Systems and methods for state of charge estimation

Abstract

The present disclosure provides a method for estimating state of charge in a lithium iron phosphate battery. The method includes initiating a relaxation period during which no current flows, measuring voltage features during relaxation at predetermined intervals, and extracting a feature vector from measured voltage features and historical current data preceding relaxation. The method applies a machine learning model to the feature vector to predict state of charge, wherein the model comprises three sequential sub-models where a first sub-model predicts voltage difference between measured voltage and equilibrium open circuit voltage, a second sub-model predicts state of charge difference based on outputs from the first sub-model, and a third sub-model generates final predictions through model fusion.

Description

SYSTEMS AND METHODS FOR STATE OF CHARGE ESTIMATIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The current application claims the benefit of and priority under 35 U.S.C. § 119(e) to U.S. Provisional Patent Application No. 63 / 729,261 entitled “Synergistic Approach Using Coulomb Counting Reset, Machine Learning, and Relaxation” filed December s, 2024. The disclosure of U.S. Provisional Patent Application No. 63 / 729,261 is hereby incorporated by reference in its entirety for all purposes.FIELD OF THE INVENTION

[0002] The present disclosure relates to battery management systems, and more particularly to machine learning-based methods for estimating state of charge (SOC) in lithium iron phosphate (LFP) batteries using voltage relaxation data during short-term rest periods.BACKGROUND

[0003] Lithium-ion batteries have become the dominant energy storage technology across consumer electronics, electric vehicles, and grid-scale applications due to their high energy density, decreasing costs, and extended cycle life. Among the various lithium-ion chemistries, lithium iron phosphate (LFP) batteries have become increasingly common based on their enhanced safety characteristics, thermal stability, reduced risk of thermal runaway, and freedom from cobalt and nickel materials. Additionally, LFP batteries demonstrate longer cycle life compared to other lithium-ion chemistries and benefit from abundant phosphate resources that support sustainable supply chains.

[0004] Battery management systems (BMS) play a fundamental role in ensuring safe and reliable operation of battery systems during real-time use. One of the primary functions of a BMS is state-of-charge (SOC) estimation, which provides information about the remaining energy capacity of the battery. Accurate SOC estimation can enable optimization of battery performance, extension of battery life, and prevention of potential safety issues.SUMMARY OF THE INVENTION

[0005] Systems and methods for charge estimation in accordance with embodiments of the invention are illustrated. One embodiment includes a method for estimating state of charge (SOC) in a battery. The method includes steps for initiating a relaxation period in the battery during which no current flows through the battery, measuring battery data during the relaxation period at predetermined time intervals, wherein the battery data includes measured voltage features and historical current data, extracting a feature vector from the battery data, predicting the SOC of the battery by applying a machine learning model to the extracted feature vector, wherein the machine learning model includes three sub-models where a first sub-model predicts voltage difference between measured voltage and equilibrium open circuit voltage (OCV), a second sub-model predicts SOC difference between measured SOC and a SOC from Coulomb counting, and a third sub-model predicts a final SOC, and generating an output includes the predicted state of charge.

[0006] In a further embodiment, the relaxation period is a short-term rest period of less than 30 minutes.

[0007] In still another embodiment, the relaxation period ranges from 30 to 600 seconds.

[0008] In a still further embodiment, the voltage features are measured at intervals of 30 seconds during the relaxation period.

[0009] In yet another embodiment, the feature vector includes at least one selected from the group consisting of measured voltage values, initial voltage measurements, mean current values, loading indications, resistance, environmental temperature measurements, and temperature difference calculations.

[0010] In a yet further embodiment, the historical current data includes at least one selected from the group consisting of a mean current value characterizing average current demand during an operational period preceding the relaxation period, a current rate indicating magnitude and direction of current flow, and loading indications providing information about operational state prior to the relaxation period.

[0011] In another additional embodiment, the machine learning model utilizes random forests.

[0012] In a further additional embodiment, the first sub-model generates a first OCV calculated as a sum of measured voltage and predicted voltage difference, and generates a first SOC obtained through inversion of an OCV to SOC relationship using the first OCV.

[0013] In another embodiment again, the second sub-model generates a second SOC calculated as a sum of measured SOC and predicted SOC difference, and generates a second OCV derived from the OCV to SOC relationship using the second SOC.

