Energy storage power station battery consistency analysis method based on Gaussian mixture model

Abstract

The invention discloses an energy storage station battery consistency analysis method based on a Gaussian mixture model, and belongs to the technical field of energy storage station battery detection. Comprising the following steps: preparing data required by consistency analysis; carrying out abnormal value and missing value preprocessing on the prepared data; extracting battery features from the preprocessed data, wherein the features comprise static features, dynamic features and working condition adaptive features; and on the basis of the extracted battery features, performing parameter updating by using a Gaussian mixture model, and outputting a feature clustering result. The method is objective in threshold setting, can distinguish'symptoms' and'causes', more accurately positions all abnormal batteries, adapts to different operation conditions, and can perform accurate analysis.

Description

Technical Field

[0001] This invention belongs to the field of battery testing technology for energy storage stations, specifically relating to a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model. Background Technology

[0002] Battery consistency analysis in energy storage power stations is crucial for ensuring system safety, extending lifespan, and maintaining performance. Several technical solutions have emerged to address this issue: Based on time-series data feature analysis (CN117930012A), the dynamic characteristics of the battery module are effectively reflected by the time-series data of each battery, and the corresponding static characteristic data are determined by the voltage and temperature data of each battery to quantify the consistency status of the battery module. The consistency evaluation results of the battery are comprehensively reflected by dynamic and static characteristic data, thereby improving the accuracy of battery consistency results.

[0003] Based on voltage sequence statistics and testing (CN115343649A), this invention obtains the voltage sequence of each cell in a battery pack during charging and / or discharging. The voltage sequence is then cleaned to obtain a cleaned voltage sequence. The voltage deviation sequence between any two cells is calculated. A white noise test is performed on the voltage deviation sequence; if the voltage deviation sequence is white noise, the two cells are in the same state; if it is not white noise, the two cells are in different states. This invention utilizes the characteristic that inconsistent cell voltages differ from other normal cell voltages, and that this difference is not random but has a temporal relationship. It proposes a novel, feasible, and simple approach to solve the problem of inconsistent cell voltages in battery packs. Compared to traditional differential voltage threshold detection methods that directly compare the voltage deviation values ​​of two sets of cells, this method does not require setting a threshold, avoiding false alarms caused by the rigid threshold application of traditional methods.

[0004] Based on statistical parameters (variance / standard deviation) (CN116087799A), voltage data of each cell in the battery module is obtained; the voltage data is preprocessed; for all preprocessed voltage data, statistical parameters of all voltage data are calculated; the statistical parameters are variance or standard deviation; for each preprocessed voltage data, statistical parameters of the remaining voltage data after removing the preprocessed voltage data are calculated; for each preprocessed voltage data, the difference between the statistical parameters of all voltage data and the statistical parameters of the remaining voltage data is calculated; based on the difference, it is determined whether there is a consistency problem within the battery module.

[0005] Based on electrical performance data (charge and discharge) (CN117706393A), this method determines the charging data of each cell in the battery pack under test after it is fully charged and left to stand for a first preset time; it also determines the discharging data of each cell in the battery pack after it is fully discharged and left to stand for a second preset time; based on the charging and discharging data of each cell, the electrical performance data of each cell is determined; and based on the electrical performance data, the method detects whether there are consistency differences in the battery pack. This embodiment of the application determines the electrical performance data based on the charging and discharging data of the cells, achieving a comprehensive analysis of the cell charging and discharging processes. It fully considers the impact of charging and discharging data on the electrical performance evaluation of the cells, improving the accuracy of the cell electrical performance data and quantifying the electrical performance of the cells, thereby improving the accuracy and efficiency of consistency difference detection.

[0006] Based on characteristic voltage and dispersion rate (CN119291551A), the detection circuit of the battery cluster is sequentially connected by the control system to collect the characteristic voltage and current signals of each battery cluster, calculate the consistency characteristic value of each battery cluster, and analyze the dispersion rate between each cluster by the cumulative root mean square deviation. Based on phase angle features and artificial intelligence (CN117420448A), this method includes the following steps: charging and discharging the battery cell, acquiring DC signals, performing Kalman filtering on the DC signals, converting the filtered DC signals into low-inductance AC signals, performing Fourier transform on the AC signals, extracting the phase angles of high, medium and low frequencies respectively, using a nonlinear programming algorithm to obtain the optimal proportional coefficient of the phase angles at different frequencies and calculating the consistency index of the battery cell, combining a recurrent neural network module to perform deep learning and training on the consistency index to obtain the graded specification lines, grading the battery cells according to the graded specification lines, and screening the battery cells with poor consistency in the graded process.

[0007] The following issues urgently need to be addressed: ① High dependence on data quality and algorithm complexity For example, voltage sequence statistics and testing (CN115343649A) is sensitive to noise and has a large computational load. White noise testing itself requires data stationarity and requires calculating the deviation sequence between any two cells. For a system with N cells, the computational complexity is O(N²), which has poor real-time performance in large-scale systems.

[0008] ② Difficulty in distinguishing between "symptoms" and "causes" Most methods (such as voltage variance analysis) can only detect the "symptoms" of inconsistency (i.e., differences in voltage and temperature), but cannot directly diagnose the root cause of the inconsistency. Is it increased internal resistance, higher self-discharge rate, or capacity decay and loose connections? It is difficult to distinguish these issues based on external electrical parameters alone.

[0009] ③ Sensitivity to operating conditions For example, data based on electrical performance (charge and discharge) (CN117706393A) has stringent requirements for operating conditions and is not suitable for online monitoring. It requires the battery to be fully charged and fully discharged, two states that are not common in actual operation. Therefore, it is more suitable for offline periodic testing and cannot achieve real-time online monitoring.

