Remote on-line capacity checking system for storage battery

Abstract

The application relates to the field of battery capacity analysis, and specifically discloses a battery remote online capacity analysis system. First, the original time sequence data uploaded by a battery management system is cleaned, and a discharge segment containing effective information is extracted therefrom. On this basis, by analyzing the dynamic pressure drop and internal resistance characteristics in the discharge segment, a short-board single cell limiting the overall capacity due to performance degradation is accurately anchored, the exclusive voltage and current sequence thereof is obtained, and the problem that the traditional method ignores the single cell difference because of focusing on the overall average is effectively solved. Further, the short-board single cell data is processed by using an incremental capacity analysis technology, the health degree is mapped by identifying the characteristic peak of the smoothed incremental capacity sequence, and then the actual available capacity of the current battery pack is calculated in combination with the nominal capacity, so that the performance of the battery pack is accurately evaluated in the remote online environment.

Description

Technical Field

[0001] This application relates to the field of battery capacity assessment, and more specifically, to a remote online battery capacity assessment system. Background Technology

[0002] As the core backup power source for critical infrastructure such as communication base stations, substations, and data centers, the accurate assessment of the health status and remaining capacity of batteries is crucial for ensuring the continuity and reliability of power supply systems. With the popularization of Industrial Internet of Things (IIoT) technology, remote online capacity assessment using operational data uploaded by Battery Management Systems (BMS) has gradually replaced the high-cost, high-risk manual on-site capacity assessment, becoming a mainstream trend for improving operation and maintenance efficiency and reducing maintenance costs. Building an efficient remote online battery capacity assessment system aims to accurately grasp the actual performance of battery packs through real-time data stream analysis without interrupting business power supply, thereby promptly identifying potential problems and making maintenance decisions.

[0003] However, existing remote online capacity assessment technologies for batteries face numerous challenges in practical applications. First, in most industrial scenarios, batteries are in a float charge state for extended periods or only experience shallow charge and discharge conditions, making it extremely difficult to obtain a complete deep discharge curve. This renders traditional capacity testing methods based on complete charge-discharge cycles ineffective. Second, many industrial-grade batteries (such as lithium iron phosphate batteries) exhibit extremely flat voltage curves within their intermediate state of charge (SOC) range, showing significant voltage plateau hysteresis. Existing schemes based on voltage lookup tables or simple ampere-hour integration cannot accurately estimate SOC changes based on minute voltage variations, leading to substantial capacity assessment errors. More critically, the actual usable capacity of a battery pack is limited by the "weakest link" effect, meaning it depends on the worst-performing individual cell. Existing remote monitoring schemes often focus on monitoring the total or average voltage of the battery pack, failing to deeply analyze the voltage drop differences and internal resistance characteristics of individual cells under dynamic loads. This results in the deteriorating weakest cells being masked by the overall data. Without deep discharge testing, existing technologies struggle to accurately pinpoint weak cells and assess the actual usable capacity of the entire battery pack based on their characteristics, resulting in low accuracy of remote capacity assessment and an inability to truly reflect the health status of the battery pack.

[0004] Therefore, a remote online capacity assessment system for batteries is desired that can accurately pinpoint the weakest individual cells from raw time-series data that is not fully charged or fully discharged, and accurately assess the actual usable capacity of the battery pack based on incremental capacity analysis technology. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. According to this application, a remote online capacity assessment system for a storage battery includes: a raw timing data stream acquisition module, used to acquire a raw timing data stream uploaded by a battery management system, wherein each data point in the raw timing data stream includes the voltage of the i-th cell, the loop current, and the ambient temperature.

[0006] The timing data stream cleaning module is used to clean the raw timing data stream to obtain a set of effective discharge segments.

[0007] The segment unit anchoring module is used to anchor the effective discharge segment set to a reference unit based on the short-board effect to obtain the short-board unit index, short-board unit voltage sequence, and current sequence.

[0008] The incremental capacity construction module is used to construct a smoothed incremental capacity sequence based on the voltage and current sequences of the short-board unit.

[0009] The short-board unit health analysis module is used to identify characteristic peaks and map health status to the smoothed incremental capacity sequence in order to obtain the current health status of the short-board unit.

[0010] The actual usable capacity analysis module is used to determine the actual usable capacity of the current battery pack based on the health of the current weakest individual cells and the nominal capacity of the battery.

[0011] Compared with existing technologies, this application provides a remote online capacity assessment system for batteries, which addresses the technical challenges of capacity assessment difficulties and the masking of bottleneck effects under shallow charge and discharge conditions. The system first cleans the raw time-series data uploaded by the battery management system, extracting discharge segments containing valid information, and then analyzes the fragmented data from non-full charge / discharge conditions. Based on this, by analyzing the dynamic voltage drop and internal resistance characteristics within the discharge segments, it accurately identifies the bottleneck cells that limit the overall capacity of the battery pack due to performance degradation, obtaining their unique voltage and current sequences, effectively solving the problem of traditional methods neglecting individual cell differences due to focusing on the overall average. Furthermore, incremental capacity analysis technology is used to process the bottleneck cell data, identifying the characteristic peaks of the smoothed incremental capacity sequence to map health status, and then combining this with the nominal capacity to calculate the actual usable capacity of the current battery pack, achieving accurate performance evaluation of the battery pack in a remote online environment. Attached Figure Description

[0012] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 This is a schematic block diagram of a remote online capacity assessment system for batteries according to an embodiment of this application.

[0014] Figure 2 This is a schematic diagram of the data flow of a remote online capacity assessment system for batteries according to an embodiment of this application.

[0015] Figure 3 This is a schematic block diagram of the timing data stream cleaning module in a remote online capacity assessment system for batteries according to an embodiment of this application.

[0016] Figure 4 This is a schematic block diagram of a segment cell anchoring module in a remote online capacity assessment system for batteries according to an embodiment of this application.

[0017] Figure 5 This is a schematic diagram of the logic flow of the incremental capacity building module in the remote online capacity assessment system for batteries according to an embodiment of this application. Detailed Implementation

[0018] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0019] It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0020] To address the resource waste and efficiency bottlenecks caused by the rigid communication strategies in existing lithium battery systems, this application proposes a remote online capacity verification system for batteries. Figure 1 This is a schematic block diagram of a remote online capacity assessment system for batteries according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in a remote online capacity assessment system for batteries according to an embodiment of this application. Specifically, as shown... Figure 1 and Figure 2As shown, the remote online capacity assessment system 100 for batteries according to an embodiment of this application includes: a raw time-series data stream acquisition module 110, used to acquire raw time-series data streams uploaded by the battery management system, wherein each data point in the raw time-series data stream includes the voltage of the i-th cell, the loop current, and the ambient temperature; a time-series data stream cleaning module 120, used to clean the raw time-series data stream to obtain a set of effective discharge segments; a segment cell anchoring module 130, used to anchor the effective discharge segment set to a reference cell based on the short-board effect to obtain a short-board cell index, a short-board cell voltage sequence, and a current sequence; an incremental capacity construction module 140, used to construct a smoothed incremental capacity sequence based on the short-board cell voltage sequence and current sequence; a short-board cell health analysis module 150, used to identify characteristic peaks and map health to the smoothed incremental capacity sequence to obtain the health of the current short-board cell; and an actual usable capacity analysis module 160, used to determine the actual usable capacity of the current battery pack based on the health of the current short-board cell and the nominal battery capacity.

