Power battery SOH prediction method based on internal resistance-SOC calibration

Abstract

The invention relates to the technical field of power batteries, and discloses a power battery SOH prediction method based on internal resistance-SOC calibration. The method comprises the following steps: firstly, acquiring multi-source data of the power battery, determining calibration characteristic parameters of each data unit, and then calculating prediction characteristic indexes and generating a prediction input set; and then loading a pre-trained SOH prediction model to determine dynamic change parameters, generating a calibration prediction map, and performing SOH prediction after data consistency detection. If the data conflict is detected, predicting SOH according to a multi-source data priority fusion parameter; after prediction, unit attribute difference values are monitored in real time, and parameters are corrected and prediction results are updated when threshold values are exceeded. The pre-training model is obtained by obtaining a historical data mapping relation and performing iterative optimization, and the internal resistance-SOC calibration model contains a three-dimensional state layer for visualization. The method makes full use of multi-source data, and improves the accuracy and reliability of power battery SOH prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries, and in particular to a power battery SOH prediction method based on internal resistance-SOC calibration. Background Art

[0002] With the booming development of electric vehicles, accurate prediction of the State of Health (SOH) of power batteries, a core component, is crucial. It directly affects the range, safety, and service life of electric vehicles, and is a key link in ensuring the reliable operation of electric vehicles.

[0003] Traditional power battery SOH prediction methods have numerous limitations. Early empirically based prediction methods often relied on simple physical models and limited experimental data, ignoring the battery's complex internal characteristics and changing operating environments. These methods can only provide a rough estimate of the battery's SOH. In actual application scenarios, the accuracy and reliability of the prediction results are significantly compromised under varying operating and environmental conditions, making it difficult to meet the needs of electric vehicle users for precise battery performance assessments.

[0004] While prediction methods based on battery internal resistance or a single parameter do account for some of the battery's physical properties, batteries are complex electrochemical systems, and their SOH is influenced by multiple factors. Relying solely on a single parameter cannot fully capture the changing state of the battery, limiting prediction accuracy. For example, under varying temperatures, the variation in battery internal resistance is nonlinear and intertwined with factors such as the battery's charge / discharge state and aging. Therefore, a single internal resistance parameter cannot accurately reflect a battery's true SOH.

[0005] Data-driven prediction methods have been widely studied in recent years. However, existing data-driven models have shortcomings when processing multi-source data. On the one hand, data collected by different monitoring devices suffer from issues such as accuracy differences, missing data, and noise interference, resulting in uneven data quality, which affects model training and prediction accuracy. On the other hand, these models often fail to fully exploit the deep connections between multi-source data and the complex mapping relationship between data and battery SOH, making it impossible to effectively utilize the complementary information from multi-source data to improve prediction performance.

[0006] Furthermore, during actual use, the battery's operating state is constantly changing, and its SOC (State of Charge) also varies over time and operating conditions. Existing prediction methods rarely account for the dynamic changes in battery SOC while incorporating key parameters such as internal resistance for calibration, resulting in deviations between the predicted results and the battery's actual state. This not only misleads users about the battery's remaining charge and health status, but can also cause safety issues such as overcharging and over-discharging, seriously impacting the safety of electric vehicles and the battery's service life. Therefore, it is urgent to develop a method that can effectively overcome these issues and accurately predict the SOH of power batteries. Summary of the Invention

[0007] The object of the present invention is to provide a power battery SOH prediction method based on internal resistance-SOC calibration to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the SOH of a power battery based on internal resistance-SOC calibration, the method comprising:

[0009] Acquire multi-source data of the power battery and determine calibration characteristic parameters for each data unit from a preset internal resistance-SOC calibration model, wherein the calibration characteristic parameters include operating state attributes, environmental attributes, and battery physical attributes of the data unit;

[0010] For each data unit, a prediction characteristic index of the data unit is calculated based on the calibration characteristic parameters, and a prediction input set for the data unit is generated based on multiple prediction characteristic indicators; wherein the prediction characteristic index is a comprehensive quantitative parameter for each data unit that integrates its operating state attributes and battery physical properties; and the prediction input set is a standardized data set composed of multiple selected prediction characteristic indicators;

[0011] For each data unit, a pre-trained SOH prediction model is loaded, and based on the correlation between multiple feature indicators included in the prediction input set and the inference rules of the SOH prediction model, a dynamic change parameter of the data unit during the prediction process is determined;

[0012] Marking the dynamic change parameter for each data unit in the internal resistance-SOC calibration model to generate a calibration prediction map;

[0013] For each data unit, perform data consistency detection according to the dynamically changing parameters to obtain a detection result;

[0014] The power battery SOH is predicted based on the detection result to generate an SOH prediction result.

[0015] Preferably, the power battery SOH prediction is performed based on the detection result to generate the SOH prediction result, including: when the detection result indicates that there is a conflict among multiple dynamically changing parameters contained in the data unit, determining the data unit as an abnormal unit; obtaining the conflict parameters of the abnormal unit as a correction object, and determining the multi-source data priority corresponding to the conflict parameters; performing parameter fusion on the data unit based on the multi-source data priority to generate the SOH prediction result.

[0016] Preferably, the multi-source data includes first monitoring device data and second monitoring device data, the multi-source data priority of the first monitoring device data is a first priority, and the multi-source data priority of the second monitoring device data is a second priority; performing parameter fusion on the data units based on the multi-source data priority to generate the SOH prediction result includes:

[0017] adjusting a prediction input set corresponding to the second monitoring device data according to an order of the first priority and the second priority;

[0018] extracting, based on the calibrated prediction map, a dynamic change parameter of the second monitoring device data in the adjusted prediction input set;

[0019] The dynamically changing parameters are combined with the calibration prediction map to perform parameter fusion to generate an SOH prediction result.