[0014] In a further embodiment again, the third sub-model estimates the final SOC based on the feature vector, the first and second OCVs, and the first and second SOCs.

[0015] In still yet another embodiment, generating the output includes transmitting control signals to a battery management system.

[0016] In a still yet further embodiment, generating the output includes providing a visualization of the final SOC to a display.

[0017] In still another additional embodiment, the method further includes steps for updating the machine learning model based on new voltage relaxation data collected during battery operation.

[0018] One embodiment includes a computer-readable medium includes instructions which, when executed by a computer, cause the computer to carry out the methods for charge estimation.

[0019] One embodiment includes a system that includes means for carrying out methods for charge estimation.

[0020] One embodiment includes a battery management system for a lithium iron phosphate battery, comprising a voltage measurement circuit configured to measure voltage of the lithium iron phosphate battery during a rest period, a processor configured to detect the rest period of the battery, collect voltage relaxation data during the rest period, extract features from the voltage relaxation data, and apply a trained machine learning model to estimate state of charge, and a memory storing the trained machine learning model, wherein the trained machine learning model includes three sequential sub-models that process extracted features to generate state of charge predictions.

[0021] In a still further additional embodiment, the voltage measurement circuit has a resolution of at least 1 millivolt.

[0022] In still another embodiment again, the processor is configured to detect the rest period by monitoring current measurements and identifying when current flow drops to zero or below a predetermined threshold value.

[0023] In a still further embodiment again, the processor is configured to collect the voltage relaxation data at predetermined intervals during the rest period ranging from 30 to 600 seconds.

[0024] In yet another additional embodiment, the processor is configured to extract the features includes voltage differences between measured values and reference open circuit voltages, temperature variations during the rest period, and historical current data preceding the rest period.

[0025] In a yet further additional embodiment, the three sequential sub-models includes a first sub-model configured to predict voltage difference between measured voltage and equilibrium open circuit voltage, a second sub-model configured to predict state of charge difference based on outputs from the first sub-model, and a third submodel configured to generate final state of charge predictions through model fusion.

[0026] Additional embodiments and features are set forth in part in the description that follows, and in part will become apparent to those skilled in the art upon examination of the specification or may be learned by the practice of the invention. A further understanding of the nature and advantages of the present invention may be realized by reference to the remaining portions of the specification and the drawings, which forms a part of this disclosure.BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The description and claims will be more fully understood with reference to the following figures and data graphs, which are presented as exemplary embodiments of the invention and should not be construed as a complete recitation of the scope of the invention.

[0028] FIG. 1 illustrates a flowchart for a process for state of charge estimation in a battery in accordance with an embodiment of the invention.

[0029] FIG. 2 depicts a diagram illustrating feature extraction from voltage measurements during a short-term rest period in accordance with an embodiment of the invention.

[0030] FIG. 3 illustrates a flowchart for a method for state of charge estimation using cascaded prediction in accordance with an embodiment of the invention.

[0031] FIG. 4 illustrates a flowchart for a method for state of charge estimation using a machine learning pipeline in accordance with an embodiment of the invention.

[0032] FIG. 5 illustrates a machine learning pipeline for battery state-of-charge estimation with training and testing processes in accordance with an embodiment of the invention.

[0033] FIG. 6 illustrates a network diagram showing a distributed system architecture for battery state of charge estimation in accordance with an embodiment of the invention.

[0034] FIG. 7 illustrates a block diagram of an SOC prediction element in accordance with an embodiment of the invention.DETAILED DESCRIPTION

[0035] The following description sets forth exemplary aspects of the present disclosure. It should be recognized, however, that such description is not intended as a limitation on the scope of the present disclosure. Rather, the description also encompasses combinations and modifications to those exemplary aspects described herein.

[0036] State-of-charge (SOC) estimation for lithium iron phosphate (LFP) batteries presents significant challenges due to their unique electrochemical characteristics. LFP batteries exhibit hysteresis effects, where voltage differences occur between charging and discharging processes at similar SOC levels. Additionally, LFP batteries demonstrate path-dependent behavior, where the battery's voltage response depends not only on the current SOC but also on the historical loading conditions.