[0021] Specifically, the raw time-series data stream acquisition module 110 is used to acquire the raw time-series data stream uploaded by the battery management system. Each data point in the raw time-series data stream includes the voltage of the i-th cell, the loop current, and the ambient temperature. It is understandable that in distributed industrial scenarios such as communication base stations, substations, and data centers, battery packs typically serve as the last line of defense for ensuring continuous power supply, remaining in a float-charge standby state for extended periods. Traditional operation and maintenance methods rely on periodic manual inspections or offline deep discharge tests, which not only consume enormous human and material resources but also often require disconnecting the battery pack from the business system during testing, posing a safety hazard of power interruption. To achieve non-intrusive remote health status assessment, it is necessary to capture the dynamic response characteristics of the battery during actual operation (even under shallow charge and discharge conditions), because internal electrochemical changes in the battery (such as increased internal resistance and loss of active materials) are directly manifested through the transient response of voltage and current to load changes. Therefore, acquiring the raw time-series data stream uploaded by the battery management system is to establish a high-frequency, multi-dimensional historical operation database, digitizing the battery electrochemical behavior in the physical world into time-series signals that can be analyzed by algorithms. This provides the most basic and indispensable observation data source for subsequent extraction of effective discharge segments from fragmented data, anchoring of weak cells, and incremental capacity analysis, ensuring that the entire capacity assessment process is based entirely on real field operating conditions without the need for additional human intervention.

[0022] In detail, in an exemplary embodiment, the raw time-series data stream acquisition module 110 processes the data as follows: The raw time-series data stream acquisition module 110 establishes a remote communication link with the battery management system (BMS) deployed on-site through an industrial IoT gateway, and performs high-frequency sampling and data transmission tasks for the battery pack's operating status. This process first involves the acquisition of physical quantities from the underlying sensor network, and then converts analog signals into digital signals through a communication protocol and uploads them to a cloud server or edge computing node. First, the data acquisition frequency and triggering mechanism need to be configured. Typically, to capture sufficient dynamic characteristics for subsequent internal resistance analysis, the sampling frequency is set to 1Hz, i.e., full data is acquired once per second. During certain periods of severe load fluctuations, the BMS may use a higher sampling rate for internal caching and upload the data in a packaged form. The core data objects in the raw time-series data stream include the voltage of the i-th cell, the loop current, and the ambient temperature. These data points are all marked with a unified timestamp t to ensure strict alignment of the data in the time dimension.

[0023] The voltage of the i-th cell refers to the potential difference between the positive and negative terminals of the i-th independent cell in the battery pack, acquired at time t. In a standard 48V communication backup power system, there are 15 or 16 lithium iron phosphate cells connected in series, so the value of i ranges from 1 to 16. The BMS reads the analog voltage value using a high-precision analog-to-digital converter (ADC) through the voltage acquisition harness connected to each cell's terminal. For example, at time t... The system will read a set of vector data simultaneously. These voltage values ​​require accuracy at the millivolt (mV) level because subsequent short-board unit anchoring is highly dependent on minute voltage differences between units. If the BMS detects a unit voltage of 3.345V, this value is recorded as the raw data for that moment.

[0024] Loop current refers to the current intensity flowing through the entire main circuit of the battery pack at time t. Since the battery pack uses a series structure, according to Kirchhoff's current law, the current flowing through each cell is theoretically equal (ignoring bypass balancing current). This data is collected by a Hall sensor or precision shunt installed on the main circuit. The system needs to preset the current direction indicator, for example, specifying the discharge direction as negative and the charging direction as positive, or vice versa, as long as it remains consistent in subsequent calculations. For example, when the base station load starts and the battery pack begins to discharge, the sensor captures a current value of -50.5A, which directly reflects the current load pressure of the battery pack. Loop current is a key denominator variable in subsequent calculations of capacity (ampere-hour integral) and internal resistance (voltage change divided by current change).

[0025] Ambient temperature refers to the thermodynamic temperature value of the environment inside or around the battery pack, collected at time t. Because the electrochemical reaction rate and internal resistance characteristics of the battery are highly sensitive to temperature, this data is used for temperature compensation correction in subsequent steps. The BMS acquires temperature information through NTC thermistors attached to the surface of the battery cells or distributed within the battery pack. In practice, values ​​from multiple temperature points may be collected (e.g., ...). , , The raw time-series data stream acquisition module can be configured to record the average or highest value of these points as the representative temperature at that moment. For example, a temperature reading of 25.8 degrees Celsius is acquired, a parameter that is crucial for correcting for capacity assessment biases caused by seasonal variations.

[0026] In the data transmission and reception phase, the raw time-series data stream acquisition module acts as a server, subscribing to data topics published by the BMS via standard industrial protocols such as MQTT, ModbusTCP, or IEC 61850. After the BMS completes one sampling cycle, it encapsulates the aforementioned voltage, current, and temperature data, along with a UTC timestamp, into a JSON or binary data packet and uploads it. For example, a data packet might contain: "Timestamp: 1678886400, Loop Current: -45.2A, Ambient Temperature: 26.1°C, Individual Voltage Set: [3.210, 3.208, ..., 3.195]". Upon receiving this data packet, the acquisition module performs a preliminary integrity check, inspecting for packet loss or format errors. After successful verification, the data is parsed and appended sequentially to a time-series database (such as InfluxDB or TimescaleDB), forming a continuous raw time-series data stream.

[0027] To illustrate more specifically, consider a 100Ah lithium iron phosphate battery pack at a remote base station. The system's preset sampling period is 1 second. At 10:00:01, the base station's mains power is briefly interrupted, and the battery pack instantly assumes the load. At this moment, the BMS detects a sudden change in the loop current from 0A to -20A. Simultaneously, the voltage of the 16 individual cells rapidly drops from approximately 3.45V in float charging mode to a plateau of approximately 3.30V. Due to inconsistent cell health, the voltage drop of the 5th cell is slightly greater than the others, falling to 3.28V, while the ambient temperature remains at 22℃. The raw timing data stream acquisition module receives and records all state values ​​for this one second in real time. Immediately afterwards, at 10:00:02, the current remains at -20A, and the voltage continues to show slight changes. This process continues, and the data stream accumulates continuously regardless of whether the battery is in a static, float charging, or discharging state. Ultimately, the module outputs a multi-channel data matrix containing a time dimension. Each row of the matrix corresponds to a time sampling point, and each column corresponds to a physical quantity (time, current, temperature, and cell voltage 1 to cell voltage N). This unprocessed raw data stream faithfully records the entire battery operation process, including noise, anomalies, and key discharge characteristics.

[0028] Specifically, the time-series data stream cleaning module 120 is used to clean the raw time-series data stream to obtain a set of effective discharge segments. It should be understood that in the complex electromagnetic environment of industrial sites, the raw time-series data stream collected and uploaded by the battery management system (BMS) is inevitably affected by high-frequency noise interference, signal drift, and wireless transmission packet loss, resulting in inconsistent data quality. Furthermore, batteries are in a float-charge standby state for most of the time, and the discharge behavior that truly contains electrochemical characteristics exhibits sparsity and fragmentation. If the raw data, containing a large amount of redundant steady-state data and signal spikes, is directly used for subsequent incremental capacity analysis or internal resistance calculation, it is very easy to introduce huge differential errors, leading to the failure of anchoring weaker cells. Therefore, cleaning the raw time-series data stream aims to eliminate the masking effect of environmental noise on minute voltage changes, accurately extract the discharge processes with analytical value from massive historical data, and repair timeline breaks caused by communication instability, thereby providing a clean, continuous, and time-ordered data foundation for building a high-precision battery health assessment model.

[0029] Figure 3 This is a schematic block diagram of the timing data stream cleaning module in a remote online capacity assessment system for batteries according to an embodiment of this application. Figure 3As shown, in an exemplary embodiment, the time-series data stream cleaning module 120 includes: a data filtering unit 121, used to perform sliding window filtering on the original time-series data stream to obtain a filtered data stream; a discharge segment identification and segmentation unit 122, used to identify and segment effective discharge segments in the filtered data stream based on a discharge current threshold and a minimum duration threshold to obtain a candidate discharge segment set; and a data resampling and alignment unit 123, used to perform data resampling and time-series alignment on the candidate discharge segment set to obtain an effective discharge segment set.