[0020] Preferably, for each data unit, respectively calculating the prediction feature index of the data unit based on the calibration feature parameters, and generating a prediction input set of the data unit according to the plurality of the prediction feature indexes, includes:

[0021] For each data unit, determining a plurality of adjacent units adjacent to the running state of the data unit;

[0022] Calculating the correlation between the calibration characteristic parameter of each adjacent cell and the physical property of the battery, and determining a first characteristic cell with the highest correlation from the adjacent cells based on a plurality of the correlation degrees;

[0023] Based on the first characteristic unit, determining a plurality of extension units adjacent to the characteristic unit, calculating a correlation between a calibration characteristic parameter of each extension unit and a physical property of the battery, and determining a next characteristic unit with the highest correlation from the extension units based on the plurality of correlations;

[0024] Repeating the steps of determining a plurality of extension units adjacent to the characteristic unit, calculating a correlation between the calibration characteristic parameter of each extension unit and the battery physical property, and determining a next characteristic unit with the highest correlation from the extension units based on the plurality of correlations, until all data units are traversed, to obtain a set of predicted characteristic indicators for the data units;

[0025] The prediction input set is generated based on the prediction feature indicator set.

[0026] Preferably, the calculating of the correlation between the calibration characteristic parameter of each adjacent cell and the physical property of the battery includes:

[0027] respectively calculating a first correlation between the operating state attribute of each of the adjacent cells and the battery physical attribute, and a second correlation between the environmental attribute of each of the adjacent cells and the battery physical attribute;

[0028] For each of the adjacent units, a weighted sum of the first degree of association and the second degree of association is used as the degree of association of the adjacent unit.

[0029] Preferably, the determining the first feature unit with the highest correlation degree from the adjacent units based on the plurality of correlation degrees includes:

[0030] storing a plurality of association degrees corresponding to a plurality of adjacent units into a feature candidate list, and sorting the association degrees in the feature candidate list in descending order to obtain a sorting result;

[0031] Based on the ranking result, the adjacent unit corresponding to the highest correlation degree is determined as the first feature unit.

[0032] Preferably, after performing power battery SOH prediction based on the detection result and generating the SOH prediction result, the method further includes:

[0033] After the SOH prediction result is generated, for each data unit, the attribute difference value between the data unit and the adjacent units is monitored in real time;

[0034] When the attribute difference value exceeds a preset threshold, the adjacent unit is used as an abnormal reference unit, and the dynamic parameter correction of the data unit is performed based on the abnormal reference unit to generate an updated SOH prediction result.

[0035] Preferably, the training process of the pre-trained SOH prediction model includes:

[0036] Acquire a historical power battery data set and extract a mapping relationship between calibration characteristic parameters and dynamic change parameters in the data set;

[0037] Constructing an initial SOH prediction model based on the mapping relationship, and iteratively optimizing the model using a cross-validation method;

[0038] When the error between the predicted parameters output by the model and the measured parameters is lower than a preset threshold, it is determined that the model training is completed.

[0039] Preferably, the internal resistance-SOC calibration model also includes a three-dimensional state layer map of battery temperature distribution, charge and discharge rate distribution, and battery aging degree. When generating the calibration prediction map, the dynamically changing parameters are superimposed on the three-dimensional state layer map for visual expression.

[0040] Preferably, the method for generating the three-dimensional state layer includes:

[0041] Collect temperature data, charge and discharge rate data, and aging degree data of power batteries;

[0042] Performing interpolation processing on the temperature data to generate a battery temperature distribution layer, performing classification coding on the charge and discharge rate data to generate a charge and discharge rate distribution layer, and performing spatial interpolation on the aging degree data to generate a battery aging degree layer;

[0043] The battery temperature distribution layer, the charge and discharge rate distribution layer, and the battery aging degree layer are spatially overlaid and analyzed to form the three-dimensional state layer.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] In terms of data processing, by acquiring multi-source data on the power battery and determining the calibration characteristic parameters of each data unit from the preset internal resistance-SOC calibration model, the various factors affecting the battery's SOH are fully considered, covering the operating state attributes, environmental attributes, and battery physical properties. This comprehensive data acquisition and characteristic parameter determination method can more accurately reflect the true state of the battery than traditional methods that rely on a single parameter or a small number of parameters. For example, in complex environments such as different temperatures and charge and discharge rates, multi-source data and calibration characteristic parameters can fully capture the operating changes of the battery, providing a richer and more accurate information basis for subsequent predictions.

[0046] In calculating predictive characteristic indicators and generating a predictive input set, predictive characteristic indicators are calculated for each data unit based on calibrated characteristic parameters, and a predictive input set is generated. By identifying adjacent units and calculating correlations, the inherent connections between data units are fully explored, making the predictive characteristic indicators comprehensive quantitative parameters that integrate operating state attributes and battery physical properties, and the predictive input set a standardized data set. This not only effectively utilizes the complementary information of multi-source data, but also improves data quality and availability, providing higher-quality input data for subsequent predictive models, thereby significantly improving prediction accuracy.