[0037] Another characteristic that complicates SOC estimation in LFP batteries is their flat open-circuit voltage (OCV) profile. Unlike other lithium-ion chemistries where OCV varies monotonically with SOC, LFP batteries exhibit a relatively flat OCV curvetypically across an SOC range of 30-80%, which can extend to 20-90% in some cases. This flatness reduces the sensitivity of voltage measurements to SOC changes, making traditional OCV-based estimation methods less effective.

[0038] Current SOC estimation approaches include model-based methods that utilize equivalent circuit models or physics-based electrochemical models combined with filtering techniques such as extended Kalman filters or unscented Kalman filters. Other approaches include hybrid approaches that integrate an electrochemical model describing the two-phase transition operation in the positive electrode, with a machine learning component designed to capture the hysteresis effects and path-dependent dynamics of the battery. The machine learning component can be trained on data that includes current profiles collected from various electric vehicle (EV) driving scenarios and electrochemical states derived from the physics-based model. However, the flat OCV characteristics of LFP batteries reduce the effectiveness of these traditional estimation methods.

[0039] Another common approach is the Coulomb counting method, which calculates SOC by integrating current over time, expressed as SOC(t) = SOC(t0) - While computationally efficient and widely used in BMS applications,Coulomb counting can suffer from various sources of error, such as drift over time due to cumulative integration errors, reliance on precise current measurements, incorrect initialization, and inaccurate capacity estimation.

[0040] The initialization of SOC in Coulomb counting typically relies on OCV measurements after prolonged rest periods to determine the starting SOC value through OCV-SOC lookup table inversion. However, the flat OCV characteristics of LFP batteries mean that even small voltage measurement errors can result in substantial SOC estimation errors. Furthermore, achieving true equilibrium voltage in LFP batteries may require extended rest periods, which are often impractical in real-world applications.A. SOC Estimation

[0041] To address these limitations, SOC estimation processes in accordance with certain embodiments of the invention leverage voltage and temperature time series data collected during short-term relaxation periods (e.g., 1 minute, 5 minutes, 30 minutes, etc.),combined with historical current data, to enhance SOC estimation accuracy without relying on prolonged rest periods or complex model-based methods. SOC estimation processes in accordance with many embodiments of the invention offer enhanced precision and reliability compared to conventional Coulomb counting techniques, providing model-free solutions that can be seamlessly integrated into existing battery management systems. Validation results across diverse operating conditions, including various temperatures, current rates, cycling scenarios, and resting periods, indicate significant improvements in estimation accuracy using relaxation voltage data recorded at intervals as short as one minute.

[0042] An example of a process for estimating state of charge (SOC) in a LFP battery in accordance with an embodiment of the invention is illustrated in FIG. 1. In various embodiments, SOC estimation processes address the unique challenges associated with LFP batteries by utilizing short-term relaxation periods, low sampling rates (e.g., 1 / 30 Hz), and / or machine learning techniques to provide accurate SOC estimation.

[0043] Process 100 initiates (105) a relaxation period of the battery. Initiating a relaxation period in accordance with various embodiments of the invention involves identifying when the battery transitions from an active operational state to a state where no current flows through the battery. In many embodiments, relaxation periods can be detected when the current measurement drops to zero or below a predetermined threshold value (e.g., 1 minute, 5 minutes, 30 minutes, etc.). Relaxation periods in accordance with certain embodiments of the invention may occur during various operational scenarios, including (but not limited to) vehicle stops at traffic lights, parking periods, charging interruptions, and / or scheduled maintenance intervals.

[0044] Process 100 measures (110) battery data during the relaxation period. In certain embodiments, battery data includes (but is not limited to) historical current data, voltage data, and temperature data. Battery data in accordance with certain embodiments of the invention is time series data. In numerous embodiments, battery data includes initial voltage measurements taken at early points in the rest period (e.g., 10 seconds after current cessation) and subsequent voltage measurements taken at later intervals (e.g., every 30 seconds from 30 to 600 seconds) throughout the relaxation process, creating atime series of voltage measurements that characterizes the relaxation behavior. In many embodiments, battery data may be captured at a low sampling rate (e g., 1 / 30 Hz), minimizing memory usage requirements in battery management systems while maintaining sufficient data resolution for accurate SOC estimation.