[0030] In detail, in the above implementation scheme, the processing procedure of the time-series data stream cleaning module 120 is as follows: First, the acquired raw time-series data stream is processed by the data filtering unit 121. The core task of this process is to eliminate random noise and pulse interference superimposed on the real physical signal, while preserving the edge features of the signal as much as possible to prevent distortion. The data filtering unit adopts a sliding window filtering technique. Specifically, for each numerical sequence in the raw time-series data stream, including the i-th unit voltage sequence, denoted as... The loop current sequence is denoted as... And the ambient temperature sequence, denoted as Sliding median filters are applied separately. The window size W is a key preset hyperparameter, and its setting depends on the data acquisition frequency and the spectral characteristics of the noise. In a typical application scenario with a sampling frequency of 1Hz, in order to filter out occasional spike interference without causing excessive phase delay, the window size W is set to 5 or 7 data points. In this application, the window size W is set to 5. For any given time point k, the data filtering unit will extract a time window centered on that point, extending two data points before and after it, i.e., a data subset with an index range of [k-2, k+2]. For example, in the base station battery pack scenario mentioned in the previous step, for the voltage of the 5th cell, at time point... The original voltage sequence segments acquired at and before time points are [3.280, 3.279, 3.350, 3.278, 3.277]. The median value of 3.350V deviates significantly from the normal voltage decrease trend, most likely due to an outlier caused by a momentary electromagnetic pulse interference to the sensor. If mean filtering is used, this outlier would raise the overall average level, distorting the filtering result. However, the median filtering algorithm used by the data filtering unit sorts the values ​​within this window, resulting in [3.277, 3.278, 3.279, 3.280, 3.350], and selects the value at the middle position, 3.279V, as the filtered value for time point k. This effectively eliminates the original impulse noise, restoring a smooth voltage curve. The unit performs this operation point-by-point on the voltage, total loop current, and temperature data of all 16 individual cells. For data points where the beginning and end of the sequence cannot form a complete window, edge copying or original value retention strategies are used. Finally, the processed sequences of physical quantities are re-aligned and combined with the original timestamps, outputting a filtered data stream. This data stream is structurally identical to the original data, but the signal-to-noise ratio is significantly improved.

[0031] After signal denoising, the discharge segment identification and segmentation unit 122 extracts the discharge interval containing effective electrochemical information from the continuous filtered data stream using physical rules. Since the voltage and current data of the battery in float charging or resting states have limited significance for capacity analysis, specific threshold logic is needed to lock the discharge behavior. This unit mainly relies on two core parameters: the discharge current threshold, denoted as... And the minimum duration threshold, denoted as Discharge current threshold This is a preset negative current value used to define whether the battery is under substantial discharge conditions. This threshold is related to the battery's rated capacity C. Considering the zero-point drift and slight leakage current of the sensor, the threshold cannot be set to 0; it is generally set to -0.05C. Taking the 100Ah battery pack in this example... Set to -5A. This means that the system only considers the battery to be responding to load demands when the discharge intensity of the loop current exceeds 5A (i.e., the current value is less than -5A). Minimum duration threshold. This is used to filter out invalid segments generated by transient disturbances or extremely short load tests (such as device self-tests lasting a few seconds), because excessively short discharge processes are insufficient to excite the polarization reaction inside the battery and cannot be used to calculate internal resistance or construct incremental capacity curves. Based on experience, this parameter is set to 60 seconds or longer; in this embodiment, it is set to 300 seconds to ensure that the extracted segments have sufficient analytical depth. The specific identification and segmentation process follows state machine logic. This unit traverses the current sequence in the filtered data stream in chronological order. For each time point k, the system records the current value as... With discharge current threshold Compare. If If the current is less than -5A, then that time point is marked as a discharge state; otherwise, if If the current is ≥-5A, it is marked as a non-discharge state. Subsequently, the cell performs continuous interval detection, scans the generated marker sequence, and finds the moment when the state changes from non-discharge to discharge as the start timestamp. The end timestamp is the moment when the state returns from discharge to non-discharge. For example, the system detects that at 10:00:01, the current jumps to -20A (less than -5A) and remains around -20A until 10:15:00 when the current returns to 0A. At this point, the system initially identifies a time interval [10:00:01, 10:15:00]. Next, the unit performs a duration check, calculating the duration of this continuous interval. In the example above, The duration is 899 seconds. Since 899s > 300s, which is the minimum duration threshold... This interval passed the verification and was determined to be a valid discharge segment. The unit then extracted all data within this time period from the filtered data stream, including timestamps, 16 individual cell voltages, loop current, and ambient temperature, forming an independent data object, namely the candidate segment. The current fluctuation is then stored in the candidate discharge segment set. If a current fluctuation lasts only 10 seconds, it will be automatically discarded by the system because it does not meet the duration requirement.

[0032] Although discharge segments are extracted, the original data is often uneven along the timeline due to limitations in the transmission quality of the on-site wireless network, and may even contain gaps caused by packet loss. For example, in the aforementioned discharge segment, although most data is sampled at 1-second intervals, data may be missing at 10:05:01 due to communication congestion, and the next data jumps directly to 10:05:03. This non-uniform sampling will seriously affect the accuracy of subsequent voltage differentiation calculations. To address this, the data resampling and alignment unit 123 performs time-series reconstruction on each segment in the candidate discharge segment set. This unit first performs a time base generation step. For each candidate segment... Based on its determined start time and end time and the preset standard sampling step size Set to 1 second, generate an absolutely uniform and continuous standard timestamp sequence. This sequence forms the skeleton of the data reconstruction. Subsequently, the cells are filled with data using a linear interpolation algorithm. The fragments are then traversed. Each of the original physical quantity channels, such as the single-unit voltage in Section 5, is for a standard timestamp sequence. Each target time point Find the previous valid time point in the original dataset that is closest to its time. and the next valid time point and their corresponding values and For example, to fill in 10:05:01, that is... The system found the voltage at 10:05:00, which was missing data. =3.250V, that is and the voltage at 10:05:03 =3.247V, that is Calculate using the following formula Interpolation at time :

[0033]

[0034] Substituting the values, we get: =3.250+(3.247-3.250)×{(10:05:01-10:05:00) / (10:05:03-10:05:00)}=3.249V. In the formula, The target value after resampling. This represents the change in value within the interval. This represents the time weight of the target time point within the interval. In this way, the system not only fills in the missing data points but also maps all non-uniformly distributed sampling points onto a standard 1Hz time grid. Finally, the data resampling and alignment unit reorganizes all interpolated and time-aligned voltage, current, and temperature data according to the standard timestamp sequence, forming well-structured discharge segments without missing values, and stores them in the set of valid discharge segments. Each segment in this set is a complete time series matrix, eliminating noise interference and timing jitter.

[0035] Specifically, the segment cell anchoring module 130 is used to anchor the effective discharge segment set to a benchmark cell based on the bottleneck effect to obtain the bottleneck cell index, bottleneck cell voltage sequence, and current sequence. Correspondingly, battery packs follow the barrel effect principle, and their actual usable capacity is often limited by the worst-performing cell. In traditional remote monitoring schemes, only the total voltage or average voltage of the battery pack is typically considered. This crude monitoring method easily masks the performance differences between cells, making it difficult to detect deteriorated cells in their early stages. Especially under shallow cycle conditions (not full charge / discharge), all cell voltages are in a plateau region with minimal static voltage differences. Only during dynamic discharge, by analyzing the voltage drop characteristics and internal resistance changes in response to load, can the subtle signs of a bottleneck be detected. Therefore, this application anchors the effective discharge segment set to a benchmark cell based on the bottleneck effect in order to accurately identify the bottleneck cell that determines the fate of the entire battery pack from massive time-series data and extract its unique voltage and current behavior sequence. This allows subsequent capacity assessment to focus on the most critical bottleneck and ensures that the capacity assessment results can truly reflect the minimum usable capacity boundary of the battery pack.