[0010] ④ The contradiction between offline analysis and online application While some methods based on full-capacity charge-discharge curve analysis (such as CN117706393A) are accurate, they require the battery to operate within a specific and complete charge-discharge range, which is difficult to meet during the daily online operation of energy storage power stations (the power station is in a floating charge or partial charge-discharge state most of the time).

[0011] ⑤ The subjectivity and fixedness of threshold setting Many statistical methods (such as variance) require setting a consistency alarm threshold. This threshold is often set empirically; setting it too low leads to false alarms, while setting it too high results in missed alarms. Furthermore, a fixed threshold cannot accommodate the increasing natural dispersion of batteries as they age. Summary of the Invention

[0012] To address the technical problems existing in the prior art, this invention provides a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model. The threshold setting is objective, which can distinguish between "symptoms" and "causes", more accurately locate all abnormal batteries, adapt to different operating conditions, and perform precise analysis.

[0013] This invention provides a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model, comprising the following steps: Step S01: Prepare the data required for consistency analysis; Step S02: Preprocess the prepared data for outliers and missing values; Step S03: Extract battery features from the preprocessed data, including static features, dynamic features, and operating condition adaptive features; Step S04: Based on the extracted battery features, update the parameters using a Gaussian mixture model and output the feature clustering results.

[0014] Preferably, the data required for consistency analysis in step S01 includes: the real-time voltage of each cell, the total current of the battery cluster, the temperature of each cell or module, and the estimated state of charge of the overall battery cluster, ensuring that all data are stamped with the corresponding timestamps; the static features in step S03 include voltage deviation features, voltage ranking features, temperature deviation features, and spatial temperature gradient; the dynamic features include voltage change rate, relaxation characteristics, estimated dynamic internal resistance, and capacity-related features; the adaptive features for operating conditions specifically involve extracting the most suitable discriminant features for each operating condition to ensure the matching between features and operating conditions.

[0015] Preferably, the rate of voltage change is obtained by calculating the derivative of voltage with respect to time during the charging and discharging process. The calculation formula is: ≈ = The unit is mV / s or V / mi; in, For voltage changes, The interval between two consecutive statistical times. and They are time points and The voltage.

[0016] Preferably, the relaxation characteristics include relaxation amplitude, relaxation speed, and time constant: relaxation amplitude is the total change in voltage recovery; relaxation speed is the rate of voltage recovery; and time constant is the time required to achieve 63.2% voltage recovery.

[0017] Preferably, obtaining the relaxation characteristics includes the following steps: Step S301, Identify the current abrupt change point: Detect the moment when the current jumps from a non-zero value to 0A, and mark this moment as the start point t0 of the relaxation process; Step S302, extract the relaxation data segment: starting from t0, continuously collect voltage data for 5-30 minutes, ensuring that the current remains zero during this period, i.e., in a static state; Step S303, establish a mathematical model: The relaxation process is fitted using a common exponential decay model, and the voltage V(t) at time t is calculated using the following formula: V(t) = + ( ) ; in: The initial voltage at the start of relaxation, in mV; The theoretical final value for the balanced voltage, in mV; is the time constant.

[0018] Preferably, the parameters of the relaxation characteristics are calculated based on the established mathematical model, as follows: relaxation range The following formula is used for calculation: Unit: mV; The time constant τ is obtained by fitting an exponential curve using the nonlinear least squares method. Specific time point recovery rate The following formula is used for calculation: 100%, of which For a specific time point The voltage, in mV; The average recovery rate is calculated using the following formula: Average recovery rate = Unit: mV / s, where t1 and t2 are two arbitrary times, with t2 being later than t1. and These are the voltages at times t1 and t2, respectively, in mV.

[0019] Preferably, the instantaneous internal resistance is calculated based on the changes in voltage and current, and the dynamic internal resistance is then determined. The estimation is performed using the following calculation formula: Unit: mΩ; in, and The voltages are the first statistical time before the mutation and the second statistical time after the mutation, respectively. and The currents are the first statistical time before the mutation and the second statistical time after the mutation, respectively.

[0020] Preferably, dynamic internal resistance The estimation specifically includes the following steps: Step 311, Identify current abrupt change events: Detect the time period during which the current changes significantly, where a significant change is defined as a change magnitude greater than a preset value, and record the stable state data before and after the abrupt change; Step 312, extract key data points: extract the voltage V1 and current I1 when the current is stable before the sudden change, and extract the voltage V2 and current I2 when the current is stable after the sudden change, to ensure that both state points are in a relatively stable period; Step 313, calculate the dynamic internal resistance; Step 314, Temperature and SOC Compensation: Record the battery temperature when calculating the dynamic internal resistance; record the SOC range when calculating the dynamic internal resistance; establish temperature compensation benchmarks under different conditions.

[0021] Preferably, the capacity-related characteristics include actual capacity, capacity retention rate, capacity decay rate, and charge / discharge capacity ratio; actual capacity refers to the total amount of electricity that the cell can store or release under specific conditions; capacity retention rate refers to the ratio of current capacity to rated capacity; capacity decay rate refers to the rate at which capacity decreases with the number of cycles; the capacity-related characteristics are obtained using the following method: Data collection conditions: Complete charge-discharge cycle: The entire process from full charge to complete discharge; Constant current: ensures stable charging and discharging current; Standard temperature: Tested within a specified temperature range; Resting period: A resting period longer than the preset time before and after charging and discharging; Actual capacity The following formula is used for calculation: ; Capacity retention The following formula is used for calculation: ; Capacity decay rate The following formula is used for calculation: ; The charge / discharge ratio (Charge_Discharge_Ratio) is calculated using the following formula: Charge_Discharge_Ratio= ; in, A constant current under specific conditions; and These are the starting and ending points of the time frame for calculating the actual capacity; Rated capacity; This is the initial capacity; Current capacity; This refers to the charging capacity. This represents the discharge capacity.