[0045] Process 100 extracts (115) features from the battery data. Extracting features in accordance with various embodiments of the invention involves deriving meaningful parameters from the measured voltage and temperature time series data that characterize the battery's electrochemical state. Features in accordance with certain embodiments of the invention can include (but are not limited to) voltage differences between measured values and reference open circuit voltages, temperature variations during the rest period, and / or historical current data preceding the rest period. In many embodiments, feature vectors may incorporate multiple types of data, including direct features from voltage measurements, intermediate features calculated from voltage-SOC relationships, and contextual features such as environmental temperature and resistance values. Feature extraction in accordance with some embodiments of the invention is described in greater detail below with reference to Fig. 2.

[0046] Process 100 applies (120) a machine learning model to the extracted features to estimate the state of charge. Applying a machine learning model in accordance with numerous embodiments of the invention involves processing the extracted feature vectors through trained sub-models to predict the battery's SOC. Machine learning models in accordance with various embodiments of the invention can include (but are not limited to) random forests, neural networks, support vector machines, and / or ensemble methods. In certain embodiments, machine learning models may include multiple sub-models that work in sequence, where each sub-model refines the SOC estimation based on outputs from previous sub-models and / or the original feature vector. Machine learning models in accordance with some embodiments of the invention are described in greater detail below with reference to FIGs. 3-5.

[0047] Process 100 generates (125) an output based on the predicted SOC. Generating an output in accordance with various embodiments of the invention includes providing the estimated SOC value for use in battery management systems and related applications. Outputs in accordance with certain embodiments of the invention mayinclude the predicted SOC value, confidence intervals, predicted open circuit voltage values, data visualizations, command signals, and / or diagnostic information about the estimation process. In many embodiments, the output may be transmitted to battery management systems, vehicle control units, energy storage system controllers, and / or remote monitoring systems.

[0048] SOC estimation processes in accordance with various embodiments of the invention can utilize machine learning techniques to specifically addresses the flat open circuit voltage characteristics and path-dependent behavior inherent to LFP batteries. Additionally, the path-dependent behavior of LFP batteries, where the voltage response depends on the battery's operational history, can be addressed through the incorporation of historical current data and relaxation dynamics into the feature extraction process, enabling the machine learning model to account for hysteresis effects and previous loading conditions when estimating SOC.

[0049] While specific processes for SOC estimation are described above, any of a variety of processes can be utilized to estimate SOC as appropriate to the requirements of specific applications. In certain embodiments, steps may be executed or performed in any order or sequence not limited to the order and sequence shown and described. In a number of embodiments, some of the above steps may be executed or performed substantially simultaneously where appropriate or in parallel to reduce latency and processing times. In some embodiments, one or more of the above steps may be omitted.

[0050] Although many of the examples described herein estimating SOC, one skilled in the art will recognize that similar systems and methods can be used in a variety of applications, including (but not limited to) estimating battery state of health (SOH), without departing from this invention. For instance, in addition to features from the relaxation period, peak shifting values derived from Incremental Capacity Analysis (ICA) or Differential Voltage Analysis (DVA) conducted at different aging cycles can be incorporated as input features. Machine learning models as in accordance with numerous embodiments of the invention can be trained and used to output both the battery’s SOC and capacity.B. Feature Extraction

[0051] In many embodiments, feature extraction processes generate a comprehensive feature vector F that multiple parameters characterizing the battery's electrochemical state during relaxation periods. Feature vectors in accordance with certain embodiments of the invention include voltage relaxation data, temperature information, and historical current characteristics. Feature vectors in accordance with various embodiments of the invention may include (but are not limited to) time parameters corresponding to the rest time t, the measured voltage Vmeacollected at various intervals, measured SOC SOCmeaderived from voltage-SOC relationships, voltage measurements such as the initial voltage 710at 10 seconds, mean current Imfrom historical data, resistance R, current rate information, loading indications Iflag, environmental temperature measurements Tenv, and / or temperature difference dT features. Temperature difference features may characterize thermal variations during the relaxation period, providing additional context for the electrochemical state of the battery.

[0052] An example of feature extraction in accordance with an embodiment of the invention is illustrated in FIG. 2. Feature extraction processes in accordance with various embodiments of the invention utilize voltage and temperature measurements collected during short-term relaxation periods, combined with historical current data, to create comprehensive feature vectors for machine learning-based SOC estimation.