[0036] Figure 4This is a schematic block diagram of a segment cell anchoring module in a remote online capacity assessment system for batteries according to an embodiment of this application. Figure 4 As shown, in an exemplary embodiment, the segment unit anchoring module 130 includes: a unit dynamic feature parameter extraction unit 131, used to extract the unit dynamic feature parameters of each effective discharge segment in the effective discharge segment set to obtain a feature index set, the feature index set including dynamic voltage drop rate, dynamic internal resistance and terminal voltage; an index set normalization unit 132, used to perform multi-segment feature aggregation and normalization on the feature index set to obtain a normalized comprehensive scoring matrix; a discharge segment weighted scoring unit 133, used to perform short-board unit decision based on weighted scoring on the effective discharge segment set based on the normalized comprehensive scoring matrix to obtain a short-board unit index; and a voltage and current sequence extraction unit 134, used to extract the short-board unit voltage sequence from the effective discharge segment set and the current sequence from the original time-series data stream based on the short-board unit index.

[0037] In detail, in the above implementation scheme, the processing procedure of the fragment single-unit anchoring module 130 is as follows: First, the single-unit dynamic feature parameter extraction unit 131 performs in-depth mining on the input set of valid discharge fragments. The core task of this unit is to transform each discharge fragment from a simple time series into a set of quantified performance indicators. In specific implementation, the unit first initializes an empty set of feature indicators. Used to store the calculation results. Subsequently, for each valid discharge segment in the set... For example, the system iterates through the 899-second discharge segment mentioned above. For this segment, the system first locks its start time. (10:00:01) and end time (10:15:00), and calculate the average discharge current in this interval. For example, the average discharge current in this segment is -20A. Next, the system enters a single-cell traversal cycle, analyzing the dynamic performance of each of the 16 cells. First, the dynamic internal resistance is calculated, a key indicator of battery aging. The system extracts the voltage of the i-th cell at the initial moment. For example, the voltage of a single cell in Section 5 is 3.30V and the voltage at the end. For example, the fifth monomer is 3.22V. Using the formula... The calculation is performed, where the numerator represents the total voltage drop during discharge, and the denominator is the absolute value of the average current. Substituting the values, we can obtain... =(3.30-3.22) / 20=0.004Ω, or 4mΩ. This indicator directly reflects the voltage holding capacity of a single cell under load. Secondly, to capture the voltage drop trend at the end of the discharge phase, the system calculates the dynamic voltage drop rate. Therefore, the system extracts a fixed time window at the end of the segment. The window length is set according to the electrochemical response time constant of the battery, and in this example, it is set to the last 30 seconds. The system extracts the monomer from section 5. The voltage sequence within the interval is obtained, and a linear regression algorithm is applied to fit the slope of the voltage change over time. The calculation formula is: If the voltage of the fifth cell drops rapidly in the last 30 seconds with a large absolute slope (e.g., -0.0005V / s), it indicates a significant polarization effect and poor performance. Finally, the system directly records the terminal voltage. For example, 3.22V serves as a direct basis for determining the cutoff capacitance. After the calculation is completed, the cell will obtain the triplet. The individual cell index i and the fragment index j are associated and stored in the feature index set. This process is repeated for all individual cells and all fragments, and the final feature index set records in detail the dynamic health check report of each individual cell in each discharge event.

[0038] After acquiring a feature index set containing characteristic parameters of multiple discharge segments, the index set normalization unit 132 first performs a feature aggregation operation. The purpose of this operation is to condense the multiple performances of a single cell under different times and operating conditions into a stable comprehensive index. For example, if the effective discharge segment set contains m=5 effective discharge segments, for the fifth cell, the system will extract the dynamic internal resistance calculated during these 5 discharge processes. Dynamic pressure drop rate and terminal voltage Subsequently, the system uses the arithmetic mean method to calculate its comprehensive characteristic value: ; ; As calculated, the average dynamic internal resistance of the monomer in Section 5 is... The voltage drop rate is 4.2 mΩ. The average terminal voltage is 0.55 mV / s. The voltage is 3.21V. The system performs the above aggregation calculation on all 16 cells in the battery pack to obtain the average feature vector of each cell. After feature aggregation, the unit then performs feature normalization processing, mapping all indicators to the dimensionless interval [0,1] to facilitate equal-weighted or weighted evaluation. To unify the scoring logic, this embodiment stipulates that the higher the normalized score, the worse the cell performance, i.e., the closer it is to the bottleneck. For internal resistance... and pressure drop rate These two metrics are inherently more accurate, with higher values ​​indicating worse performance. Therefore, a positive maximum-minimum normalization formula is used: For example, among the set of average internal resistances of all 16 individual cells, the maximum value is... It is 5.0 mΩ, from the 8th monomer, the minimum value. The resistance is 3.0 mΩ, from the first monomer. For the fifth monomer (4.2 mΩ), its normalized internal resistance score is: For the terminal voltage A lower value for this indicator indicates faster discharge and smaller capacity (i.e., worse performance), which is the opposite of the aforementioned indicators. To unify the direction, an inverse normalization formula is used:

[0039]

[0040] The maximum value of the average terminal voltage of all cells It is 3.28V, the minimum value. The voltage is 3.18V. For the fifth cell (3.21V), its normalized voltage score is: The higher the score (closer to 1), the closer its voltage is to the minimum value, and the worse its performance. Ultimately, this unit uses the three normalized scores of all 16 individual cells. A normalized comprehensive score matrix is ​​constructed by combining these components. Each row of this matrix corresponds to a single unit, and each column corresponds to a normalized feature dimension. This matrix eliminates differences in dimensions and inconsistencies in evaluation direction, and intuitively displays the relative degree of degradation of each unit across different performance dimensions.

[0041] The discharge segment weighted scoring unit 133 processes the normalized comprehensive scoring matrix output from the previous stage. The core logic of this unit is to assign corresponding weight coefficients based on the different degrees of influence of each characteristic index on the battery's SOH (State of Health), and calculate the comprehensive degradation score for each individual cell. In an exemplary embodiment, the discharge segment weighted scoring unit 133 is used to: perform a weighted scoring-based decision on the weakest cell in the effective discharge segment set using the following formula:

[0042]

[0043]

[0044] in, The overall health score for individual i. For dynamic internal resistance, For dynamic voltage drop rate, This is the terminal voltage. , and These are the dynamic internal resistance weighting coefficient, the dynamic voltage drop rate weighting coefficient, and the terminal voltage weighting coefficient, respectively. For normalization processing, To find Index to get the maximum value , This is for indexing the weakest link. In practice, the weight vector must first be determined. These weighting coefficients are derived from correlation analysis of expert experience bases or large-scale historical data, and satisfy normalization conditions. In the communication base station backup power supply scenario of this embodiment, considering that the change in internal resistance under shallow charge and discharge conditions is most sensitive to aging, the following is assigned: Higher weights, such as 0.4, result in a higher voltage drop rate. Next, such as 0.4, the terminal voltage Because it is significantly affected by fluctuations in the depth of discharge, its weight is slightly lower, such as 0.2. After determining the weight, this unit traverses each row of the matrix (i.e., each individual cell) and calculates its comprehensive health score. Continuing with the data example above, for the monomer in Section 5, its normalized indices are: internal resistance score. =0.60, pressure drop rate score =0.55, terminal voltage score =0.70. Substitute the weights to calculate its overall score: =0.4×0.60+0.4×0.55+0.2×0.70=0.60. Meanwhile, the system also calculates the scores of other individual cells. For example, although cell 8 has the highest internal resistance (normalized to 1.0), its voltage drop rate and terminal voltage performance are acceptable (normalized to 0.4), and its score is 0.4×1.0+0.4×0.4+0.2×0.4=0.64. Cell 12 exhibits more significant degradation characteristics, with an average dynamic internal resistance reaching 4.8mΩ (normalized score as high as 0.90), an average terminal voltage as low as 3.19V (normalized score as high as 0.80), and a voltage drop rate normalized score of 0.90. Substituting these parameters into the weighting formula, the overall health score of cell 12 is calculated to be 0.4×0.90+0.4×0.90+0.2×0.80=0.88. After the calculation is completed, the unit performs a short-board individual locking operation, using the formula... Find the index corresponding to the highest score. In the above comparison, if the score of the 12th cell is 0.88, it means that its overall performance is the worst and it is the bottleneck restricting the battery pack capacity. Therefore, the system will... The index number is locked at 12. This index number 12 is the final determined index of the weakest link cell, which clearly indicates which cell needs to be given special attention.