[0022] Preferably, step S04 specifically includes the following steps: Step S401: Initialize the extracted battery features; Step S402: Calculate the probability that each cell belongs to each cluster; Step S403: Update the Gaussian mixture model parameters; Step S404: Determine whether the Gaussian mixture model has converged. If it has not converged, return to step S402. If it has converged, output the updated Gaussian mixture model and the clustering results of the battery features. Step S405: Analyze and diagnose the output clustering results.

[0023] Compared with the prior art, the present invention has the following beneficial effects: This invention deeply integrates probabilistic modeling, multi-dimensional feature engineering, dynamic operating condition adaptation, and closed-loop diagnostics to construct a more intelligent, accurate, and practical battery consistency analysis system than traditional solutions. Its technical advantage lies in elevating battery consistency management from a "remedial" maintenance method to a proactive asset optimization strategy of "early warning and in-process diagnosis," thereby providing core guarantees for the safety, lifespan, and economic benefits of energy storage power stations. Specifically: 1. Objective threshold setting Gaussian mixture models automatically discover natural clusters in data through probabilistic models. Instead of relying on fixed variance or scatter rate thresholds to determine "consistency," they calculate the posterior probability of each data point (battery) belonging to various Gaussian distributions. This fundamentally avoids the problem of subjectively setting thresholds, making it more adaptable.

[0024] 2. Able to distinguish between "symptoms" and "causes" Traditional methods may only provide a binary judgment of "consistent" or "inconsistent." Gaussian mixture models, however, can identify multiple different "inconsistency patterns." For example, it might automatically classify batteries into: 1) the main cluster (healthy and consistent group); 2) the slightly aged group; 3) the group with high internal resistance; and 4) the potential fault group. This provides deeper insights into pinpointing the "cause of the problem."

[0025] 3. More accurately locate all abnormal batteries When there are multiple different anomalous groups in the system, the method based on the overall variance will fail. Gaussian mixture models are designed by nature to model multimodal distributions. They can capture and separate multiple different subgroups well, so as to more accurately locate all anomalous batteries, rather than just finding the most anomalous one.

[0026] 4. Adaptable to different operating conditions Gaussian mixture models can be established for different operating conditions (such as different SOC ranges, temperature ranges, and current rates) to analyze the dynamic changes of the battery cluster under different operating conditions. If a battery "jumps" from the main cluster to the abnormal cluster at a high rate, this information is valuable and reflects the problem more clearly than static analysis.

[0027] 5. Capable of performing precise analysis Gaussian mixture models can directly model characteristic data at the cell level (such as voltage, internal resistance, and temperature change rate), thus enabling state assessment and clustering of each cell for refined analysis. Attached Figure Description

[0028] Figure 1 This is a flowchart of a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model, as an embodiment of the present invention.

[0029] Figure 2 This is an embodiment of the present invention. Figure 2 This is a flowchart of the core steps of GMM clustering. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0031] This invention provides a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model, comprising the following steps: Step S01: Prepare the data required for consistency analysis; Step S02: Preprocess the prepared data for outliers and missing values; Step S03: Extract battery features from the preprocessed data, including static features, dynamic features, and operating condition adaptive features; Step S04: Based on the extracted battery features, update the parameters using a Gaussian mixture model and output the feature clustering results.

[0032] In one embodiment of the present invention, the data required for consistency analysis in step S01 includes: the real-time voltage of each cell, the total current of the battery cluster, the temperature of each cell or module, and the estimated state of charge of the battery cluster as a whole, ensuring that all data are timestamped accordingly; the static features in step S03 include voltage deviation features, voltage ranking features, temperature deviation features, and spatial temperature gradient; the dynamic features include voltage change rate, relaxation characteristics, estimated dynamic internal resistance, and capacity-related features; the adaptive operating condition features specifically extract the most suitable discriminant features for each operating condition to ensure the matching between features and operating conditions.

[0033] In one embodiment of the present invention, the rate of voltage change is obtained by calculating the derivative of voltage with respect to time during the charging and discharging process. The calculation formula is: ≈ = The unit is mV / s or V / mi; in, For voltage changes, The interval between two consecutive statistical times. and They are time points and The voltage.

[0034] In one embodiment of the present invention, the relaxation characteristics include relaxation amplitude, relaxation speed and time constant: relaxation amplitude is the total change in voltage recovery; relaxation speed is the rate of voltage recovery; and time constant is the time required to achieve 63.2% voltage recovery.

[0035] In one embodiment of the present invention, obtaining the relaxation characteristics includes the following steps: Step S301, Identify the current abrupt change point: Detect the moment when the current jumps from a non-zero value to 0A, and mark this moment as the start point t0 of the relaxation process; Step S302, extract the relaxation data segment: starting from t0, continuously collect voltage data for 5-30 minutes, ensuring that the current remains zero during this period, i.e., in a static state; Step S303, establish a mathematical model: The relaxation process is fitted using a common exponential decay model, and the voltage V(t) at time t is calculated using the following formula: V(t) = + ( ) ; in: The initial voltage at the start of relaxation, in mV; The theoretical final value for the balanced voltage, in mV; is the time constant.

[0036] In one embodiment of the present invention, the parameters of the relaxation characteristics are calculated according to the established mathematical model, as follows: relaxation range The following formula is used for calculation: Unit: mV; The time constant τ is obtained by fitting an exponential curve using the nonlinear least squares method. Specific time point recovery rate The following formula is used for calculation: 100%, of which For a specific time point The voltage, in mV; The average recovery rate is calculated using the following formula: Average recovery rate = Unit: mV / s, where t1 and t2 are two arbitrary times, with t2 being later than t1. and These are the voltages at times t1 and t2, respectively, in mV.

