_peak|". Among the ranges that meet the conditions, select the one with the "smallest range". The range size equals the full-scale span, such as ±1V or 0-2V, both spanning 2V. Prioritize the range that best matches the signal polarity. For example: Scenario 1: V_peak = 1.8V (positive voltage only, no negative values). The suitable voltage levels are: 0-2V (upper limit 2V ≥ 1.8V), ±2V (absolute value 2V ≥ 1.8V), 0-5V, ±5V; the minimum suitable voltage level is: 0-2V (span 2V, unipolar pure positive voltage, no redundant coverage of negative voltage).

[0034] Scenario 2: V_peak = -2.1V (negative voltage only, no positive value), the suitable range is: ±5V (absolute value 5V ≥ 2.1V) (±2V absolute value 2V < 2.1V, 0-2V / 0-5V unipolar cannot cover negative voltage, so they are excluded); the smallest suitable range is: ±5V (the only range that can cover -2.1V).

[0035] Scenario 3: V_peak = 0.9V (positive voltage, small amplitude). The suitable ranges are: ±1V (absolute value 1V ≥ 0.9V), 0-2V (upper limit 2V ≥ 0.9V), ±2V, 0-5V, and ±5V. The smallest suitable range is: ±1V (span of 2V, consistent with the 0-2V span, but with a smaller absolute value at full scale and higher sampling accuracy).

[0036] Scenario 4: V_peak = 3.6V (positive voltage, exceeding 2V), the suitable ranges are: 0-5V (upper limit 5V ≥ 3.6V) and ±5V (absolute value 5V ≥ 3.6V); the minimum suitable range is: 0-5V (5V range, unipolar fit to pure positive voltage, no need for redundant coverage of negative voltage).

[0037] Scenario 5: V_peak = ±1.5V (including positive and negative voltages, bidirectional fluctuations). The suitable ranges are: ±2V (absolute value 2V ≥ 1.5V) and ±5V (0-2V / 0-5V unipolar cannot cover negative voltages, so they are excluded). The smallest suitable range is: ±2V (span of 4V, which is the smallest bipolar range that can cover ±1.5V).

[0038] When configuring the gain, if the range is 0-2V: the PGA gain is 2 times, and the selection logic is that the full scale after amplification is 2×2=4V<5V (ADC full scale), which will not cause overflow and can improve the sampling accuracy of the voltage signal. When the range is 0-5V: the PGA gain is 1x, and the selection logic is that this range is already matched with the full range of the ADC, so no additional amplification is needed and it can be directly sampled. When the range is ±1V: the PGA gain is 5 times, and the selection logic is 5×1=5V after amplification, which is just right to match the full range of the ADC, and can maximize the sampling accuracy of small voltage signals. When the range is ±2V: the PGA gain is 2 times, and the selection logic is that the full scale after amplification is 2×2=4V<5V, which can avoid bipolar voltage signal overflow and ensure sampling accuracy. When the range is ±5V: the PGA gain is 1x, and the selection logic is that this range is consistent with the full-scale range of the ADC. The bipolar voltage signal does not need to be amplified and can be directly acquired.

[0039] For example: Scenario 1: V_peak = 1.8V (positive voltage only), select the minimum suitable range as: 0-2V (can cover the minimum range of 1.8V); configure PGA gain: 2 times; amplified voltage: 1.8×2=3.6V (<5V, no overflow, improved sampling accuracy).

[0040] Scenario 2: V_peak=0.9V (positive voltage only, small signal), select the minimum suitable range: ±1V (can cover the minimum range of 0.9V); configure PGA gain: 5 times; amplified voltage: 0.9×5=4.5V (close to the full scale of ADC, maximizing the accuracy of small signal).

[0041] Scenario 3: V_peak = ±1.5V (bipolar voltage), select the minimum suitable range: ±2V (the minimum range that can cover ±1.5V); configure PGA gain: 2 times; amplified voltage: 1.5×2=3V (<5V, no overflow of bipolar signal).

[0042] Scenario 4: V_peak = 3.6V (positive voltage only, exceeding 2V), select the minimum suitable range: 0-5V (can cover the minimum range of 3.6V); configure PGA gain: 1x; amplified voltage: 3.6×1=3.6V (no amplification required, directly adapts to ADC range).

[0043] Scenario 5: V_peak = -2.1V (negative voltage only), select the minimum suitable range: ±5V (the only range that can cover -2.1V); configure PGA gain: 1x; amplified voltage: 2.1×1=2.1V (no overflow, bipolar signal directly acquired).