[0045] In particular, during the aging process of a battery's entire lifecycle, performance degradation at different stages is often dominated by different electrochemical mechanisms. For example, in the early stages of battery use, increased contact resistance may be the primary issue, while in the later stages, the loss of active lithium or the collapse of electrode material structures may become key factors limiting capacity. This means that the sensitivity and importance of characteristic indicators that accurately indicate the battery's state of health (SOH), such as dynamic internal resistance, voltage drop rate, and terminal voltage, change dynamically at different times. If fixed weighting coefficients are used to determine the weakest cells, it is easy to make misjudgments by "marking the boat where it came from"—that is, indicators given high weights are not actually sensitive in the current state, while indicators that truly reflect the current degradation mechanism are underestimated. To solve this problem, an adaptive mechanism that can sense the specific state of the current battery pack needs to be established, tightly coupling the decision-making process of weakest cells with incremental capacity (IC) analysis that reflects the electrochemical essence. The diagnostic capability of the IC curve for SOH should be used as a "ruler" to measure the importance of each basic screening indicator at the current moment. Therefore, the adaptive weighting coefficient determination based on IC degradation feedback is implemented in order to dynamically tilt the decision weights toward those indicators that best match the current aging mode by quantitatively analyzing the statistical correlation between each basic indicator and the actual electrochemical degradation of the battery, thereby ensuring that the identified weak cells truly represent the performance bottleneck of the battery pack.

[0046] In an exemplary preferred embodiment, the determination of the weighting coefficients includes:

[0047] The system extracts voltage and current sequences of each cell from the effective discharge segment set within a representative discharge segment, and extracts characteristic peak parameters based on these sequences, including the IC peak position and peak height. This step aims to perform a rapid virtual IC analysis on each cell within the battery pack to obtain benchmark data for measuring its electrochemical health. Specifically, the system selects the representative segment with the best data quality and deepest discharge from the effective discharge segment set, and extracts the voltage sequences of all cells within it, such as those of 16 cells. and current sequence Subsequently, using voltage domain resampling and Gaussian filtering algorithms, the smoothed IC curve for each individual cell is calculated. The specific calculations will be explained in detail in the subsequent incremental capacity construction module. Although this process is computationally intensive, it is necessary for accurate weight allocation. For each generated IC curve, the system identifies its main characteristic peak and extracts two key parameters: IC peak position. The voltage and peak height corresponding to the peak value The peak value corresponds to the capacity change rate. Taking the 12th monomer as an example, calculations show its characteristic peak is located at 3.280V with a height of 45.5 Ah / V; while the relatively healthy 1st monomer has a peak at 3.295V with a height of 48 Ah / V. These parameters intuitively reflect the integrity and polarization degree of the phase transition process within each monomer.

[0048] An IC deterioration score is calculated for each monomer based on its characteristic peak parameters. This step aims to transform the complex differences in curve morphology into a single quantifiable numerical indicator, i.e., deterioration, to facilitate subsequent statistical analysis. The system defines an IC curve deterioration scoring function. This function quantifies the degree to which the current monomeric IC curve deviates from an ideal (healthy) state. The calculation formula is as follows:

[0049]

[0050] In the formula, Score the degree of IC deterioration. and These are reference values ​​for the battery at the factory or in its healthy condition, for example... =50Ah / V, =3.300V. and These are preset meta-weights used to balance the effects of peak height attenuation and peak position shift, for example, each is set to 0.5. This term reflects the relative decay rate of the peak height and characterizes the capacity loss; This term reflects the relative shift of the peak position, characterizing the increase in internal resistance. Substituting the data from the monomer in Section 12: =0.5×(1-45.5 / 50)+0.5×|3.280-3.300| / 3.300≈0.047, where a higher score indicates a worse IC curve shape. The system calculates for each of the 16 monomers individually, resulting in a degradation vector. .

[0051] Based on the correlation between the dynamic internal resistance, dynamic voltage drop rate, and terminal voltage of each individual cell and the IC degradation score, the correlation coefficients for internal resistance-IC degradation, voltage drop rate-IC degradation, and terminal voltage-IC degradation are calculated. This step is crucial for establishing the logical connection between basic indicators and the final diagnostic results. Since the relationship between each indicator and degradation may not be strictly linear, but is usually monotonic (e.g., higher internal resistance corresponds to higher degradation), the Spearman rank correlation coefficient is preferred for calculation. This method is based on the rank (sorting) of the data rather than the original values, and has stronger robustness. The correlation coefficient between internal resistance and IC degradation is then calculated. For example, the formula is: Where n is the total number of individuals, i.e., 16. This is the squared difference between the ranking of the internal resistance value and the ranking of the IC degradation of the i-th monomer. If, in the current batch of data, the internal resistance distribution and IC degradation distribution of the monomers highly match (i.e., monomers with high internal resistance also have high IC degradation), the calculated... It could be as high as 0.85. Similarly, the system calculates the correlation coefficient between the voltage drop rate and the degree of degradation. For example, a value of 0.45, and the correlation coefficient between terminal voltage and degradation degree. For example, a value of 0.20. These coefficients directly reveal which indicator is the main cause of the deterioration in the IC curve at this moment.

[0052] Adaptive weighting is performed on the correlation coefficients of internal resistance-IC degradation, voltage drop rate-IC degradation, and terminal voltage-IC degradation to obtain dynamic internal resistance weighting coefficients, dynamic voltage drop rate weighting coefficients, and terminal voltage weighting coefficients. This step transforms the statistical correlations into weight parameters in the decision model, ensuring that the scoring system automatically focuses on the most critical features. The calculation formulas are as follows: ; ; In the formula, the numerator is the absolute value of the correlation coefficient, and the denominator is the sum of the absolute values ​​of all correlation coefficients to ensure weight normalization. Substituting the values ​​from the example above: the sum of the denominators = 0.85 + 0.45 + 0.20 = 1.50. Dynamic internal resistance weighting coefficient. =0.85 / 1.50≈0.57; Dynamic pressure drop rate weighting coefficient =0.45 / 1.50≈0.30; End-point voltage weighting coefficient =0.20 / 1.50≈0.13. Compared to traditional fixed weights (such as 0.4, 0.4, 0.2), the value calculated in this embodiment is... The score was significantly improved to 0.57 because the system, through data analysis, discovered that the current battery pack's degradation was mainly manifested in increased internal resistance. Therefore, it automatically assigned a higher weight to internal resistance, thereby ensuring that the weakest cells selected based on this weight were identified. It is the most accurate representation of the current physical aging state and is a true shortcoming.

[0053] After identifying the weakest link cell, the voltage-current sequence extraction unit 134 is responsible for extracting a streamlined dataset from the massive historical data for subsequent algorithms. Although the system identified the 12th segment as the weakest link cell, this segment may contain hundreds or thousands of discharge fragments in its historical records. To improve the signal-to-noise ratio and characteristic peak recognition rate of incremental capacity analysis (ICA), this unit needs to select the most representative fragment. A common strategy is to select the fragment with the largest depth of discharge (DOD) or the longest duration from the effective set of discharge fragments. For example, among all valid fragments, the one numbered is... The longest duration of this segment was 45 minutes, and the discharge current remained stable. The unit was first indexed based on the short-board unit. =12, from Extract the data from the 12th column of the data matrix, namely the voltage sequence of the short-board unit. This is a voltage vector that varies with time t, for example [3.310, 3.309, ..., 3.150], which accurately depicts the voltage drop trajectory of the worst-performing cell during this optimal discharge process. Simultaneously, the cell is derived from the corresponding raw time-series data stream (after cleaning). Extract the corresponding loop current sequence from ) These two sequences— and Strict alignment in time constitutes the core input for subsequent incremental capacity building modules.