[0037] In one embodiment of the present invention, the instantaneous internal resistance is calculated based on the changes in voltage and current, and the dynamic internal resistance is then determined. The estimation is performed using the following calculation formula: Unit: mΩ; in, and The voltages are the first statistical time before the mutation and the second statistical time after the mutation, respectively. and The currents are the first statistical time before the mutation and the second statistical time after the mutation, respectively.

[0038] In one embodiment of the present invention, dynamic internal resistance The estimation specifically includes the following steps: Step 311, Identify current abrupt change events: Detect the time period during which the current changes significantly, where a significant change is defined as a change magnitude greater than a preset value, and record the stable state data before and after the abrupt change; Step 312, extract key data points: extract the voltage V1 and current I1 when the current is stable before the sudden change, and extract the voltage V2 and current I2 when the current is stable after the sudden change, to ensure that both state points are in a relatively stable period; Step 313, calculate the dynamic internal resistance; Step 314, Temperature and SOC Compensation: Record the battery temperature when calculating the dynamic internal resistance; record the SOC range when calculating the dynamic internal resistance; establish temperature compensation benchmarks under different conditions.

[0039] In one embodiment of the present invention, capacity-related characteristics include actual capacity, capacity retention rate, capacity decay rate, and charge-discharge capacity ratio; actual capacity refers to the total amount of electricity that the cell can store or release under specific conditions; capacity retention rate refers to the ratio of current capacity to rated capacity; capacity decay rate refers to the rate at which capacity decreases with the number of cycles; the capacity-related characteristics are obtained using the following method: Data collection conditions: Complete charge-discharge cycle: The entire process from full charge to complete discharge; Constant current: ensures stable charging and discharging current; Standard temperature: Tested within a specified temperature range; Resting period: A resting period longer than the preset time before and after charging and discharging; Actual capacity The following formula is used for calculation: ; Capacity retention The following formula is used for calculation: ; Capacity decay rate The following formula is used for calculation: ; The charge / discharge ratio (Charge_Discharge_Ratio) is calculated using the following formula: Charge_Discharge_Ratio= ; in, A constant current under specific conditions; and These are the starting and ending points of the time frame for calculating the actual capacity; Rated capacity; This is the initial capacity; Current capacity; This refers to the charging capacity. This represents the discharge capacity.

[0040] In one embodiment of the present invention, step S04 specifically includes the following steps: Step S401: Initialize the extracted battery features; Step S402: Calculate the probability that each cell belongs to each cluster; Step S403: Update the Gaussian mixture model parameters; Step S404: Determine whether the Gaussian mixture model has converged. If it has not converged, return to step S402. If it has converged, output the updated Gaussian mixture model and the clustering results of the battery features. Step S405: Analyze and diagnose the output clustering results.

[0041] Example 1 This invention provides a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model, comprising the following steps: Step S01: Prepare the data required for consistency analysis; Step S02: Preprocess the prepared data for outliers and missing values; Step S03: Extract battery features from the preprocessed data, including static features, dynamic features, and operating condition adaptive features; Step S04: Based on the extracted battery features, update the parameters using a Gaussian mixture model and output the feature clustering results.

[0042] Example 2 This invention provides a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model, comprising the following steps: Step S01: Prepare the data required for consistency analysis; Step S02: Preprocess the prepared data for outliers and missing values; Step S03: Extract battery features from the preprocessed data, including static features, dynamic features, and operating condition adaptive features; Step S04: Based on the extracted battery features, update the parameters using a Gaussian mixture model and output the feature clustering results.

[0043] In this embodiment, the data required for consistency analysis in step S01 includes: the real-time voltage of each cell, the total current of the battery cluster, the temperature of each cell or module, and the estimated state of charge of the battery cluster as a whole, ensuring that all data are stamped with the corresponding timestamps; the static features in step S03 include voltage deviation features, voltage ranking features, temperature deviation features, and spatial temperature gradient; the dynamic features include voltage change rate, relaxation characteristics, estimated dynamic internal resistance, and capacity-related features; the adaptive operating condition features specifically extract the most suitable discriminant features for each operating condition to ensure the matching between features and operating conditions.

[0044] In this embodiment, the rate of voltage change is obtained by calculating the derivative of voltage with respect to time during the charging and discharging process. The calculation formula is: ≈ = The unit is mV / s or V / mi; in, For voltage changes, The interval between two consecutive statistical times. and They are time points and The voltage.

[0045] In this embodiment, the relaxation characteristics include relaxation amplitude, relaxation speed, and time constant: relaxation amplitude is the total change in voltage recovery; relaxation speed is the rate of voltage recovery; and time constant is the time required to achieve 63.2% voltage recovery.

[0046] In this embodiment, obtaining the relaxation characteristics includes the following steps: Step S301, Identify the current abrupt change point: Detect the moment when the current jumps from a non-zero value to 0A, and mark this moment as the start point t0 of the relaxation process; Step S302, extract the relaxation data segment: starting from t0, continuously collect voltage data for 5-30 minutes, ensuring that the current remains zero during this period, i.e., in a static state; Step S303, establish a mathematical model: The relaxation process is fitted using a common exponential decay model, and the voltage V(t) at time t is calculated using the following formula: V(t) = + ( ) ; in: The initial voltage at the start of relaxation, in mV; The theoretical final value for the balanced voltage, in mV; is the time constant.

[0047] In this embodiment, the parameters of the relaxation characteristics are calculated based on the established mathematical model, as follows: relaxation range The following formula is used for calculation: Unit: mV; The time constant τ is obtained by fitting an exponential curve using the nonlinear least squares method. Specific time point recovery rate The following formula is used for calculation: 100%, of which For a specific time point The voltage, in mV; The average recovery rate is calculated using the following formula: Average recovery rate = Unit: mV / s, where t1 and t2 are two arbitrary times, with t2 being later than t1. and These are the voltages at times t1 and t2, respectively, in mV.