[0047] During the prediction process, a pre-trained SOH prediction model is loaded. Dynamically changing parameters are determined based on the correlation between the feature indicators in the prediction input set and the model's inference rules. A calibration prediction map is then generated within the internal resistance-SOC calibration model. This process fully leverages the model's learning and reasoning capabilities. Combined with the calibration model's visual representation (for example, overlaying dynamically changing parameters onto a three-dimensional state map that displays battery temperature distribution, charge / discharge rate distribution, and battery aging), the prediction process is more intuitive and interpretable. The calibration prediction map allows users and engineers to clearly understand the battery's changing trends under different conditions, enabling them to promptly identify potential issues and take appropriate measures.

[0048] In terms of data consistency detection and exception handling, data consistency detection is performed based on dynamically changing parameters. When a conflict in the dynamically changing parameters contained in a data unit is detected, the abnormal unit can be identified and parameter fusion is performed based on the priority of multi-source data to generate the SOH prediction result. This mechanism effectively solves the problems of conflict and error in multi-source data and improves the reliability of the prediction results. In actual applications, the data collected by different monitoring devices may differ. Through data priority and parameter fusion, these inconsistent data can be handled more reasonably, ensuring that the prediction results are closer to the actual SOH of the battery.

[0049] Furthermore, after the SOH prediction results are generated, the difference in attribute values ​​between the data unit and its adjacent units is monitored in real time. When the difference exceeds a preset threshold, the data unit's parameters are dynamically corrected and an updated SOH prediction result is generated. This function enables dynamic tracking and continuous optimization of the battery status, promptly reflecting changes in the battery status and ensuring that the prediction results maintain a high level of accuracy. Whether the battery is in operation or undergoing gradual performance changes due to long-term use, it can provide users with reliable battery SOH information, ensuring the safe operation of the electric vehicle and the rational use of the battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a working principle diagram of the power battery SOH prediction method based on internal resistance-SOC calibration according to the present invention;

[0051] Figure 2 This is a working principle diagram of power battery SOH prediction when data conflict exists;

[0052] Figure 3 A diagram showing the working principle for calculating the correlation between adjacent cell calibration characteristic parameters and battery physical properties;

[0053] Figure 4 Workflow diagram for determining the first feature unit with the highest correlation;

[0054] Figure 5 A diagram showing the working principle of correcting prediction results based on the attribute differences between data cells and adjacent cells. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] See also Figure 1-Figure 5 The present invention relates to a power battery SOH prediction method based on internal resistance-SOC calibration. The specific implementation scheme is as follows:

[0057] Acquire multi-source data on power batteries. This data comes from a wide range of sources, encompassing a wide range of information collected by different monitoring devices. The calibration characteristic parameters for each data unit are determined from a preset internal resistance-SOC calibration model. These calibration characteristic parameters include the unit's operating state attributes, such as charge and discharge current and voltage fluctuations; environmental attributes, such as ambient temperature and humidity; and battery physical properties, such as internal resistance and capacity.

[0058] For each data unit, a predictive characteristic index is calculated based on the calibrated characteristic parameters. This predictive characteristic index is a comprehensive quantitative parameter that integrates the data unit's operating state attributes and the battery's physical properties. After calculating multiple predictive characteristic indicators, a prediction input set is generated based on these indicators. The prediction input set is a standardized data set composed of multiple selected predictive characteristic indicators.

[0059] For each data unit, a pre-trained SOH prediction model is loaded. Based on the correlation between multiple feature indicators in the prediction input set and the SOH prediction model's own inference rules, the dynamic parameters of the data unit during the prediction process are determined. These dynamic parameters reflect the changes in the characteristics of the data unit under different circumstances.

[0060] In the internal resistance-SOC calibration model, each data unit is marked with the above-determined dynamic change parameter, and a calibration prediction map is generated through this marking. The map presents the relationship between the data unit and the dynamic change parameter in an intuitive form.

[0061] For each data unit, data consistency detection is performed based on dynamically changing parameters to determine whether there are any abnormalities in the data, and then the detection results are obtained.

[0062] The power battery SOH is predicted based on the test results, and the SOH prediction result is finally generated.

[0063] The present invention will be further described below in conjunction with Examples 1 to 5:

[0064] Example 1:

[0065] In this embodiment, in the calculation of the prediction characteristic index, the specific operations of determining adjacent units, calculating the correlation degree, and determining the first characteristic unit are described in detail. After obtaining multi-source data of the power battery, for a specific data unit, the multiple adjacent units adjacent to it in time or space are first found based on its operating status. In actual application scenarios, such as in a real-time monitoring system for a power battery pack, the adjacent units may be battery data units corresponding to the previous and next acquisition moments that are close in time, or data units that are spatially close to the battery unit (such as adjacent batteries in the same battery module).

[0066] After determining the adjacent cells, the first correlation between the operating state attributes and the battery's physical properties, as well as the second correlation between the environmental attributes and the battery's physical properties, are calculated for each adjacent cell. There's a certain correlation between the current magnitude and the battery's internal resistance, two attributes of the operating state attribute. When current is transmitted within the battery, it is hindered by the internal resistance, which in turn affects the battery's performance and state. The correlation between these two attributes can be determined by analyzing their changing trends. There's also a correlation between temperature and battery capacity, an environmental attribute. Different ambient temperatures affect the battery's chemical reaction rate, thus affecting the battery's capacity. Similarly, a suitable algorithm can be used to calculate the correlation between these two attributes. For each adjacent cell, the first and second correlations are weighted and calculated according to a certain weight to obtain the correlation for that adjacent cell.