[0053] The first chart of this figure illustrates changes in voltage during operation as well as through a short-term rest, along with direct and indirect features that can be extracted from the voltage data. Extracted features from the voltage data may include direct features from the battery data, such as (but not limited to) an initial voltage (e.g., 710) measured at some time (e.g., 10 seconds) after load disconnection to capture the early voltage response following the cessation of current flow, a measured voltage Vmeacollected at predetermined intervals throughout the relaxation period to characterize the voltage trajectory as the battery approaches its equilibrium state, and / or a rest time t that represents the duration of the relaxation period during which voltage measurements are collected, providing temporal context for the voltage recovery process. In numerous embodiments, extracted features also include intermediate features from the voltagedata, such as (but not limited to) the measured SOC, SOCmea, which is extracted based on the measured voltage Vmea.

[0054] The second chart of this figure illustrates changes in current during operation as well as through a short-term rest, along with direct and indirect features that can be extracted from the current data. Extracted features in accordance with numerous embodiments of the invention include historical current data to enhance SOC prediction accuracy by providing context about the battery's operational history prior to the relaxation period. Historical current features in accordance with certain embodiments of the invention include (but are not limited to) direct features (e.g., mean current value and / or loading indications), as well as intermediate features (e.g., calculated resistance). Mean current values may characterize the average current demand during the operational period preceding the rest time t. In several embodiments, utilizing mean current values can provide historical current information while reducing the impact of lower sampling rates. Loading indications may provide binary or categorical information about the historical current condition, such as whether the current is constant or dynamic. Calculated resistance is based on the difference between voltage measurements during the relaxation period (e.g., V10and V30) and the mean current value. The voltage difference provides information about the magnitude of voltage recovery during the relaxation process, characterizing the extent of polarization effects that dissipate as the battery approaches equilibrium.

[0055] The third chart of this figure illustrates changes in temperature during operation as well as through a short-term rest, along with direct features that can be extracted from the temperature data. In this example, the direct features include an ambient temperature and a temperature variation at the cell surface. Temperature data in accordance with various embodiments of the invention can provide insight to operating conditions, and can be used to estimate battery capacity Q.C. SOC Estimation Models

[0056] SOC estimation in accordance with a variety of embodiments of the invention utilizes machine learning models. Machine learning models in accordance with a variety of embodiments of the invention may include one or more sub-models. In avariety of embodiments, machine learning models (or their sub-models) may include various model architectures, such as (but not limited to) neural networks, support vector machines, random forests, and / or gradient boosting algorithms.

[0057] Random forests in accordance with certain embodiments of the invention provide advantages for SOC estimation applications due to their ensemble nature, which combines predictions from multiple decision trees to generate robust outputs that are less susceptible to overfitting and noise in the input data. Random forests in accordance with various embodiments of the invention can effectively handle non-linear relationships between voltage relaxation patterns, temperature variations, historical current data, and / or actual SOC values without requiring explicit mathematical models of the underlying electrochemical processes. The tree-based structure of random forests in accordance with numerous embodiments of the invention enables interpretable feature importance rankings that can guide feature selection processes and provide insights into the relative contributions of different input parameters to SOC prediction accuracy.

[0058] In a variety of embodiments, machine learning models include three submodels. In various embodiments, the sub-models can include (but are not limited to) a voltage difference sub-model that predicts voltage difference, a SOC difference submodel that predicts SOC difference, and a final estimation sub-model that generates final SOC and / or OCV predictions.

[0059] Voltage difference sub-models in accordance with many embodiments of the invention address voltage-related uncertainties by predicting the difference between measured voltage values and equilibrium open circuit voltage values, accounting for polarization effects that persist during short-term relaxation periods.

[0060] SOC difference sub-models in accordance with numerous embodiments of the invention process the refined voltage and SOC estimates from the first sub-model along with the original feature vector to predict SOC difference values that correct for remaining estimation errors. SOC difference predictions in accordance with certain embodiments of the invention may account for path-dependent behavior and hysteresis effects that influence the relationship between voltage measurements and actual state of charge values. The second sub-model generates SOC2calculated as SOC2= SOCmea+SOCdiff, where SOCmearepresents the measured SOC obtained from voltage-SOC inversion and SOCdiffrepresents the predicted correction factor.