[0044] Such dynamic range settings apply only to the voltage sampling stage, with two key limitations: First, they limit the upper and lower limits of the voltage signal range, ensuring that the peak voltage (V_peak) of the voltage fluctuation in the most recent 10ms does not exceed the selected range (such as 0-2V, ±2V, 0-5V, etc.) to prevent voltage signal overflow and distortion. Second, they limit the range size, forcing the selection of the smallest range while covering the peak voltage, to avoid excessively large ranges that reduce voltage sampling resolution (waste of accuracy).

[0045] Noise adaptive filtering is performed on the dynamically range-optimized power battery voltage and current sampling signals (acquired at a 1kHz sampling rate). Specifically, the required cutoff frequency is calculated, and the cutoff frequency is dynamically adjusted according to the equivalent ratio to achieve accurate noise reduction under different operating conditions. The formula for calculating the required cutoff frequency is as follows: ; in, The cutoff frequency (Hz) ranges from 300 to 1500 Hz. A fourth-order Butterworth filter is used to filter noise from the sensor data. Its transfer function is expressed as: ; Where is the cutoff angular frequency, s is the complex frequency variable, and is the input variable of H(s).

[0046] The fourth-order Butterworth filter is discretized using a bilinear transform, and the filter coefficients are updated. The specific process is as follows: The core mapping relationship of the bilinear transform is determined to resolve frequency aliasing, mapping the analog domain (s-domain) transfer function to the digital domain (z-domain). The mapping relationship is expressed as follows: ; Where T is the optimized ADC sampling period, determined by the ADC sampling frequency. Decide, ; This is a digital domain delay operator, indicating a delay of one sampling period.

[0047] Substituting the mapping relationship into the transfer function H(s) completes the discretization, yielding the z-domain transfer function of the digital filter. Combined with the existing parameters k=2 / T, After simplification, we obtain the coefficient expression for the fourth-order Butterworth filter, which will not be elaborated here.

[0048] The filter coefficients are dynamically updated according to the operating conditions of the power battery. When the equivalent rate changes, the adaptive cutoff frequency is recalculated. Low magnification =50Hz, high magnification =1kHz.

[0049] Synchronously update ADC sampling frequency This yields a new sampling period. The following process is repeated: write the new coefficients to the digital filter register, overwriting the old coefficients; check the equivalent multiplier every 10ms; if the change is >0.1C, update the coefficients immediately; repeat this process once more, then substitute the new coefficients... Calculate the new filter coefficients using T.

[0050] After the coefficients are updated, the digital filter processes the sampled signal in real time according to the existing difference equation.

[0051] Composite drift compensation is performed using temperature gradient and aging drift term. The aging drift term is the "aging offset compensation value" of the voltage signal. The aging drift term uses the state of health (SOH) of the battery as the core input parameter. The lower the SOH, the larger the aging drift term and the greater the correction of the aging compensation factor to the rate. This eliminates systematic errors such as temperature drift and aging effects, performs wavelet denoising, and updates the filter parameters.

[0052] The composite drift compensation process involves measuring the current temperature T and the rate of change dT / dt; calculating the temperature drift; and obtaining the state of equilibrium (SOH) and current. Calculate aging drift and output the compensated voltage. The composite drift compensation is expressed as: ; The temperature drift term is represented as follows: ; The aging drift term is represented as: ; Here are the temperature polynomial coefficients, and here is the calibration value; is the temperature change rate coefficient, 0.1-0.5 mV·s / ℃; c is the aging sensitivity coefficient, 10-50 mV; The characteristic current is 1C; SOH represents the healthy state, 80%-100%.