[0067] The multiple correlation degrees corresponding to the multiple adjacent units are stored in a feature candidate list, and then the correlation degrees in the feature candidate list are sorted in descending order. The sorting algorithm can use common algorithms such as quick sort to arrange the correlation degrees from high to low. Based on the sorting result, the adjacent unit corresponding to the highest correlation degree is determined as the first feature unit. Then, based on the first feature unit, multiple extended units adjacent to the feature unit are determined, the correlation degree of each extended unit is calculated again, and the next feature unit with the highest correlation degree is determined. This cycle is repeated until all data units are traversed, and a set of prediction feature indicators of the data units is obtained, and then a prediction input set is generated.

[0068] Suppose there is an electric vehicle power battery pack, which contains multiple battery cells. During the vehicle's driving, various types of battery data are continuously collected through the on-board monitoring system. These data will be divided into data units for processing.

[0069] At a given moment, we focus on a specific data unit, A. To calculate the predictive characteristic indicators for data unit A, we first need to identify multiple neighboring units adjacent to its operating state. For example, if the vehicle monitoring system collects data once per second, then the neighboring units of data unit A might be the data units collected one second before and one second after it, denoted as B and C, respectively.

[0070] Next, the correlation between the calibration characteristic parameters of adjacent cells and the battery's physical properties is calculated. For example, consider the current (operating state attribute) and the internal resistance (battery physical attribute). Assume that during data collection, the battery's operating current varies at different times, and the internal resistance also changes with battery usage and status. The correlation can be calculated by analyzing the trend relationship between current and internal resistance changes. If, over a certain period of time, the internal resistance shows a significant upward trend as the current increases, the correlation is considered high. Conversely, if the current change has little effect on the internal resistance, the correlation is low. For the temperature (environmental attribute) and the capacity (battery physical attribute), if the battery capacity shows a downward trend as the ambient temperature increases, this correlation can also be calculated by analyzing this relationship. The correlation between the operating state attribute and the battery's physical properties for each adjacent cell is defined as the first correlation, and the correlation between the environmental attribute and the battery's physical properties is defined as the second correlation. For each adjacent cell, the first and second correlations are weighted and summed according to a certain weight to obtain the correlation for that adjacent cell. Assume that the first correlation of adjacent unit B is 0.6, the second correlation is 0.4, and the weights are set to 0.7 and 0.3 respectively, then the correlation of B is ; Similarly, the correlation degree of adjacent unit C is calculated to be 0.48.

[0071] After calculating the associations of adjacent units, these associations are stored in a feature candidate list. This list includes an association of 0.54 for B and 0.48 for C. The associations in the feature candidate list are then sorted in descending order. B has the highest association. Based on this, B is selected as the first feature unit.

[0072] Based on the first characteristic cell B, identify multiple extended cells adjacent to B. Assume that, in addition to cells A and C, B's adjacent cells also include D and E, which are spatially close to each other. Next, calculate the correlation between the calibration characteristic parameters and the battery's physical properties for each extended cell (D and E), using the same method used to calculate the correlation between B and C. Assuming the correlation for D is 0.5 and for E is 0.45, sort these correlations in descending order again, and determine that the next characteristic cell with the highest correlation is D.

[0073] In this way, multiple extension units adjacent to the feature unit are continuously identified, the correlation of each extension unit is calculated, and the next feature unit with the highest correlation is determined. This process continues until all data units have been traversed. Ultimately, a set of predicted feature indicators for data unit A is obtained, and then a set of predicted inputs is generated based on this set of predicted feature indicators. This approach provides more accurate and representative input data for subsequent SOH prediction models, thereby improving prediction accuracy.

[0074] Example 2:

[0075] This embodiment focuses on generating a calibration prediction map. After determining the dynamic change parameters of the data unit, the calibration prediction map is generated. The internal resistance-SOC calibration model includes a three-dimensional state map of the battery temperature distribution, charge and discharge rate distribution, and battery aging. First, the power battery's temperature data, charge and discharge rate data, and aging data are collected. These data can be obtained through various sensors installed on the battery.

[0076] For temperature data, interpolation is used to generate a battery temperature distribution layer. Interpolation is a mathematical method that uses known discrete temperature data points and a specific interpolation algorithm (such as linear interpolation or spline interpolation) to estimate the temperature values ​​at other locations, thereby constructing a continuous temperature distribution layer. This allows us to intuitively see the temperature distribution of the battery at different locations or at different times.

[0077] Classify and encode the charge and discharge rate data to generate a charge and discharge rate distribution layer. Classification and coding divides the charge and discharge rate ranges into different categories and assigns each category a specific code. This allows the charge and discharge rate data to be converted into a form suitable for visualization, facilitating subsequent analysis and viewing.

[0078] For aging data, we use spatial interpolation to generate a battery aging layer. Similar to interpolation of temperature data, by processing battery aging data at known locations, we estimate the aging at other locations, thereby generating a layer that reflects the overall aging distribution of the batteries.

[0079] Finally, the battery temperature distribution layer, charge / discharge rate distribution layer, and battery aging layer are spatially overlaid to form a three-dimensional state layer. During this spatial overlay analysis, the three layers are superimposed according to their respective spatial locations, so that each location contains information on temperature, charge / discharge rate, and aging. Dynamically changing parameters are then overlaid onto the three-dimensional state layer for visualization, generating a calibration prediction map. This generated calibration prediction map comprehensively and intuitively displays various battery state information and the dynamically changing parameters during the prediction process, providing strong support for subsequent analysis and prediction.