[0048] In this embodiment, the instantaneous internal resistance is calculated based on the changes in voltage and current, and the dynamic internal resistance is then determined. The estimation is performed using the following calculation formula: Unit: mΩ; in, and The voltages are the first statistical time before the mutation and the second statistical time after the mutation, respectively. and The currents are the first statistical time before the mutation and the second statistical time after the mutation, respectively.

[0049] In this embodiment, dynamic internal resistance The estimation specifically includes the following steps: Step 311, Identify current abrupt change events: Detect the time period during which the current changes significantly, where a significant change is defined as a change magnitude greater than a preset value, and record the stable state data before and after the abrupt change; Step 312, extract key data points: extract the voltage V1 and current I1 when the current is stable before the sudden change, and extract the voltage V2 and current I2 when the current is stable after the sudden change, to ensure that both state points are in a relatively stable period; Step 313, calculate the dynamic internal resistance; Step 314, Temperature and SOC Compensation: Record the battery temperature when calculating the dynamic internal resistance; record the SOC range when calculating the dynamic internal resistance; establish temperature compensation benchmarks under different conditions.

[0050] In this embodiment, capacity-related characteristics include actual capacity, capacity retention rate, capacity decay rate, and charge / discharge capacity ratio; actual capacity refers to the total amount of electricity that a cell can store or release under specific conditions; capacity retention rate refers to the ratio of current capacity to rated capacity; capacity decay rate refers to the rate at which capacity decreases with the number of cycles; the following method is used to obtain capacity-related characteristics: Data collection conditions: Complete charge-discharge cycle: The entire process from full charge to complete discharge; Constant current: ensures stable charging and discharging current; Standard temperature: Tested within a specified temperature range; Resting period: A resting period longer than the preset time before and after charging and discharging; Actual capacity The following formula is used for calculation: ; Capacity retention The following formula is used for calculation: ; Capacity decay rate The following formula is used for calculation: ; The charge / discharge ratio (Charge_Discharge_Ratio) is calculated using the following formula: Charge_Discharge_Ratio= ; in, A constant current under specific conditions; and These are the starting and ending points of the time frame for calculating the actual capacity; Rated capacity; This is the initial capacity; Current capacity; This refers to the charging capacity. This represents the discharge capacity.

[0051] In this embodiment, step S04 specifically includes the following steps: Step S401: Initialize the extracted battery features; Step S402: Calculate the probability that each cell belongs to each cluster; Step S403: Update the Gaussian mixture model parameters; Step S404: Determine whether the Gaussian mixture model has converged. If it has not converged, return to step S402. If it has converged, output the updated Gaussian mixture model and the clustering results of the battery features. Step S405: Analyze and diagnose the output clustering results.

[0052] Example 3 This invention provides a battery consistency analysis method for energy storage power stations based on a Gaussian mixture model, comprising the following steps: (a) Data preparation The data required for battery consistency analysis includes: ① Voltage data: Real-time voltage of each cell, with a sampling frequency of no less than 1Hz; ② Current data: Total current of the battery cluster, used to identify the system's operating conditions; ③Temperature data: Temperature of each cell or module, monitoring thermal distribution characteristics; ④SOC data: Estimated overall state of charge of the battery cluster; Ensure that all data is timestamped and that the clock on the data acquisition server is synchronized.

[0053] The following table shows examples of voltage data (all other data are based on voltage data, with cell and time series as the axes): Table 1. Sample Voltage Data

[0054] (ii) Data preprocessing 1. Outlier filtering: The 3σ principle is used to identify and remove obvious outlier data points.

[0055] The 3σ rule, also known as the 68-95-99.7 rule, is an empirical rule based on the normal distribution. It describes the probability that a data point falls within a specific range around the mean (μ) of normally distributed data. For a dataset that conforms to a normal distribution, the distribution of its data points is as follows: ①1σ range (mean ± 1 standard deviation) Approximately 68.27% of the data points will fall within this range, P(μ - σ ≤ X ≤ μ + σ) ≈68.27%.

[0056] ②2σ range (mean ± 2 standard deviations) Approximately 95.45% of the data points will fall within this range, P(μ - 2σ ≤ X ≤ μ + 2σ) ≈95.45%.

[0057] ③ 3σ range (mean ± 3 standard deviations) Approximately 99.73% of the data points will fall within this range, P(μ - 3σ ≤ X ≤ μ + 3σ) ≈99.73%.

[0058] The specific steps are as follows: Step 1 (Calculation): For a data sequence (such as the voltage value of a certain battery cell over a period of time), first calculate its mean (μ) and standard deviation (σ).

[0059] Step 2 (Setting Boundaries): Then calculate the upper boundary (μ + 3σ) and the lower boundary (μ - 3σ).

[0060] Step 3 (Identification and Processing): Next, iterate through all data points. Any data point that is greater than μ + 3σ or less than μ - 3σ will be identified as an outlier.

[0061] 2. Missing value handling: Use linear interpolation to fill in data gaps.

[0062] Linear interpolation is a very simple and commonly used numerical method for estimating new data points between two known data points. The core idea is to assume that the change between two known points is linear (the state data of the battery cell all satisfy linearity), and then calculate the value of the intermediate point according to the proportional relationship.

[0063] If there are two known points: (x0, y0) and (x1, y1), to estimate the corresponding y value at position x, the formula is: y = + * ( ) The specific processing steps are as follows: Suppose that in the battery voltage data, the data with timestamp T3 in the table below is identified as an outlier (e.g., by the 3σ principle) and needs to be corrected by interpolation.