[0053] The signal preprocessing before noise reduction involves inputting a power battery voltage signal after composite drift compensation (eliminating systematic errors from temperature drift and aging drift). However, this signal still contains high-frequency random noise (such as vehicle electromagnetic interference and sampling circuit thermal noise). This noise interferes with the accuracy of subsequent peak voltage calculations and equivalent rate analysis. Therefore, multi-scale wavelet denoising is necessary to "preserve signal characteristics and suppress noise." The multi-scale wavelet denoising process is as follows: Balancing temporal locality and frequency domain resolution, and adapting to the "non-stationary, abrupt" characteristics of power battery voltage, a 5-level wavelet decomposition is performed on the signal. This 5-level decomposition avoids the increased computational delay caused by too many levels, and also avoids the inability to distinguish different frequencies of noise / signal due to too few levels. During decomposition, the Mallat algorithm is used to extract the low-frequency "core profile" (approximation coefficients) of the signal at each level using a low-pass filter, and to extract the high-frequency "details + noise" (detail coefficients) using a high-pass filter, thus separating different frequency components layer by layer. After decomposition, one low-frequency approximation coefficient is obtained (denoted as ca_5: corresponding to the core features of the voltage signal (such as the voltage plateau of the battery, the stable discharge segment); after decomposition, five high-frequency detail coefficients are obtained (denoted as cd_1~ cd_5): cd_1 (the first layer) corresponds to the highest frequency component, mainly consisting of random noise (such as glitches from electromagnetic interference); cd_5 (the fifth layer) corresponds to the lower frequency component, containing a small amount of signal details (such as small voltage fluctuations) + weak noise.

[0054] Estimate the noise variance at each scale, quantify the noise level at each layer, and determine that the noise follows a Gaussian distribution in the wavelet domain. And detail coefficient From signal components +Noise components The noise variance is estimated by using the statistical properties of the detail coefficients, specifically: using the detail coefficients of the j-th layer. For example, take The absolute values ​​of all coefficients are calculated to eliminate the influence of the sign on the statistics; the median of the absolute value sequence is calculated to avoid interference from signal details; the noise standard deviation is calculated and expressed as: ; The noise variance is obtained as follows: .

[0055] Calculate adaptive weights, perform adaptive detail coefficient weighting, and intelligently filter signal / noise. Assign an adaptive weight of 0-1 to each detail coefficient; signal-dominant coefficients are given higher weights (retention), and noise-dominant coefficients are given lower weights (suppression). The weight calculation formula is as follows: The k-th detail coefficient at layer j The weights are: ; The coefficient is energy.

[0056] like Signal-driven, such as voltage spikes during charging. The coefficients were fully preserved.

[0057] like Noise-dominated, such as electromagnetic interference glitches. The coefficient was significantly suppressed.

[0058] The final weighted detail coefficients are: .

[0059] The reconstructed signal is represented as: ; The formula for optimizing wavelet coefficients is: ; in, For wavelet coefficients at scale j; For noise variance estimation (real-time calculation); For scale weights, satisfying j represents the decomposition level, j = 1-5.

[0060] Step 4: Calculate the median of each sensor group and the standard deviation of all sensors, then combine the three confidence components, normalize and calculate the confidence of each sensor; calculate the weight of each sensor, calculate the weighted average, estimate the second derivative (five-point difference method), apply dynamic smoothing constraints to perform adaptive weighted fusion; calculate each detection index, determine the abnormal state and isolate the faulty sensor, and trigger the calibration procedure to perform abnormal detection and isolation of the power battery.

[0061] The formula for calculating sensor confidence level is: ; in, This is the calibration accuracy factor, ranging from 0.9 to 1.0; Here, is the recent variance, and is the sliding window variance; This is the median of the sensor group; The standard deviation for all sensors; The confidence level is 0-1.

[0062] The formula for adaptive weighted fusion is expressed as: ; Among them, smoothing factor The value range is 0.05-0.15, expressed as: ; The normalized rate of change of current is expressed as: ; This is the average value from the sensor.

[0063] Abnormal detection of power batteries is indicated as follows: ; in, The median; This represents the absolute deviation of the median. The maximum permissible rate of change (related to operating conditions) is 0.1 V / s at low magnification and 5 V / s at high magnification. This is the Pearson correlation coefficient.

[0064] Step 5: Publish high-precision data (using both CAN and Ethernet links), generate a health report, and trigger the early warning mechanism. The published data list consists of high-precision data after noise reduction and compensation, specifically: When the data type is voltage data, the accuracy requirement is ±0.01V, and the data source is after dynamic range optimization and wavelet noise reduction. When the data type is current data, the accuracy requirement is ±0.05A, and the data source is 1kHz sampling and adaptive filtering. When the data type is temperature data, the accuracy requirement is ±0.1℃, and the data source is the data after temperature gradient calculation. When the data type is an equivalent multiple, the accuracy requirement is ±0.01C, and the data source is the calculation result of step 2. When the data type is Health Status (SOH), the accuracy requirement is ±1%, and the data source is the real-time updated value from the BMS. When the data type is State of Charge (SOC), the accuracy requirement is ±2%, and the data source is combined with voltage / current estimates. When the data type is the residual noise amplitude after filtering, the accuracy requirement is <5mV, and the data source is the residual noise from wavelet denoising.