[0080] Imagine we are monitoring and analyzing a power battery system used in an energy storage power station to generate a calibrated prediction map. In this energy storage power station, the power battery pack consists of a large number of battery cells. Sensors installed at various key locations continuously collect battery temperature data, charge and discharge rate data, and aging data.

[0081] For temperature data processing, consider a situation where, at a given moment, sensors collect temperature values ​​at multiple discrete points within a battery pack. Assume the battery pack is arranged in a rectangle, with temperature sensors installed at five points: the top left corner, top right corner, bottom left corner, bottom right corner, and center. The collected temperatures are 25°C, 26°C, 24°C, 25°C, and 25.5°C, respectively. To generate a continuous battery temperature distribution map, we employ a linear interpolation algorithm. For example, we use the area between the top left and top right corners of the battery pack to estimate the temperature values ​​of each intermediate point based on the temperature values ​​of these two points and the distance between them, using the linear interpolation rule. Assuming the two points are 10 cm apart horizontally, we estimate the temperature 3 cm from the top left corner. Using the linear interpolation formula (assuming a linear relationship holds), we calculate the temperature at that point to be approximately 25.3°C. Similarly, interpolation calculations are performed for each region of the battery pack, ultimately generating a temperature distribution map for the entire pack. This map allows us to visually visualize the temperature variations at different locations within the pack.

[0082] Regarding charge and discharge rate data processing, the battery charge and discharge rate range of energy storage power stations is between 0.5C and 2C (C represents the rated capacity rate of the battery). We divide this range into three categories: low rate (0.5C-1C), medium rate (1C-1.5C), and high rate (1.5C-2C), and assign them codes 0, 1, and 2, respectively. When the battery charge and discharge rate is 1.2C at a certain moment, after classification and coding, the corresponding code for this data is 1. After this classification and coding process is performed on all charge and discharge rate data collected over a period of time, a charge and discharge rate distribution layer is generated. This layer can clearly show the distribution of battery charge and discharge rate categories at different times and locations.

[0083] For aging data, we assess the battery's aging by analyzing factors such as the number of charge and discharge cycles and usage time. Assume that through long-term monitoring and analysis, we obtain aging data for several key locations in the battery pack (expressed as a percentage, with 0 representing brand new and 100% representing fully aged). Using spatial interpolation, similar to the processing of temperature data, we estimate aging values ​​at other locations based on the aging data at known locations, thereby generating a battery aging layer.

[0084] Finally, a spatial overlay analysis is performed, overlaying the battery temperature distribution layer, the charge / discharge rate distribution layer, and the battery aging layer. In the overlaid three-dimensional status layer, each location simultaneously contains all three aspects of information. For example, at a specific location in the battery pack, the three-dimensional status layer shows that the current temperature is 25°C, the charge / discharge rate is coded as 1 (i.e., in a medium-rate charge / discharge state), and the aging level is 30%. After determining the dynamic change parameters for each data unit, these parameters are overlaid onto the three-dimensional status layer to generate a calibrated prediction map. By viewing the calibrated prediction map, operations and maintenance personnel can fully understand the operating status and prediction information of the battery pack, providing a strong basis for the management and maintenance of the energy storage power station.

[0085] Example 3:

[0086] In this embodiment, the specific situation of predicting the power battery SOH based on the test results is emphasized. After completing the data consistency test and obtaining the test results, if the test results show that there are conflicts in multiple dynamically changing parameters contained in the data unit, a series of processing is required to generate the SOH prediction result.

[0087] The data unit with parameter conflict is determined as an abnormal unit, and the conflict parameter of the abnormal unit is obtained as the correction object. Assume that the multi-source data includes the first monitoring device data and the second monitoring device data, and the data priority of the first monitoring device data is the first priority, and the data priority of the second monitoring device data is the second priority. According to the order of the first priority and the second priority, the prediction input set corresponding to the second monitoring device data is adjusted. During the adjustment process, the arrangement order of the second monitoring device data in the prediction input set, data selection, etc. may be changed according to certain characteristics of the first monitoring device data.

[0088] Based on the calibration prediction map, the dynamic change parameters of the second monitoring device data in the adjusted prediction input set are extracted. The calibration prediction map contains rich information, and a specific extraction algorithm can accurately obtain the dynamic change parameters related to the second monitoring device data.

[0089] Parameter fusion is performed by combining the acquired dynamically changing parameters with the calibrated prediction map. Various methods can be used for parameter fusion, such as weighted fusion, which assigns weights to different parameters based on their importance and then fuses these parameters together. This parameter fusion ultimately generates the SOH prediction result. This approach, which targets outliers, effectively resolves data conflicts and improves the accuracy of SOH prediction.

[0090] Assume that there is a large energy storage power station, which is composed of multiple battery modules. Each battery module is equipped with a first monitoring device and a second monitoring device to collect battery data for subsequent SOH prediction and analysis.

[0091] At a certain point, while processing data from one of the battery modules, data consistency testing discovered that a data unit contained multiple conflicting dynamically changing parameters. For example, the battery internal resistance calculated from the battery voltage and current collected by the first monitoring device differed significantly from the battery internal resistance calculated using a different algorithm by the second monitoring device, exceeding the allowable error range. At this point, the data unit was identified as an abnormal unit, and its conflicting parameters (the two different internal resistance values) became the subject of correction.