[0054] Specifically, the incremental capacity construction module 140 is used to construct a smoothed incremental capacity sequence based on the voltage and current sequences of the short-platen cells. It is understandable that in the field of battery electrochemical analysis, traditional voltage-capacity (VQ) curves often lack distinctiveness due to the flatness of the voltage plateau, especially in lithium iron phosphate batteries, where long voltage plateaus mask subtle features of internal phase transitions. To reveal these masked electrochemical processes, incremental capacity analysis (ICA) technology has emerged. It transforms the originally flat voltage plateau into significant characteristic peaks by calculating the differential of capacity relative to voltage (dQ / dV). The position, height, and shape of each characteristic peak correspond to a specific electrochemical reaction stage or aging mechanism (such as active material loss or lithium inventory loss) within the battery. However, directly performing numerical differentiation on discrete and noisy raw data drastically amplifies the noise, generating a large number of meaningless spurious peaks, making it difficult to extract true health characteristics. Therefore, constructing a smoothed incremental capacity sequence based on the voltage and current sequences of the short-board cells is to transform the macroscopic time-domain data obtained from the field into a microscopic voltage-domain electrochemical fingerprint, and to suppress differential noise through advanced filtering algorithms, thereby obtaining a clear and interpretable IC curve, providing a high signal-to-noise ratio feature map for subsequent accurate assessment of battery health.

[0055] Figure 5 This is a schematic diagram illustrating the logical flow of the incremental capacity building module in a remote online capacity assessment system for batteries according to an embodiment of this application. Figure 5As shown, in an exemplary embodiment, the incremental capacity construction module 140 includes: a voltage-capacity data point set construction unit 141, used to construct a voltage-capacity data point set based on the short-board unit voltage sequence and current sequence; a voltage domain resampling unit 142, used to perform voltage domain resampling and original IC curve calculation on the voltage-capacity data point set to obtain the original incremental capacity sequence; and a curve Gaussian filtering and smoothing unit 143, used to perform IC curve Gaussian filtering and smoothing on the original incremental capacity sequence based on the Gaussian standard deviation to obtain the smoothed incremental capacity sequence.

[0056] In detail, in the above implementation scheme, the incremental capacity construction module 140 processes the following: First, the voltage-capacity data point set construction unit 141 processes the input short-board unit voltage sequence. and loop current sequence The task of this unit is to map the original time-axis-based data onto a voltage-capacity two-dimensional plane. In practice, the unit first initializes an empty voltage-capacity data point set container. And the cumulative discharge capacity counter Set to zero. Then, the unit iterates along the time axis through each data point k of the input sequence, from 0 to N-1. At each time step k, the system calculates the tiny time interval using the ampere-hour integration method. This application specifies a time of 1 second, consistent with the aforementioned data cleaning steps, for the amount of electricity discharged from the battery. The calculation formula is: , such as at time Current If it is -20A, then =|-20|×(1 / 3600)≈0.00556Ah, where the time unit is converted to hours to match the Ah unit. Next, the system updates the cumulative capacity:

[0057]

[0058] If a cumulative total of 1.200 Ah was released in the previous moment, then at this moment... =1.200 + 0.00556 = 1.20556 Ah. Simultaneously, the system extracts the voltage of the short-board unit corresponding to this moment. For example, 3.295V. At this point, a data pair consisting of voltage and accumulated capacity (3.295V, 1.20556Ah) is generated and stored in the set. As the traversal proceeds, for example, if the entire discharge process lasts 45 minutes, The data set contains 2,700 such data pairs, which are discretely distributed in a coordinate system with voltage as the horizontal axis and capacity as the vertical axis.

[0059] However, because the original sampling is time-based and the battery voltage drop is non-linear (slow drop in the plateau region, dense sampling points; fast drop at both ends, sparse sampling points), it leads to... The data points are extremely unevenly distributed along the voltage axis. To calculate the differential... The data must be converted to a uniform voltage grid. This process is performed by the voltage domain resampling unit. This unit first determines the voltage range and scans... Find the maximum voltage Such as 3.35V and minimum voltage For example, 3.15V. Then, construct a fixed step size. For example, a 1mV, or 0.001V, equally spaced decreasing voltage grid sequence. =[3.350,3.349,3.348,...,3.150]. This grid sequence becomes the new x-axis. Next, the elements are interpolated using linear interpolation. The discrete capacity values ​​in the mapping are mapped to At each grid point. For any voltage point in the grid. For example, 3.300V, if in the original set If no perfectly equal voltage value can be found, the system will find the two closest original points. and ,like =3.3005V, =5.0Ah; =3.2995V, =5.1Ah, calculated through interpolation corresponding capacity This step yields a continuous function defined in the uniform voltage domain. Subsequently, in order to obtain the incremental capacity curve (IC curve), it is necessary to... Differentiation. To overcome the sensitivity of numerical differentiation to noise, this element employs the central difference method. For grid points... Its original IC value The calculation formula is as follows:

[0060]

[0061] Special attention needs to be paid to the symbols here. In a physical sense, This represents the capacity released for each unit drop in voltage. Since the voltage V decreases as the capacity Q increases during discharge, dV itself is negative. To make the IC value positive (for easier observation of the peak value), it is defined as the capacity increment divided by the absolute value of the voltage drop. The denominator in the formula... This is precisely what is reflected, because It is a positive step size, while the voltage sequence is decreasing. For example, for =3.300V, if Q(3.301V)=5.00Ah, Q(3.299V)=5.02Ah, step size =0.001V. Therefore: The calculated series The values ​​constitute the original incremental capacity sequence. Although the curve at this point reflects the rate of change of capacity with voltage, the curve is filled with sharp spikes and spurious peaks because the tiny voltage fluctuations in the original data (such as sensor noise) are amplified under the differential operation. This will seriously interfere with the identification of the true electrochemical characteristic peaks.

[0062] To obtain a smooth and realistic IC curve, a Gaussian filtering smoothing unit 143 is used. Gaussian filtering is a linear smoothing filter; its essence is to perform a weighted average of the signal, with the weight distribution following a Gaussian function (normal distribution). This unit first uses a preset Gaussian standard deviation... Generate a one-dimensional Gaussian kernel function : , This determines the degree of smoothness: The larger the value, the smoother the curve, but this may lead to peak broadening and position shift; The smaller the value, the more details are preserved, but the denoising effect deteriorates. In the scenario described in this application, to balance denoising and feature preservation, Set to 10-20 times the voltage grid step size, such as =10. The window size W of the Gaussian kernel is set to 3. Or 6 This covers the vast majority of the weights. Then, the unit performs a convolution operation. For each point in the original sequence... its smoothed value It is each point in its neighborhood. Convolution with Gaussian kernel:

[0063]

[0064] This process is equivalent to sliding a bell-shaped Gaussian window across the original curve, replacing the value of the center point with a weighted average of all points within the window. For example, for the calculated value above... =10Ah / V. If the points in its neighborhood are also close to 10, the smoothed value will not change much. However, if a point in its neighborhood suddenly becomes 50 due to noise, Gaussian weighting will assign very small weights to distant points and larger weights to nearby points, thus pulling this isolated noise point back to the average level of the surrounding points. After Gaussian filtering, the output smoothed incremental capacity sequence is... It exhibits a smooth and continuous shape. The originally messy spikes are filtered out, while the broad characteristic peaks representing the internal phase transition of the battery (such as the typical FePO4 / LiFePO4 phase transition in lithium iron phosphate batteries at around 3.30V) are clearly preserved. This smooth curve not only has an extremely high signal-to-noise ratio, but its geometric features such as peak position, height, and full width at half maximum can stably and accurately map the current SOH state of the battery.