[0065] The transmission method and protocol for publishing high-precision data are as follows: When published via CAN bus (in-vehicle communication): The protocol is CAN 2.0B (extended frame), with a baud rate of 500kbps / 1Mbps (suitable for automotive environments). The transmission targets are: MCU (vehicle controller), BMS main controller, and vehicle instrument cluster; The frame format includes a data ID (e.g., voltage data 0x201, current data 0x202), a timestamp, and a checksum (CRC-8).

[0066] When publishing via Ethernet (cloud / host computer communication): The protocol is TCP / IP+UDP, supporting 100Mbps Ethernet; Transmission targets: Battery management cloud platform, test host computer; Data encapsulation: JSON format (including device ID, sampling time, and full high-precision data), supports breakpoint resume.

[0067] The transmission cycle for publishing high-precision data (dynamically adapted operating conditions) is as follows: Low-rate operation (0.01C~1C): 100ms / frame (reduce bus load). Medium-rate operation (1C~5C): 50ms / frame (balancing real-time performance and load); High-rate operation (5C~15C): 10ms / frame (ensuring real-time data for fast charging / rapid acceleration).

[0068] Generate health reports by combining structured data with trend analysis: The report's core metrics (covering all dimensions of battery status) include: Basic status: current SOH, SOC, and cycle count (cumulative charge-discharge cycles); Performance metrics: maximum allowable charge-discharge rate (dynamically adjusted based on SOH), voltage balance (maximum difference in cell voltage); Aging trend: SOH degradation over the past 30 days, aging drift (…). Growth rate; Environmental adaptation: current temperature gradient level (normal / warning / dangerous), number of extreme working condition tolerances; Data quality: filtering effectiveness (noise suppression rate ≥90%), sampling rate adaptation accuracy.

[0069] The report generation rules and presentation format are as follows: The report generation cycle is 1 hour / report for normal operating conditions and 10 minutes / report for high-rate operating conditions (≥5C); The presentation format is as follows: Vehicle terminal: Simplified version (only displaying SOH, SOC, and abnormal prompts), transmitted to the vehicle instrument panel via CAN; Cloud / Host computer: Full version (including all indicators + trend curves), supporting Excel export and web page visualization; Core logic: The report data is from the same source as the "high-precision release data", and the trend analysis is based on fitting of the historical data of the past 7 days.

[0070] The early warning mechanism is triggered through tiered responses and multi-dimensional triggering conditions: The warning trigger conditions (related to the threshold parameters mentioned above) for different warning types are as follows: When the warning type is temperature gradient warning, the trigger threshold is: gradient > 2℃ / cm (warning) and > 5℃ / cm (danger). The data comes from the temperature gradient calculation results in step 1. When the warning type is voltage anomaly warning, the trigger threshold is: voltage change rate at high magnification > 5V / s, and the data comes from voltage data after dynamic range optimization. When the warning type is aging warning, the trigger threshold is: SOH < 85% (warning) and < 80% (danger), and the data comes from the real-time update value of BMS; When the warning type is noise exceeding the standard warning, the trigger threshold is: residual noise after filtering > 10mV, and the data comes from the detection results after wavelet denoising. When the warning type is sampling anomaly warning, the trigger threshold is: ADC sampling frequency adaptation error, and the data comes from the sampling rate configuration result of step 3.

[0071] The warning level and response actions are as follows: When the warning level is a minor warning, the response is as follows: the yellow light on the vehicle's instrument panel is displayed, the warning log is recorded in the cloud, it does not affect the normal use of the battery, and the abnormal item is marked in the next report; When the warning level is severe, the response actions are as follows: a red light is displayed on the vehicle's instrument panel, an audible alarm is added, the information is pushed to the cloud via Ethernet, and the charging / discharging rate is limited (≤3C). When the warning level is an emergency warning, the response actions are as follows: disconnect the charging and discharging circuit (if the temperature gradient is >5℃ / cm), trigger the vehicle fault code, and push an SMS to the maintenance personnel via the cloud.

[0072] The warning reset conditions are as follows: for minor warnings, the warning will automatically reset after the abnormal indicators return to normal; for severe / emergency warnings, after manual confirmation that the fault has been eliminated, a reset command will be sent through the host computer, and the warning will be cleared after the BMS restarts.