[0092] It is known that the data priority of the first monitoring device data is the first priority, and the data priority of the second monitoring device data is the second priority. According to the priority order, the prediction input set corresponding to the second monitoring device data is adjusted. The prediction input set contains multiple battery-related parameter information, such as voltage, current, temperature, etc. During the adjustment process, the weight of the second monitoring device data in the prediction input set is adjusted according to the reliability and priority of the first monitoring device data. Assume that the prediction input set can be expressed in the form of a vector ,in ( ) represent different parameters. For the parameters collected by the second monitoring device (Assume is the parameter affected by the conflict, such as internal resistance), the adjusted weight According to the formula Calculate, where yes The original weight, is the priority weight of the first monitoring device data (assuming ), is the priority weight of the second monitoring device data (assuming ). By adjusting the relevant parameter weights through this formula, the prediction input set corresponding to the second monitoring device data can be adjusted.

[0093] After the adjustment is complete, the dynamically changing parameters of the second monitoring device data within the adjusted prediction input set are extracted based on the calibrated prediction map. The calibrated prediction map contains a large amount of information about parameter changes under different battery states. Using a specific index and query algorithm, the dynamically changing parameters corresponding to the adjusted second monitoring device data are found.

[0094] Finally, the extracted dynamic parameters are combined with the calibration prediction map for parameter fusion. This fusion process takes into account the importance and interrelationships of each parameter. For example, the adjusted internal resistance parameter is analyzed against the standard internal resistance range for the corresponding state in the calibration prediction map, as well as other relevant battery properties (such as temperature and charge / discharge rate). These parameters are then fused using methods such as weighted averaging (assuming the weights are determined based on the battery's operating status and historical data experience), ultimately generating a more accurate SOH prediction result. This prediction helps energy storage plant managers understand the battery's health status in a timely manner, plan maintenance plans in advance, and ensure stable operation of the plant.

[0095] Example 4:

[0096] This example details the training process of a pre-trained SOH prediction model. First, a historical power battery dataset is acquired. This dataset can be derived from a large number of operational data records of power batteries of different types and usage scenarios. A mapping relationship between calibration characteristic parameters and dynamically changing parameters is extracted from the historical dataset. In practice, through in-depth analysis of historical data, we observe how changes in calibration characteristic parameters (such as operating state attributes, environmental attributes, and battery physical properties) cause changes in dynamically changing parameters. This allows us to identify the inherent connections between these parameters and form a mapping relationship.

[0097] An initial SOH prediction model is constructed based on the obtained mapping relationship. The model can be constructed using some existing model architectures, such as neural network models and machine learning models, and the mapping relationship can be integrated into the model parameter settings or structural design.

[0098] Cross-validation is used to iteratively optimize the initial model. Cross-validation is a commonly used method for model evaluation and optimization. It divides the dataset into multiple subsets, and performs training and validation on different subsets. During each training session, a subset is used as the training set, and the remaining subsets as the validation set. Through repeated training and validation, the model parameters (such as the weights of the neural network and the hyperparameters of the machine learning algorithm) are continuously adjusted to ensure that the model performs well on different validation sets.

[0099] During the iterative optimization process, the predicted parameters output by the model are continuously compared with the measured parameters. When the error between the predicted and measured parameters falls below a preset threshold, the model is considered to have achieved good performance and training is complete. This preset threshold can be set based on the actual application needs and accuracy requirements. After multiple trials and adjustments, an appropriate value is found to ensure that the trained model meets the accuracy requirements for power battery SOH prediction.

[0100] Suppose we want to build and train a SOH prediction model for a certain brand of electric vehicle power batteries. This brand has a large amount of historical operating data of electric vehicles under different usage scenarios, which constitutes a historical power battery dataset.

[0101] During the data collection phase, various sensors on the vehicle record a wealth of information. For example, during a daily drive, from the moment the vehicle starts to the end of the drive, sensors continuously collect the battery's operating state properties. For example, the current increases during acceleration and decreases during deceleration; the voltage also changes continuously during the charging and discharging process. Regarding environmental properties, the outside temperature gradually rises from 15°C in the morning to 20°C at noon. Battery physical properties, such as internal resistance, change with battery use and temperature, and capacity slowly decays due to long-term charging and discharging. This data is recorded at regular intervals (such as every minute), forming data units. Numerous such data units comprise a historical dataset.

[0102] From this massive dataset, we began extracting mappings between calibration characteristic parameters and dynamically changing parameters. For example, we observed that over a certain period of time, when the battery temperature increased from 18°C ​​to 20°C (an environmental attribute change) and the current increased from 10A to 15A (an operating state attribute change), the battery's internal resistance increased from 50 milliohms to 55 milliohms (a physical attribute change). The corresponding dynamically changing parameter (such as the coefficient of variation of the battery's internal chemical reaction rate) also adjusted from 0.8 to 0.9. By analyzing large amounts of similar data, we were able to identify the inherent connections between these parameters and form a mapping relationship.

[0103] Based on these mapping relationships, we selected a neural network model to construct the initial SOH prediction model. In the neural network's architectural design, calibrated characteristic parameters are used as nodes in the input layer. For example, parameters such as temperature, current, voltage, and internal resistance are mapped to different input nodes. The appropriate number of layers and nodes in the hidden layer is determined based on experience and experimentation. These hidden layers are responsible for performing complex nonlinear transformations on the input data to extract deep-level features from the data. The output layer is set to predict relevant parameters for the battery's SOH, such as the percentage of remaining battery capacity.