[0064] Table 2. Examples of voltage data with outliers

[0065] Step 1: Determine the known points Previous normal value: (T2, 3.305); Next normal value: (T4, 3.285).

[0066] Step 2: Calculate the scaling factor Ratio = = = 0.5 (assuming uniform time intervals: T2→T3→T4).

[0067] Step 3: Apply the linear interpolation formula = 3.305 + 0.5 × (3.285 - 3.305); = 3.305 + 0.5 × ( -0.020 ); = 3.305 - 0.010 = 3.295V.

[0068] The corrected data is shown in the table below: Table 3. Voltage data after outlier correction

[0069] 3. Data preprocessing needs to focus on whether the data is real abnormal data or noisy data. If it is noisy data, it must be removed, otherwise it will cause errors in subsequent inference and training. Battery monitoring of energy storage stations is an extremely rigorous process, and any small error may lead to a safety accident.

[0070] The principle for differentiation is to look at the "persistence" and "pattern" of the anomaly. If a voltage anomaly is just an isolated, instantaneous spike (for example, the data is normal one second and then there is a significant change in that one second), it is likely noise and should be eliminated. However, if the voltage of a cell deviates persistently (for example, its voltage is consistently significantly lower than other cells throughout the discharge process), or if its anomaly exhibits a clear pattern (such as a sudden voltage drop and temperature surge when subjected to high current discharge), then it is highly likely to be a real fault and must be retained and analyzed in detail. In short: "A fleeting anomaly is noise, a persistent one is a fault."

[0071] (III) Feature Extraction 1. Static Feature Extraction Voltage deviation characteristics: Calculate the absolute and relative deviations of each cell's voltage from the average value; Voltage ranking feature: Sort the battery cells by voltage and calculate the percentile ranking position; Temperature deviation characteristics: Calculate the difference between the temperature of each cell and the average temperature of the system; Spatial temperature gradient: Analyze the rate of temperature change between adjacent cells.

[0072] 2. Dynamic Feature Extraction ① Rate of change of voltage: The derivative of voltage with respect to time during the charging and discharging process.

[0073] The calculation formula is: ≈ =

[0074] Units: mV / s or V / min.

[0075] Sample data is shown in the table below: Table 4. Example table of voltage change rate data

[0076] ②Relaxation characteristics: Characteristic parameters for analyzing the voltage recovery process after a sudden change in current. It refers to the process by which the battery voltage gradually recovers to an equilibrium state when the charging and discharging current suddenly becomes zero. This process reflects the reduction of polarization phenomena inside the battery.

[0077] Key parameters: Relaxation amplitude: The total change in voltage recovery; Relaxation speed: the rate at which the voltage recovers; Time constant: The time required to achieve 63.2% recovery.

[0078] Detailed analysis steps Step S301: Identify the current abrupt change point: Detect the moment when the current jumps from a non-zero value to 0A, and mark this moment as the start point t0 of the relaxation process; Step S302: Extract relaxation data segment: Starting from t0, continuously collect voltage data for 5-30 minutes to ensure that the current remains zero during this period (resting state); Step S303: Establish a mathematical model: The relaxation process is usually fitted using an exponential decay model. The voltage V(t) at time t is calculated using the following formula: V(t) = + ( ) *

[0079] in: The initial voltage at the start of relaxation, in mV; The theoretical final value for the balanced voltage, in mV; is the time constant.

[0080] The formulas for calculating key feature parameters are as follows: 1. Relaxation amplitude ; 2. Time constant τ The result was obtained by fitting the exponential curve using the nonlinear least squares method. 3. Recovery rate at a specific time point * 100%; Common time points: 30 seconds, 1 minute, 5 minutes, 10 minutes 4. Mean recovery rate Average speed = (Unit: mV / s); Sample data is shown in the table below: Table 5. Example of average recovery rate

[0081] ③ Dynamic internal resistance estimation: Calculate instantaneous internal resistance based on changes in voltage and current; Dynamic internal resistance reflects the battery's ability to maintain voltage stability under varying load conditions and is a key indicator of battery health.

[0082] Calculation formula: Unit: mΩ (milliohm).

[0083] Detailed calculation steps Step S311: Identify current abrupt change events: Detect the time period during which the current changes significantly, ensuring that the change amplitude is large enough (usually >0.1C), and record the steady state before and after the abrupt change; Step S312: Extract key data points: Before the mutation: voltage V1 and current I1 when the current is stable; After the mutation: voltage V2 and current I2 when the current is stable; Ensure that both state points are in a relatively stable period.

[0084] Step S313: Calculate the dynamic internal resistance using the following formula:

[0085] Step S314: Temperature and SOC Compensation: Record the battery temperature during calculation; record the SOC range during calculation; establish temperature compensation benchmarks under different conditions.

[0086] The calculation examples are shown in the table below: Calculation of internal resistance at the start of discharge. Table 6. Example of dynamic internal resistance calculation

[0087] Calculation process: ; ; = 0.00025Ω = 0.25mΩ; ④ Capacity-related characteristics: Analyzing the capacity consistency through charge-discharge curves reflects the differences in the actual usable capacity of each cell in the battery pack, which is a key factor affecting the overall performance of the system.

[0088] Key metrics include: Actual capacity: The total amount of electricity that a battery cell can store / release under specific conditions; Capacity retention rate: The ratio of current capacity to rated capacity; Capacity decay rate: The rate at which capacity decreases with the number of cycles; The specific analysis steps are as follows: Step 1: Data Acquisition Conditions Complete charge-discharge cycle: The entire process from full charge to complete discharge; Constant current: ensures stable charging and discharging current; Standard temperature: Tested within a specified temperature range; Resting period: Sufficient resting time before and after charging and discharging.