[0065] Specifically, the short-board cell health analysis module 150 is used to identify characteristic peaks and map health status to the smoothed incremental capacity sequence to obtain the current health status of the short-board cell. Correspondingly, after obtaining a high-quality smoothed incremental capacity (IC) curve, each peak, valley, and inflection point on the curve contains rich information about the battery's internal aging. Electrochemical studies have shown that battery capacity decay does not occur uniformly, but is closely related to specific phase transition processes (such as lithium-ion intercalation / deintercalation in the graphite anode and structural degradation of the cathode material). These microscopic degradation mechanisms are directly mapped onto the IC curve, manifested as a decrease in the height of characteristic peaks (corresponding to loss of active material) or a shift in peak position (corresponding to increased internal resistance or lithium inventory loss). Traditional capacity estimation methods only focus on total discharge, cannot distinguish specific degradation mechanisms, and are easily affected by operating conditions. In contrast, the analysis method based on the characteristic peaks of the IC curve can delve into the mechanistic level, establishing a strong correlation model with SOH (state of health) by capturing changes in key characteristic parameters. However, these characteristic parameters are extremely sensitive to temperature. Without decoupling correction, capacity deviations caused by environmental temperature fluctuations may be misjudged as irreversible aging. Therefore, identifying characteristic peaks and mapping health status to the smoothed incremental capacity sequence, and introducing temperature compensation correction, aims to transform the qualitative curve shape into quantitative health status values. This constructs a high-precision assessment closed loop that is robust to environmental factors and sensitive to aging mechanisms, thereby accurately determining the true remaining lifespan of the weakest link cells.

[0066] In one exemplary embodiment, the short-board cell health analysis module 150 includes: a health indicator feature extraction unit 151, used to extract IC curve health indicator features from the smoothed incremental capacity sequence based on a peak significance threshold to obtain a feature peak parameter set; an initial health calculation unit 152, used to perform initial SOH estimation on the feature peak parameter set based on a feature mapping model based on a standard aging model library to obtain an initial health; and a temperature compensation correction unit 153, used to perform temperature compensation correction on the initial health based on the average discharge temperature to obtain the current health of the short-board cell.

[0067] In detail, in the above implementation scheme, the process of the short-board cell health analysis module 150 is as follows: First, the input smooth incremental capacity sequence is deeply analyzed by the health indicator feature extraction unit 151. The core task of this unit is to automatically identify and extract the key geometric features that best characterize the battery health status from the continuous curve data. To ensure the accuracy of the extraction, the unit presets a peak significance threshold. This is an empirical value used to screen out the main peak that truly represents the major electrochemical reaction from numerous local maxima, filtering out minor fluctuations that may be caused by residual noise or secondary side reactions. This threshold is set to 5%–10% of the theoretical height of the main peak. In practice, the unit first applies a peak detection algorithm to the smoothed IC curve. This algorithm scans and marks all local maximum positions by calculating the zero-crossing points of the first derivative and the negative points of the second derivative of the curve. For example, in one detection, the algorithm identified three candidate peaks: peak A (height 45.5 Ah / V), peak B (height 12 Ah / V), and peak C (height 5 Ah / V). The set value is... Since the intensity is 10 Ah / V, peak C is directly removed due to insufficient significance (5 < 10). Among the remaining peaks A and B, the system further filters out peak A, which has the highest intensity, as the main characteristic peak. For lithium iron phosphate batteries, this main peak corresponds to the most intense phase transition process in the voltage plateau region (around 3.30V). After selecting the main peak, the cell accurately extracts its two core parameters: peak position. and peak intensity . This refers to the voltage value corresponding to the peak value on the horizontal axis, such as 3.285V; This refers to the IC value corresponding to the peak apex on the vertical axis, such as 45.5 Ah / V. These two parameters are encapsulated within the characteristic peak parameter set. ={3.285V, 45.5Ah / V}. These two parameters have clear physical meanings: The decrease directly reflects the loss of battery capacity, while The shift of the polarization towards lower voltage reflects an increase in polarization resistance.

[0068] Subsequently, the initial health calculation unit 152 processes the extracted feature parameters based on the standard aging model library. The standard aging model library is a core knowledge base that stores experimental data obtained through extensive offline full-lifecycle cyclic testing of the same battery model. By performing multivariate regression analysis on this experimental data, the quantitative mapping relationship between SOH and IC feature parameters was determined. The library contains a specific set of regression coefficients. For example, after calibration, the obtained model coefficients are: =0.015, =-0.2, =0.85. These coefficients solidify the mathematical relationship between characteristic changes and SOH decay. This unit loads these coefficients and performs SOH estimation based on the eigenmap model. The calculation formula is a multiple linear regression equation: Substitute the previously extracted feature values: =0.015×45.5+(-0.2)×3.285+0.85=0.8755, meaning the initial estimated SOH is 87.55%. At this point, the unit will perform a boundary constraint check on the calculation result to confirm whether it is within the reasonable physical range of [0,1]. If the calculation result exceeds this range (e.g., a calculated value of 1.05 due to abnormal characteristic parameters), the system will forcibly clamp it to a boundary value (such as 1.00) to ensure the physical validity of the output. While it reflects the state of the battery, it incorporates the effects of temperature and has not been normalized.

[0069] Finally, the temperature compensation correction unit 153 intervenes to eliminate the interference of ambient temperature on the evaluation results. The electrochemical reaction rate of the battery follows the Arrhenius equation; the lower the temperature, the lower the reaction activity and the smaller the exhibited capacity; and vice versa. To obtain the true SOH at a standard temperature (e.g., 25°C), the temperature deviation during measurement must be eliminated. In an exemplary embodiment, the temperature compensation correction unit 153 is used to: perform temperature compensation correction on the initial health status using the following formula:

[0070]

[0071]

[0072] in, The reference temperature used for model calibration. The average discharge temperature. The apparent activation energy during battery aging. Let be the ideal gas constant. This is the temperature correction factor. For initial health, The health status of the current bottleneck cell is assessed. The unit first obtains the average discharge temperature. For example, 20°C, which is 293.15K, and the reference temperature in the standard aging model library. The aging activation energy is 25℃, or 298.15K. For example, 30000 J / mol. Simultaneously, the ideal gas constant is introduced. =8.314 J / (mol·K). The temperature correction factor was calculated based on the Arrhenius formula. Substitute the numerical values ​​into the calculation: =exp{30000 / 8.314×(1 / 298.15-1 / 293.15)}≈0.814, this factor indicates that the performance measured at 20℃ is only equivalent to 81.4% of that under standard conditions. Next, the initial SOH is corrected using a correction factor. Since temperature primarily affects the apparent capacity decay of the battery... The actual attenuation is greater than that under standard conditions, therefore the correction logic should remove the additional attenuation component caused by temperature. Substitute the values: =1-{(1-0.8755)×0.814}=0.8987, the final calculated value is... The result is 89.87%. This result reasonably reflects the true health level of the weak monomer under standard conditions after removing the inhibitory factors of low temperature.

[0073] Specifically, the actual available capacity analysis module 160 is used to determine the actual available capacity of the current battery pack based on the health status of the current weakest individual cell and the nominal capacity of the battery. In other words, for critical application scenarios such as communication base stations, substations, and data centers, the core requirement for maintenance personnel is to clearly know the specific duration for which the backup power supply can maintain load operation after a mains power outage. This directly depends on the absolute amount of electricity (ampere-hours) that the battery pack can currently release, rather than just a relative health percentage. Although the preceding steps have accurately assessed the health status of the weakest individual cell limiting the overall performance of the pack through complex feature extraction and model mapping, this dimensionless indicator cannot directly estimate the backup power duration based on load power consumption. Furthermore, because the battery pack follows the series bottleneck principle, its overall external capacity limit is constrained by the actual capacity of the worst-performing individual cell; any estimation based on the average level will bring significant safety risks. Therefore, determining the actual usable capacity of the current battery pack based on the health of the current weakest individual cells and the nominal capacity of the battery is to convert the relative SOH index obtained from electrochemical analysis into a capacity value with actual physical meaning in engineering applications, thereby quantifying the actual power supply capability of the battery pack under the current environmental conditions and providing the final quantitative basis for subsequent replacement decisions or backup power duration prediction.