[0061] Final estimation sub-models in accordance with various embodiments of the invention receives inputs from previous sub-models, including the feature vector, SOCi, OCVi, SOC2, and OCV2, to perform final model fusion and generate the most accurate SOC and OCV predictions. Model fusion in accordance with certain embodiments of the invention combines the complementary information from multiple prediction stages to produce final estimates that are more accurate than any individual sub-model output. Final estimation sub-models in accordance with many embodiments of the invention may utilize ensemble learning techniques to weight the contributions of different input sources based on their reliability and relevance to the final prediction task.

[0062] An example of SOC estimation using machine learning models in accordance with an embodiment of the invention is illustrated in FIG. 3. SOC estimation in accordance with various embodiments of the invention utilize a sequential approach where multiple prediction steps build upon the results of preceding steps to generate increasingly refined SOC estimates. SOC estimation in accordance with certain embodiments of the invention address the complex electrochemical behavior of LFP batteries by decomposing the SOC estimation problem into multiple stages, each focusing on specific aspects of the battery's state. Examples of machine learning models for SOC estimation are described in greater detail below with reference to Fig. 4.

[0063] Process 300 predicts (305) a preliminary open circuit voltage (OCV) based on extracted features from the relaxation period. Predicting a preliminary open circuit voltage in accordance with numerous embodiments of the invention involves processing the feature vector through a voltage difference sub-model to estimate the voltage difference between the measured voltage and the equilibrium open circuit voltage. Predicting a preliminary OCV in accordance with several embodiments of the invention can be computed based on the estimated voltage difference from a voltage difference sub-model. A preliminary OCV may be calculated as OCVt= Vdiff+ Vmea, where Vmearepresents the measured voltage during the relaxation period and Vdiffrepresents the predicted voltage difference between the measured voltage and the equilibrium opencircuit voltage. In numerous embodiments, predicting a preliminary OCV can include computing a preliminary SOC SOCi through inversion of the SOC - OCVGITTrelationship using the predicted OCVi value. In many embodiments, SOCi may be obtained by applying the predicted OCVi value to a lookup table or mathematical relationship that correlates open circuit voltage values with corresponding state of charge levels.

[0064] Process 300 predicts (310) an initial state of charge (SOC) based on the predicted preliminary OCV. Predicting an initial SOC SOC2in accordance with various embodiments of the invention includes processing the feature vector along with OCVi and / or SOCi (from the voltage difference sub-model) with a SOC difference sub-model to predict SOC difference values SOCdiffthat further refine the state of charge estimation. SOC difference models in accordance with several embodiments of the invention analyze the relationship between voltage-based SOC estimates and actual SOC values to generate correction factors that account for path-dependent behavior and hysteresis effects characteristic of LFP batteries. In numerous embodiments, initial SOCs SOC2can be calculated as SOC2= SOCmea+ SOCdlff, where SOCmearepresents the measured SOC obtained from OCV-SOC map inversion using measured voltage values and SOCdiffrepresents the predicted correction factor. In a number of embodiments, predicting an initial SOC2includes generating a corresponding OCV OCV2derived from the SOC - OCVGITTrelationship using the predicted SOC2value.

[0065] Process 300 predicts (315) a final SOC based on the predicted open circuit voltage and the predicted state of charge from the previous sub-models. Predicting final state of charge in accordance with numerous embodiments of the invention involves utilizing a final estimation sub-model that combines outputs from both the voltage difference and SOC difference sub-models with the original feature vector to generate the most accurate SOC and OCV predictions.

[0066] While specific processes for SOC estimation are described above, any of a variety of processes can be utilized to estimate SOC as appropriate to the requirements of specific applications. In certain embodiments, steps may be executed or performed in any order or sequence not limited to the order and sequence shown and described. In a number of embodiments, some of the above steps may be executed or performedsubstantially simultaneously where appropriate or in parallel to reduce latency and processing times. In some embodiments, one or more of the above steps may be omitted.

[0067] An example of training a machine learning model for estimating SOC prediction in accordance with an embodiment of the invention is illustrated in FIG. 4. Training processes for machine learning models in accordance with various embodiments of the invention may utilize datasets collected across multiple temperature conditions, current rates, cycling scenarios, and resting periods to ensure robust performance across diverse operational environments encountered in practical battery management applications. Machine learning models in accordance with various embodiments of the invention utilize a systematic approach where multiple sub-models work sequentially to refine SOC predictions through iterative processing and model updating mechanisms.