[0104] After building the initial model, we iteratively optimize it using cross-validation. Suppose we divide the historical dataset into five subsets, selecting four of them as training sets and one as validation sets. In the first round of training, we use subsets 1-4 as training sets, and subset 5 as validation sets. During training, we continuously adjust the neural network's weights and biases to ensure that the model's predictions on the training set are as close to the actual SOH values ​​as possible. After training, we use the validation set to evaluate the model's performance, examining the error between the model's predicted SOH values ​​and the measured SOH values ​​in the validation set.

[0105] Suppose that in this round of validation, the average error between the model's predicted remaining battery capacity percentage and the actual value is 10%. Our pre-set error threshold is 5%, so the model's error falls short of the required threshold. Therefore, we proceed to the next round of training, selecting subsets 2-5 as the training set and subset 1 as the validation set, and repeating the training and validation process. With each round of training, the model's performance gradually improves, and the error decreases.

[0106] After multiple rounds of training, when the error between the predicted parameters output by the model in the validation set and the measured parameters is less than a preset threshold of 5%, the model training is considered complete. At this point, the trained SOH prediction model can be applied to actual EV battery SOH prediction, providing owners and automakers with accurate battery health information, enabling timely battery maintenance and replacement decisions.

[0107] Example 5:

[0108] This embodiment focuses on the subsequent processing after the SOH prediction results are generated. Once the SOH prediction results are successfully generated, the real-time monitoring phase begins. For each data unit, the attribute difference values ​​between the data unit and adjacent units are continuously monitored in real time. Attribute difference values ​​can include differences in operating state attributes (such as voltage and current), environmental attributes (such as temperature and humidity), and battery physical attributes (such as internal resistance and capacity).

[0109] During the monitoring process, a preset threshold is set. When the attribute difference value exceeds the preset threshold, the adjacent unit is used as an abnormal reference unit. This means that the attributes of the adjacent unit differ significantly from those of the current data unit, which may indicate an abnormality and require further processing of the current data unit.

[0110] Dynamic parameter correction is performed on the data unit based on the abnormal reference unit. The specific method of dynamic parameter correction can be determined based on the different attribute differences and the characteristics of the prediction model. For example, if the voltage difference is too large, the relevant voltage parameters of the current data unit in the prediction model may need to be adjusted based on the voltage value of the abnormal reference unit and the degree of difference between the two.

[0111] After dynamic parameter correction, the updated SOH prediction results are regenerated. This allows for timely detection of abnormalities in data units during actual battery operation, and updates of the prediction results by correcting dynamic parameters. This ensures that the SOH prediction results reflect the actual battery status in real time, improving the accuracy and reliability of the prediction and providing a more precise basis for power battery management and maintenance.

[0112] Assume that a fleet of electric buses is in operation. Each bus's power battery is equipped with a real-time monitoring system that continuously collects and processes battery data to predict the battery's state of health (SOH). The implementation of Example 5 will now be described in detail, using the power battery data processing process for one of these buses as an example.

[0113] At a certain moment, the bus's power battery monitoring system completes a SOH prediction and obtains the current SOH prediction result, for example, indicating that the battery's current state of health is 80%. From this moment on, the monitoring system begins to monitor the attribute difference between each data unit and its adjacent units in real time. Assume that the data units are collected in chronological order with a collection interval of 1 minute. The current data unit of interest is data unit 100, and its adjacent data units are data units 99 and 101.

[0114] The monitoring system primarily monitors battery voltage, current, temperature, and internal resistance. After a period of operation, the monitoring system discovered that the voltage difference between data unit 100 and data unit 101 exceeded a preset threshold. This threshold is set based on the battery's normal operating characteristics and historical data. For example, under normal circumstances, the voltage difference between adjacent data units should not exceed 0.1V. However, this time, the voltage difference between data units 100 and 101 reached 0.2V. At this point, the monitoring system identified data unit 101 as an abnormal reference unit.

[0115] Based on this anomalous reference cell, the system begins dynamic parameter correction for data cell #100. Because large voltage variations can affect the accuracy of SOH predictions, the system adjusts the relevant voltage parameters for data cell #100 in the prediction model based on the voltage value of data cell #101 and the degree of difference between the two. For example, in the prediction model, voltage parameters have a certain weight when calculating SOH, and the system will readjust this weight based on the voltage variation, or directly correct the voltage input value of data cell #100.

[0116] After dynamic parameter correction, the monitoring system reruns the SOH prediction model using the adjusted parameters to generate an updated SOH prediction. For example, if the original predicted SOH was 80%, after correction, the new prediction shows the battery's SOH to be 78%. This updated result more accurately reflects the battery's current actual health status. Based on this more accurate prediction, bus operators can rationally arrange bus charging plans, maintenance schedules, and operating routes, avoiding operational issues caused by misjudgments of battery health status and improving the operational efficiency and safety of the entire fleet.

[0117] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0118] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A power battery SOH prediction method based on internal resistance-SOC calibration, characterized in that: The method comprises: Acquire multi-source data of the power battery and determine calibration characteristic parameters for each data unit from a preset internal resistance-SOC calibration model, wherein the calibration characteristic parameters include operating state attributes, environmental attributes, and battery physical attributes of the data unit; For each data unit, a prediction characteristic index of the data unit is calculated based on the calibration characteristic parameters, and a prediction input set for the data unit is generated based on multiple prediction characteristic indicators; wherein the prediction characteristic index is a comprehensive quantitative parameter for each data unit that integrates its operating state attributes and battery physical properties; and the prediction input set is a standardized data set composed of multiple selected prediction characteristic indicators; For each data unit, a pre-trained SOH prediction model is loaded, and based on the correlation between multiple feature indicators included in the prediction input set and the inference rules of the SOH prediction model, a dynamic change parameter of the data unit during the prediction process is determined; Marking the dynamic change parameter for each data unit in the internal resistance-SOC calibration model to generate a calibration prediction map; For each data unit, perform data consistency detection according to the dynamically changing parameters to obtain a detection result; The power battery SOH is predicted based on the detection result to generate an SOH prediction result.

2. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 1, characterized in that: The power battery SOH prediction based on the detection result and the generation of the SOH prediction result include: when the detection result indicates that there is a conflict among multiple dynamically changing parameters contained in the data unit, determining the data unit as an abnormal unit; obtaining the conflict parameters of the abnormal unit as a correction object, and determining the multi-source data priority corresponding to the conflict parameters; and performing parameter fusion on the data unit based on the multi-source data priority to generate the SOH prediction result.

3. The method for predicting SOH of a power battery based on internal resistance-SOC calibration according to claim 2, characterized in that: The multi-source data includes first monitoring device data and second monitoring device data, the multi-source data priority of the first monitoring device data is a first priority, and the multi-source data priority of the second monitoring device data is a second priority; performing parameter fusion on the data units based on the multi-source data priority to generate an SOH prediction result includes: adjusting a prediction input set corresponding to the second monitoring device data according to an order of the first priority and the second priority; extracting, based on the calibrated prediction map, a dynamic change parameter of the second monitoring device data in the adjusted prediction input set; The dynamically changing parameters are combined with the calibration prediction map to perform parameter fusion to generate an SOH prediction result.

4. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 1, characterized in that: The step of calculating, for each data unit, a prediction feature index of the data unit based on the calibration feature parameters, and generating a prediction input set for the data unit according to a plurality of the prediction feature indexes includes: For each data unit, determining a plurality of adjacent units adjacent to the running state of the data unit; Calculating the correlation between the calibration characteristic parameter of each adjacent cell and the physical property of the battery, and determining a first characteristic cell with the highest correlation from the adjacent cells based on a plurality of the correlation degrees; Based on the first characteristic unit, determining a plurality of extension units adjacent to the characteristic unit, calculating a correlation between a calibration characteristic parameter of each extension unit and a physical property of the battery, and determining a next characteristic unit with the highest correlation from the extension units based on the plurality of correlations; Repeating the steps of determining a plurality of extension units adjacent to the characteristic unit, calculating a correlation between the calibration characteristic parameter of each extension unit and the battery physical property, and determining a next characteristic unit with the highest correlation from the extension units based on the plurality of correlations, until all data units are traversed, to obtain a set of predicted characteristic indicators for the data units; The prediction input set is generated based on the prediction feature indicator set.

5. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 4, characterized in that: Calculating the correlation between the calibration characteristic parameter of each adjacent cell and the physical property of the battery includes: respectively calculating a first correlation between the operating state attribute of each of the adjacent cells and the battery physical attribute, and a second correlation between the environmental attribute of each of the adjacent cells and the battery physical attribute; For each of the adjacent units, a weighted sum of the first degree of association and the second degree of association is used as the degree of association of the adjacent unit.

6. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 4, characterized in that: The determining, based on the plurality of correlation degrees, a first feature unit having the highest correlation degree from the adjacent units comprises: storing a plurality of association degrees corresponding to a plurality of adjacent units into a feature candidate list, and sorting the association degrees in the feature candidate list in descending order to obtain a sorting result; Based on the ranking result, the adjacent unit corresponding to the highest correlation degree is determined as the first feature unit.

7. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 1, characterized in that: After performing power battery SOH prediction based on the detection result and generating an SOH prediction result, the method further includes: After the SOH prediction result is generated, for each data unit, the attribute difference value between the data unit and the adjacent units is monitored in real time; When the attribute difference value exceeds a preset threshold, the adjacent unit is used as an abnormal reference unit, and the dynamic parameter correction of the data unit is performed based on the abnormal reference unit to generate an updated SOH prediction result.

8. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 1, characterized in that: The training process of the pre-trained SOH prediction model includes: Acquire a historical power battery data set and extract a mapping relationship between calibration characteristic parameters and dynamic change parameters in the data set; Constructing an initial SOH prediction model based on the mapping relationship, and iteratively optimizing the model using a cross-validation method; When the error between the predicted parameters output by the model and the measured parameters is lower than a preset threshold, it is determined that the model training is completed.

9. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 1, characterized in that: The internal resistance-SOC calibration model also includes a three-dimensional state layer of battery temperature distribution, charge and discharge rate distribution, and battery aging degree. When generating the calibration prediction map, the dynamically changing parameters are superimposed on the three-dimensional state layer for visual expression.

10. The method for predicting power battery SOH based on internal resistance-SOC calibration according to claim 9, characterized in that: The method for generating the three-dimensional state layer includes: Collect temperature data, charge and discharge rate data, and aging degree data of power batteries; Performing interpolation processing on the temperature data to generate a battery temperature distribution layer, performing classification coding on the charge and discharge rate data to generate a charge and discharge rate distribution layer, and performing spatial interpolation on the aging degree data to generate a battery aging degree layer; The battery temperature distribution layer, the charge and discharge rate distribution layer, and the battery aging degree layer are spatially overlaid and analyzed to form the three-dimensional state layer.

Images