[0073] Step 6: Check the calibration conditions, which include stable operating conditions, no abnormalities in the acquisition system, and calibration trigger conditions. Collect K samples (K≥100), calculate the average offset, perform zero-point / gain calibration, and update the compensation parameters.

[0074] The specific calibration conditions are as follows: The stable operating conditions (prerequisite for the validity of the calibration benchmark) are as follows: Battery resting: no charging or discharging operation, absolute current value ≤0.05A; SOC range: 20%≤SOC≤80% (avoiding nonlinear extreme value range); temperature stability: core temperature 20℃±5℃, fluctuation within 10min ≤0.2℃; health status: SOH≥80% (avoiding the distortion of aging battery characteristics affecting calibration).

[0075] The acquisition system is free of abnormal conditions (data quality assurance) as follows: Hardware is normal: ADC, PGA, and DMA modules are fault-free, and the sampling frequency is consistent with the configuration; Noise meets the standard: residual noise after filtering is ≤5mV (matching the wavelet noise reduction accuracy requirements mentioned above); Sensors are intact: voltage / current / temperature sensors have no open circuits, short circuits, or signal overflow warnings.

[0076] The calibration trigger conditions (necessary scenarios for starting calibration) are as follows: periodic trigger: every 3 months or 500 cumulative charge-discharge cycles; abnormal trigger: zero offset > 0.05V (voltage) / > 0.1A (current), or trigger the "sampling abnormality warning" mentioned above; passive trigger: after replacing the voltage / current sensor, PGA module or ADC chip.

[0077] Zero-point drift calibration is represented as: ; The zero-point drift calibration is triggered if any one of the following conditions is met: (1) continued ; (2) ; (3) .

[0078] After selecting the minimum suitable range in the dynamic range optimization step, the gain error compensation process for the PGA (Programmable Gain Amplifier) ​​is as follows: The theoretically configured gain (1x / 2x / 5x / 10x) tied to the range is used; a reference voltage source is connected; the output voltage is measured; the gain error is calculated; and the gain coefficient is updated. The formula is: ; in, This is the reference voltage; The updated PGA gain coefficient is used to compensate for the actual effective gain written into the PGA module, which is then used for precise amplification of the subsequent voltage signal. The initial gain configured according to the PGA theory before compensation is the initial gain bound to the minimum suitable range in the "Dynamic Range Optimization" sub-step of step 1, such as 2 times for the 0-2V range and 5 times for the ±1V range. The output voltage of the standard reference voltage source is known in precise value and needs to be matched with the smallest suitable range selected in step 1. For example, when selecting the 0-2V range, Take the precise value within this range to provide a calibration reference; Reference voltage Based on the current PGA ( After amplification, the actual output voltage measured by the ADC (i.e., the actual amplification result under gain error).

[0079] The triggering conditions for gain error compensation are: after the external reference source is available, the constant voltage charging stage is met, or the monthly scheduled execution is met.

[0080] Compared with existing technologies, this embodiment, through three-dimensional temperature field modeling and aging sensing rate calculation, sigmoid sampling rate control and predictive range management, multi-physics coupling compensation and wavelet-Kalman denoising, confidence dynamic learning and multi-dimensional anomaly detection, and background online calibration and drift prediction, can detect power battery anomalies in real time, dynamically and accurately under complex operating conditions.

[0081] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics is not elaborated upon here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A multi-condition sensor method suitable for power batteries, characterized in that, The method comprises the following steps: Step 1, synchronously reading voltage, current and temperature data of the power battery, setting sampling frequency and range; Step 2, calculating temperature gradient according to temperature data, calculating equivalent rate according to current data, and determining working condition type according to gradient temperature, equivalent rate and internal resistance change rate; Step 3, applying temperature gradient and aging drift term for compound drift compensation, performing wavelet denoising, and updating filter parameters; Step 4, calculating median of each sensor group and standard deviation of all sensors, combining three parts of confidence components, and performing normalization processing to calculate confidence of each sensor; calculating weight of each sensor, calculating weighted average value, estimating second derivative, applying dynamic smoothing constraint to perform adaptive weighted fusion, and performing abnormal detection and isolation of the power battery; Step 5, publishing high-precision data, generating health report, and triggering early warning mechanism; Step 6, checking calibration condition, performing zero point / gain calibration, and updating compensation parameters.