[0068] Process 400 predicts (405) a voltage difference with a voltage difference sub-model. Predicting a voltage difference in accordance with numerous embodiments of the invention involves processing an extracted feature vector to estimate voltage differences between measured relaxation voltages and a true OCV values. In numerous embodiments, true OCV values are measured through lab tests.

[0069] Process 400 predicts (410) a SOC difference based on a predicted OCV with a SOC difference sub-model. Predicting a SOC difference in accordance with many embodiments of the invention involves processing the extracted feature vector and the predicted OCV to estimate SOC differences between a measured SOC and a true SOC. In certain embodiments, true SOC values are measured through lab tests. True SOC values in accordance with many embodiments of the invention are computed based on the SOC - OCVG1TTrelationship. The SOC - OCVGITTrelationship in accordance with certain embodiments of the invention represents the correlation between state of charge values and open circuit voltage measurements obtained through Galvanostatic Intermittent Titration Technique testing under laboratory conditions.

[0070] Process 400 predicts (415) final SOC based on the predicted OCV and predicted SOC with a third sub-model. Predicting final SOC with a third sub-model in accordance with numerous embodiments of the invention involves processing the feature vector, SOCi, OCVi from the voltage difference sub-model, and SOC2, OCV2from theSOC difference sub-model to produce final SOC and OCV predictions through ensemble learning techniques.

[0071] Process 400 updates (420) the sub-models based on predictions and true SOC and OCV values. Updating sub-models in accordance with various embodiments of the invention involves utilizing feedback mechanisms where actual SOC and OCV measurements are compared with the predicted voltage difference, SOC difference, final SOC, and / or final OCV to refine each of the sub-models. Model updating processes in accordance with certain embodiments of the invention may incorporate new training data collected during battery operation to continuously improve prediction accuracy and adapt to changing battery characteristics over time.

[0072] Training processes in accordance with numerous embodiments of the invention utilize training labels derived from laboratory testing and operational measurements to establish ground truth values for machine learning algorithm development. Training labels in accordance with certain embodiments of the invention include OCVGITTvalues obtained from laboratory tests and SOC values calculated from Coulomb counting after charge or discharge operations. OCVGITTvalues in accordance with various embodiments of the invention represent equilibrium open circuit voltage measurements obtained through Galvanostatic Intermittent Titration Technique (GITT) testing, providing accurate reference values for voltage-related predictions. SOC values calculated from Coulomb counting in accordance with many embodiments of the invention provide accurate state of charge measurements obtained through current integration during controlled charge and discharge operations under laboratory conditions.

[0073] While specific processes for training SOC estimation models are described above, any of a variety of processes can be utilized to train SOC estimation models as appropriate to the requirements of specific applications. In certain embodiments, steps may be executed or performed in any order or sequence not limited to the order and sequence shown and described. In a number of embodiments, some of the above steps may be executed or performed substantially simultaneously where appropriate or in parallel to reduce latency and processing times. In some embodiments, one or more of the above steps may be omitted.

[0074] An example of a machine learning pipeline for battery state-of-charge estimation in accordance with an embodiment of the invention is illustrated in FIG. 5. Machine learning pipelines in accordance with various embodiments of the invention utilize a comprehensive training process that incorporates three sequential sub-models working in coordination to generate accurate SOC and OCV predictions through systematic feature processing and model fusion techniques. The training process and testing process are described in greater detail above with reference to FIGs. 3 and 4.

[0075] Machine learning models in accordance with many embodiments of the invention can be updated based on new voltage relaxation data collected during battery operation to continuously improve prediction accuracy and adapt to changing battery characteristics over time. Model updating processes in accordance with certain embodiments of the invention may incorporate incremental learning techniques that allow machine learning algorithms to assimilate new training data without requiring complete retraining of the entire model. Battery management systems in accordance with various embodiments of the invention may implement processors configured to update machine learning models based on new voltage relaxation data, enabling adaptive SOC estimation capabilities that maintain accuracy as battery conditions evolve through aging and usage patterns.D. Systems for SOC Estimation1. SOC Estimation System

[0076] An example of an SOC estimation system that estimates SOC in accordance with an embodiment of the invention is illustrated in Figure 6. Network 600 includes a communications network 660. The communications network 660 is a network such as the Internet that allows devices connected to the network 660 to communicate with other connected devices. Server systems 610, 640, and 670 are connected to the network 660. Each of the server systems 610, 640, and 670 is a group of one or more servers communicatively connected to one another via internal networks that execute processes that provide cloud services to users over the network 660. One skilled in the art will recognize that an SOC estimation system may exclude certain components and / orinclude other components that are omitted for brevity without departing from this invention.