[0089] Step 2: Capacity Calculation Method Ah-hour integration:

[0090] Discretization formula:

[0091] Capacity estimation based on voltage curves:

[0092] The key capacity characteristic parameters are calculated as follows: Actual capacity The following formula is used for calculation: ; Capacity retention The following formula is used for calculation: ; Capacity decay rate The following formula is used for calculation: ; The charge / discharge ratio (Charge_Discharge_Ratio) is calculated using the following formula: Charge_Discharge_Ratio= ; in, A constant current under specific conditions; and These are the starting and ending points of the time frame for calculating the actual capacity; Rated capacity; This is the initial capacity; Current capacity; This refers to the charging capacity. This represents the discharge capacity.

[0093] The following table shows sample data from the charging process: Table 7. Sample Data Table of Charging Process

[0094] The sample data of the discharge process is shown in the table below: Table 8. Sample Data Table of Discharge Process

[0095] Capacity characteristic calculation: Actual discharge capacity: 95Ah; Capacity retention rate: 95 / 100 × 100% = 95%; Charge / discharge efficiency: 95 / 105 × 100% = 90.5% 3. Adaptive operating condition feature Based on different operating conditions (such as high-rate discharge, float charge, low SOC range, etc.), the most suitable discrimination features for each operating condition are extracted to ensure the matching between features and operating conditions.

[0096] (iv) Core algorithm flow of Gaussian mixture model The Gaussian Mixture Model assumes that each cluster corresponds to a Gaussian distribution, and the entire dataset is a mixture of multiple Gaussian distributions. The EM algorithm iteratively searches for the parameters of these mixture distributions. Figure 2 This is a flowchart of the core steps of GMM clustering, which shows the complete iterative process from data preparation to model convergence: Let's explain in detail below. Figure 2 Each step in: 1. Initialization parameters: Initial parameters need to be set for each Gaussian component (cluster), including: Mean vector: It can usually be initialized with the result of K-Means clustering, or by randomly selecting data points.

[0097] Covariance matrix: Usually initialized as the identity matrix.

[0098] Mixed weights: can be initialized to equal values, for example, if the number of clusters K=3, then each weight is 1 / 3.

[0099] 2. E-step (Expectation Step): Based on the current model parameters, calculate the posterior probability of each data point belonging to each cluster.

[0100] For the feature vector x of the i-th cell i The probability γ that it belongs to the kth cluster. ik The calculation formula is: /

[0101] Here g(x) i | μ k , Σ k ) is the probability density function of the k-th Gaussian distribution. In battery analysis, x i The feature vector (such as voltage, internal resistance, temperature, etc.) represents the i-th cell. The E-step calculates the probability that each cell belongs to a cluster such as "healthy", "high internal resistance", or "capacity decay".

[0102] 3. M-step (Maximization Step): Using the attribution probabilities calculated in the E-step, update the parameters of each Gaussian component to maximize the expectation of the log-likelihood function.

[0103] Update the mixed weights ω k :ω k = (1 / N) ∑ i γ ik (N is the total number of battery cells) Update the mean vector μ k μ k = (∑ i γ ik x i ) / (∑ i γ ik ) Update covariance matrix Σ k :Σ k = [∑ i γ ik (x i - μ k ) (x i - μ k ) ] / (∑i γ ik ) In battery analysis, the M-step involves re-characterizing the feature center (mean) and distribution shape (covariance) of each cluster (e.g., "healthy cluster" or "high internal resistance cluster") based on the probability that each component belongs to a particular cluster, as well as the proportion of each cluster in the overall analysis (mixed weights). 4. Check convergence conditions: Repeat steps E and M until convergence conditions are met. Common convergence conditions include: The changes in model parameters (such as mean and weight) are less than the set threshold.

[0104] The log-likelihood function value changes very little.

[0105] The preset maximum number of iterations has been reached.

[0106] 5. Key parameters and setting suggestions (1) Number of clusters K: This is the most important parameter, representing how many clusters are expected to be found from the data. In battery consistency analysis, K can be set to 3 or 4, for example, to represent several categories such as "healthy", "slightly aged", and "obviously abnormal". The value of K can be selected by means of the elbow method (observing the inflection point of the model loss function under different K values) or information criteria (such as BIC).

[0107] (2) Covariance type: This determines the morphological flexibility of each Gaussian distribution. Common types include 'full' (complete covariance, the most flexible), 'tied' (all components share the same covariance matrix), 'diag' (the covariance matrix of each component is a diagonal matrix), and 'spherical' (spherical covariance). In battery analysis, 'full' or 'diag' is usually used to capture the complex relationships between features.

[0108] (3) Convergence threshold and maximum number of iterations: These parameters control the stopping conditions of training. The convergence threshold is usually set to a small positive number, such as 1e-3, while the maximum number of iterations prevents the program from running indefinitely, such as 100-200 times.

[0109] Specific execution process: Suppose we have characteristic data such as voltage deviation, internal resistance, and temperature deviation for 92 battery cells, and we set the cluster number K=3. We hope that GMM can classify them into "healthy main cluster", "high internal resistance cluster" and "capacity decay cluster".

[0110] Initialization: Randomly initialize the parameters of the three Gaussian components, or use K-Means for a rough initialization.

[0111] Iteration: Step E: Calculate the probability that each cell belongs to one of the three clusters. For example, cell 3 may belong to the "high internal resistance cluster" with a 90% probability and the "healthy cluster" with a 10% probability.

[0112] M-step: Update the mean, covariance, and weights of the three clusters based on the probability of all cells belonging to each cluster.

[0113] Convergence: Iteration stops when the parameter changes very little. Assuming the final weights are [0.75, 0.15, 0.10], this means that approximately 75% of the cells are healthy, 15% of the cells have high internal resistance, and 10% of the cells may have capacity degradation.