2. The multi-condition sensor method for power battery of claim 1, wherein: In the step 2, the calculation process of the temperature gradient is: a three-dimensional coordinate system mapping the sensor position is established for the temperature data, central difference method is used to calculate the temperature partial derivative in each direction, and a three-dimensional temperature gradient vector is synthesized based on the temperature partial derivative, which is expressed as: ； wherein, x is the rate of temperature change in the direction, is the sensor spacing, ranging from 1-2 cm, with a typical value of 1.5 cm, is the temperature value at position (i,j,k) in °C, is the temperature gradient (°C / cm), with a range of: normal operating condition: 0-2 °C / cm; pre-warning state: 2-5 °C / cm; dangerous state: >5 °C / cm; the calculation process of the equivalent rate is: Calculation based on current data The moving root mean square is used to obtain the current SOH estimate, and the aging compensation factor is applied to correct the equivalent ratio. The output is the equivalent ratio. , represented as: ； wherein, is the real-time current obtained by sampling; is the rated capacity (Ah), which is a parameter provided by the battery; is the time window, with a value range of 10-60 s, and a default value of 30 s; is the state of health, with a value range of 80%-100%; is the aging sensitivity coefficient, with a value range of 0.15-0.25, and a typical value of 0.2; the determination process of the working condition type is: a feature vector F is calculated according to the gradient temperature, the equivalent rate and the internal resistance change rate, which is expressed as: ； wherein, is the rate of change of capacity, in C / min, ranging from -10 to 10; is the rate of change of internal resistance (ranging from 0.1 to 0.5; a pre-trained classification model is used to output the current dominant working condition, and the expression of the current dominant working condition is: ； Wherein, the number of working conditions The working condition categories include low-temperature low-multiple, low-temperature high-multiple, normal-temperature low-multiple, normal-temperature high-multiple, high-temperature low-multiple, and high-temperature high-multiple, is the classification weight.

3. The multi-condition sensor method for power cells of claim 2, wherein: In step 3, the composite drift compensation process is to measure the current temperature T and the rate of change dT / dt; calculate the temperature drift; obtain SOH and current ; computing an aging drift, outputting a compensated voltage , the composite drift compensation is expressed as: ； wherein the temperature drift term is expressed as: ； the aging drift term is expressed as: ； is a temperature polynomial coefficient; is a temperature rate coefficient; c is an aging sensitivity coefficient; is a characteristic current; SOH is a state of health; the multi-scale wavelet denoising process is: a signal is decomposed by 5 layers of wavelets, noise variance of each scale is estimated, adaptive weight is calculated, and the reconstructed signal is expressed as: ； the wavelet coefficient optimization formula is: ； wherein is a wavelet coefficient at scale j; is a noise variance estimate; is a scale weight satisfying ; j is a decomposition level, j = 1-5.

4. The multi-condition sensor method for power cells of claim 3, wherein: in the step 4, the sensor confidence evaluation calculation formula is: ； wherein, is the calibrated precision factor; is the recent variance, a sliding window variance; is the median of the sensor group; is the standard deviation of all sensors; is the confidence level; the formula of the adaptive weighted fusion is: ； wherein the smoothing factor is in the range 0.05-0.15, expressed as: ； the normalized current change rate is expressed as: ； is the sensor mean.

5. The multi-condition sensor method for power cells of claim 4, wherein: in the step 4, the abnormal detection of the power battery is expressed as: ； wherein, is the median; is the median absolute deviation; is the maximum allowed rate of change (conditioned on the operating condition), 0.1 V / s for low rates and 5 V / s for high rates; is the Pearson correlation coefficient.

6. The multi-condition sensor method for power cells of claim 5, wherein: in the step 6, the zero point drift calibration is expressed as: ； the trigger condition of the zero point drift calibration is: continuously ; ； 。 7. The multi-condition sensor method for power cells of claim 5, wherein: in the step 6, the gain error compensation process is: connecting a reference voltage source, measuring output voltage, calculating gain error, and updating gain coefficient, and the expression formula is: ； wherein Vref is a reference voltage; the trigger condition of the gain error compensation is: under the premise that the external reference source is available, the constant voltage charging stage is met, or the monthly regular execution is met. 8.A multi-condition sensor system for a power battery, comprising at least one processor, and a memory connected with the at least one processor in communication, the memory storing a computer program executable by the at least one processor, characterized in that: The computer program is executed by the at least one processor, so that the at least one processor can perform the multi-working condition sensor method according to any one of claims 1-7.

Images