[0074] In detail, in the above implementation scheme, the actual available capacity analysis module 160 processes the following: This module first obtains two key input parameters: one is the current shortest slab unit health status calculated and corrected by the shortest slab unit health status analysis module. Continuing the calculation results from the previous embodiments, this value is 0.8987, or 89.87; secondly, the nominal capacity of the battery. The nominal capacity of a battery refers to the rated amount of electricity that the battery should be able to release under standard test conditions, as defined by the battery manufacturer in its specifications, and is measured in ampere-hours (Ah). This parameter is known system configuration information and is stored as static asset data in the non-volatile memory of the battery management system (BMS) or in the database of the cloud-based operations and maintenance platform. It is read and loaded by modules during system initialization or when a capacity verification task is started. Taking the lithium iron phosphate battery pack of the communication base station monitored in this embodiment as an example, its nominal capacity is set at 100Ah at the factory.

[0075] After obtaining the above parameters, the module executes the core capacity conversion logic. Given that industrial battery packs generally use a series topology, the discharge cutoff point of the entire battery pack depends on the cell whose voltage first drops to the lower limit, i.e., the shortest cell identified in the aforementioned steps. Therefore, the current actual capacity of the shortest cell is equivalent to the system capacity presented to the outside world by the entire battery pack. In an exemplary embodiment, the actual usable capacity analysis module 160 is used to: determine the current actual usable capacity of the battery pack using the following formula:

[0076]

[0077] in, This refers to the battery's nominal capacity. This represents the current actual usable capacity of the battery pack. In the formula, This represents the current actual available capacity and is the ultimate assessment target; This refers to the battery's nominal capacity, representing the design baseline. The health coefficient represents the percentage of the battery's current condition relative to its design baseline. The physical meaning of this formula is that the battery's current effective maximum capacity is its theoretical capacity in its brand-new state multiplied by its current overall health coefficient. Substituting specific values ​​into the formula for calculation: =100Ah × 0.8987 = 89.87Ah. The calculation results show that the actual usable capacity of the current battery pack is 89.87Ah. This means that although the battery pack is labeled as 100Ah, under the combined effects of accumulated aging damage and a low temperature of 20℃, the maximum amount of electricity it can actually release to the load is only 89.87Ah. This value not only takes into account irreversible physical aging factors such as the loss of active materials inside the battery, but also rigorously includes the temporary limitation of discharge capacity caused by ambient temperature. The system finally reports this value to the operation and maintenance center, and the operation and maintenance personnel can use this to accurately calculate that if the current base station load current is 50A, the battery pack can only provide about 1.8 hours (89.87Ah / 50A) of battery life, thus providing accurate data support for whether to dispatch an emergency power generation vehicle.

[0078] In summary, the remote online capacity assessment system 100 for batteries based on the embodiments of this application is explained, addressing the technical challenges of capacity assessment difficulties and the masking of bottleneck effects under shallow charge and discharge conditions. The system first cleans the raw time-series data uploaded by the battery management system, extracting discharge segments containing valid information, and then analyzes the fragmented data from non-full charge / discharge conditions. Based on this, by analyzing the dynamic voltage drop and internal resistance characteristics within the discharge segments, it accurately identifies the bottleneck cells that limit the overall capacity of the battery pack due to performance degradation, obtaining their unique voltage and current sequences, effectively solving the problem of traditional methods neglecting individual cell differences due to focusing on the overall average. Furthermore, incremental capacity analysis technology is used to process the bottleneck cell data, identifying the characteristic peaks of the smoothed incremental capacity sequence to map health status, and then combining this with the nominal capacity to calculate the actual usable capacity of the current battery pack, achieving accurate evaluation of battery pack performance in a remote online environment.

Claims

1. A remote on-line battery capacity system, comprising: The method comprises the following steps: An original time series data stream acquisition module is configured to acquire an original time series data stream uploaded by a battery management system, wherein each data point in the original time series data stream comprises an i-th single cell voltage, a loop current and an ambient temperature; a time series data stream cleaning module is configured to clean the original time series data stream to obtain a set of valid discharge segments; a segment single cell anchoring module is configured to anchor the set of valid discharge segments based on a short board effect to obtain a short board single cell index, a short board single cell voltage sequence and a current sequence; an incremental capacity construction module is configured to construct a smoothed incremental capacity sequence based on the short board single cell voltage sequence and the current sequence; and a short board single cell health degree analysis module is configured to identify a feature peak of the smoothed incremental capacity sequence and map the feature peak to a health degree to obtain a current short board single cell health degree. An actual available capacity analysis module is configured to determine a current battery pack actual available capacity based on the current short board single cell health degree and a battery nominal capacity.

2. The battery remote on-line state-of-charge system according to claim 1, wherein, The time series data stream cleaning module comprises a data filtering unit configured to filter the original time series data stream by a sliding window to obtain a filtered data stream; a discharge segment identification and segmentation unit configured to identify and segment valid discharge segments from the filtered data stream based on a discharge current threshold and a minimum duration threshold to obtain a set of candidate discharge segments; and a data resampling and alignment unit configured to resample and align the set of candidate discharge segments to obtain the set of valid discharge segments.

3. The battery remote on-line state-of-charge system according to claim 2, wherein, The segment single cell anchoring module comprises a single cell dynamic characteristic parameter extraction unit configured to extract single cell dynamic characteristic parameters from each valid discharge segment in the set of valid discharge segments to obtain a feature index set, wherein the feature index set comprises a dynamic pressure drop rate, a dynamic internal resistance and a terminal voltage; an index set normalization unit configured to aggregate and normalize the feature index set to obtain a normalized comprehensive score matrix; a discharge segment weighted scoring unit configured to determine a short board single cell based on a weighted score of the normalized comprehensive score matrix to obtain the short board single cell index; and a voltage and current sequence extraction unit configured to extract the short board single cell voltage sequence from the set of valid discharge segments and the current sequence from the original time series data stream based on the short board single cell index.

4. The battery remote on-line state-of-charge system according to claim 3, wherein, The discharge segment weighted score unit is configured to perform short-board single cell decision based on weighted score on the effective discharge segment set according to the following formula: ; ; wherein, is a comprehensive health score of single cell i, is a dynamic internal resistance, is a dynamic pressure drop rate, is a terminal voltage, , and are a dynamic internal resistance weight coefficient, a dynamic pressure drop rate weight coefficient and a terminal voltage weight coefficient respectively, is a normalization processing, is an index found to make maximum, , is a short-board single cell index.

5. The battery remote on-line state-of-charge system according to claim 1, wherein, The incremental capacity construction module comprises a voltage-capacity data point set construction unit configured to construct a voltage-capacity data point set based on the short board single cell voltage sequence and the current sequence; a voltage domain resampling unit configured to resample the voltage domain and calculate an original IC curve to obtain an original incremental capacity sequence; and a curve Gaussian filtering smoothing unit configured to perform Gaussian filtering smoothing on the original incremental capacity sequence based on a Gaussian standard deviation to obtain the smoothed incremental capacity sequence.

6. The battery remote on-line state-of-charge system of claim 1, wherein, The short plate monomer health degree analysis module comprises: a health indication feature extraction unit, configured to perform IC curve health indication feature extraction on the smoothed incremental capacity sequence based on a peak significance threshold to obtain a feature peak parameter set; an initial health degree calculation unit, configured to perform initial SOH estimation based on a feature mapping model on the feature peak parameter set based on a standard aging model library to obtain an initial health degree; and a temperature compensation correction unit, configured to perform temperature compensation correction on the initial health degree based on a discharge average temperature to obtain the health degree of the current short plate monomer.

7. The battery remote on-line state-of-charge system according to claim 6, wherein, The temperature compensation correction unit is configured to perform temperature compensation correction on the initial health degree according to the following formula: ; ; wherein, is a reference temperature used for model calibration, is a discharge average temperature, is an apparent activation energy for battery aging, is an ideal gas constant, is a temperature correction factor, is an initial state of health, is a current state of health of the weakest cell.

8. The battery remote on-line state-of-charge system of claim 1, wherein, The actual available capacity analysis module is configured to determine the actual available capacity of the current battery pack according to the following formula: ; wherein, is the battery nominal capacity, is the current battery pack actual available capacity.

Images