[0114] Analysis: Inspecting the cells in the "high internal resistance cluster" may reveal that their voltage is significantly lower and their temperature is higher during discharge. Maintenance personnel can then focus their inspection and maintenance on these cells.

[0115] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A battery consistency analysis method for energy storage power stations based on Gaussian mixture models, characterized in that, Includes the following steps: Step S01: Prepare the data required for consistency analysis; Step S02: Preprocess the prepared data for outliers and missing values; Step S03: Extract battery features from the preprocessed data, including static features, dynamic features, and operating condition adaptive features; Step S04: Based on the extracted battery features, update the parameters using a Gaussian mixture model and output the feature clustering results.

2. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 1, characterized in that, The data required for consistency analysis in step S01 includes: the real-time voltage of each cell, the total current of the battery cluster, the temperature of each cell or module, and the estimated state of charge of the overall battery cluster, ensuring that all data are stamped with the corresponding timestamps; the static features in step S03 include voltage deviation features, voltage ranking features, temperature deviation features, and spatial temperature gradient; the dynamic features include voltage change rate, relaxation characteristics, estimated dynamic internal resistance, and capacity-related features; the adaptive features for operating conditions specifically involve extracting the most suitable discriminant features for each operating condition to ensure the matching between features and operating conditions.

3. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 2, characterized in that, The rate of change of voltage is obtained by calculating the derivative of voltage with respect to time during the charging and discharging process. The calculation formula is: ≈ = The unit is mV / s or V / mi; in, For voltage changes, The interval between two consecutive statistical times. and They are time points and The voltage.

4. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 2, characterized in that, Relaxation characteristics include relaxation amplitude, relaxation rate, and time constant: relaxation amplitude is the total change in voltage recovery; relaxation rate is the speed at which voltage recovers; and time constant is the time required to achieve 63.2% voltage recovery.

5. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 4, characterized in that, The process of obtaining relaxation properties includes the following steps: Step S301, Identify the current abrupt change point: Detect the moment when the current jumps from a non-zero value to 0A, and mark this moment as the start point t0 of the relaxation process; Step S302, extract the relaxation data segment: starting from t0, continuously collect voltage data for 5-30 minutes, ensuring that the current remains zero during this period, i.e., in a static state; Step S303, establish a mathematical model: The relaxation process is fitted using a common exponential decay model, and the voltage V(t) at time t is calculated using the following formula: V(t)= + ( ) ; in: The initial voltage at the start of relaxation, in mV; The theoretical final value for the balanced voltage, in mV; is the time constant.

6. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 5, characterized in that, The parameters of the relaxation characteristics are calculated based on the established mathematical model, as follows: relaxation range The following formula is used for calculation: Unit: mV; The time constant τ is obtained by fitting an exponential curve using the nonlinear least squares method. Specific time point recovery rate The following formula is used for calculation: 100%, of which For a specific time point The voltage, in mV; The average recovery rate is calculated using the following formula: Average recovery rate = Unit: mV / s, where t1 and t2 are two arbitrary times, with t2 being later than t1. and These are the voltages at times t1 and t2, respectively, in mV.

7. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 2, characterized in that, Calculate instantaneous internal resistance based on voltage and current changes, and perform dynamic internal resistance calculation. The estimation is performed using the following calculation formula: Unit: mΩ; in, and The voltages are the first statistical time before the mutation and the second statistical time after the mutation, respectively. and The currents are the first statistical time before the mutation and the second statistical time after the mutation, respectively.

8. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 7, characterized in that, Dynamic internal resistance The estimation specifically includes the following steps: Step 311, Identify current abrupt change events: Detect the time period during which the current changes significantly, where a significant change is defined as a change magnitude greater than a preset value, and record the stable state data before and after the abrupt change; Step 312, extract key data points: extract the voltage V1 and current I1 when the current is stable before the sudden change, and extract the voltage V2 and current I2 when the current is stable after the sudden change, to ensure that both state points are in a relatively stable period; Step 313, calculate the dynamic internal resistance; Step 314, Temperature and SOC Compensation: Record the battery temperature when calculating the dynamic internal resistance; record the SOC range when calculating the dynamic internal resistance; establish temperature compensation benchmarks under different conditions.

9. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 2, characterized in that, Capacity-related characteristics include actual capacity, capacity retention rate, capacity decay rate, and charge / discharge capacity ratio; actual capacity refers to the total amount of electricity that a cell can store or release under specific conditions; capacity retention rate refers to the ratio of current capacity to rated capacity; Capacity decay rate refers to the rate at which capacity decreases with the number of cycles; the following method is used to obtain capacity-related characteristics: Data collection conditions: Complete charge-discharge cycle: The entire process from full charge to complete discharge; Constant current: ensures stable charging and discharging current; Standard temperature: Tested within a specified temperature range; Resting period: A resting period longer than the preset time before and after charging and discharging; Actual capacity The following formula is used for calculation: ； Capacity retention The following formula is used for calculation: ； Capacity decay rate The following formula is used for calculation: ； The charge / discharge ratio (Charge_Discharge_Ratio) is calculated using the following formula: Charge_Discharge_Ratio= ; in, A constant current under specific conditions; and These are the starting and ending points of the time frame for calculating the actual capacity; Rated capacity; This is the initial capacity; Current capacity; This refers to the charging capacity. This represents the discharge capacity.

10. The battery consistency analysis method for energy storage power stations based on Gaussian mixture model according to claim 2, characterized in that, Step S04 specifically includes the following steps: Step S401: Initialize the extracted battery features; Step S402: Calculate the probability that each cell belongs to each cluster; Step S403: Update the Gaussian mixture model parameters; Step S404: Determine whether the Gaussian mixture model has converged. If it has not converged, return to step S402. If it has converged, output the updated Gaussian mixture model and the clustering results of the battery features. Step S405: Analyze and diagnose the output clustering results.

Images