[0077] For purposes of this discussion, cloud services are one or more applications that are executed by one or more server systems to provide data and / or executable applications to devices over a network. The server systems 610, 640, and 670 are shown each having three servers in the internal network. However, the server systems 610, 640 and 670 may include any number of servers and any additional number of server systems may be connected to the network 660 to provide cloud services. In accordance with various embodiments of this invention, an SOC estimation system that uses systems and methods that estimates SOC in accordance with an embodiment of the invention may be provided by a process being executed on a single server system and / or a group of server systems communicating over network 660.

[0078] Users may use personal devices 680 and 620 that connect to the network 660 to perform processes that estimates SOC in accordance with various embodiments of the invention. In the shown embodiment, the personal devices 680 are shown as desktop computers that are connected via a conventional “wired” connection to the network 660. However, the personal device 680 may be a desktop computer, a laptop computer, a smart television, an entertainment gaming console, or any other device that connects to the network 660 via a “wired” connection. The mobile device 620 connects to network 660 using a wireless connection. A wireless connection is a connection that uses Radio Frequency (RF) signals, Infrared signals, or any other form of wireless signaling to connect to the network 660. In the example of this figure, the mobile device 620 is a mobile telephone. However, mobile device 620 may be a mobile phone, Personal Digital Assistant (PDA), a tablet, a smartphone, or any other type of device that connects to network 660 via wireless connection without departing from this invention.

[0079] As can readily be appreciated the specific computing system used to estimate SOC is largely dependent upon the requirements of a given application and should not be considered as limited to any specific computing system(s) implementation.2. SOC Estimation Element

[0080] An example of an SOC estimation element that executes instructions to perform processes that estimates SOC in accordance with an embodiment of the invention is illustrated in Figure 7. SOC estimation elements in accordance with many embodiments of the invention can include (but are not limited to) one or more of mobile devices, cameras, and / or computers. SOC estimation element 700 includes processor 705, peripherals 710, network interface 715, and memory 720. One skilled in the art will recognize that an SOC estimation element may exclude certain components and / or include other components that are omitted for brevity without departing from this invention.

[0081] The processor 705 can include (but is not limited to) a processor, microprocessor, controller, or a combination of processors, microprocessor, and / or controllers that performs instructions stored in the memory 720 to manipulate data stored in the memory. Processor instructions can configure the processor 705 to perform processes in accordance with certain embodiments of the invention. In various embodiments, processor instructions can be stored on a non-transitory machine readable medium.

[0082] Peripherals 710 can include any of a variety of components for capturing data, such as (but not limited to) cameras, displays, measurement elements (e.g., for temperature, current, and / or voltage), and / or other sensors. In a variety of embodiments, peripherals can be used to gather inputs and / or provide outputs. SOC estimation element 700 can utilize network interface 715 to transmit and receive data over a network based upon the instructions performed by processor 705. Peripherals and / or network interfaces in accordance with many embodiments of the invention can be used to gather inputs that can be used to estimate SOC.

[0083] Memory 720 includes an SOC estimation application 725, model data 730, and training data 735. SOC estimation applications in accordance with several embodiments of the invention can be used to estimate SOC and / or train models as described throughout this specification.

[0084] Measurement data 730 in accordance with many embodiments of the invention stores voltage, current, temperature, and other sensor measurements collected from battery systems during operational and rest periods. Measurement data 730 in accordance with certain embodiments of the invention includes time-series voltage measurements corresponding to the initial voltage and the measured voltage collected during the rest time, along with associated temperature and current measurements.

[0085] Model data 735 in accordance with numerous embodiments of the invention contains machine learning models and parameters used for SOC estimation, including trained random forests, feature extraction parameters, and SOC-OCV lookup tables. The model data 735 in accordance with certain embodiments of the invention stores trained sub-models for predicting voltage differences, SOC corrections, and final state of charge values based on extracted feature vectors. In several embodiments, model data can store various parameters and / or weights for various models that can be used for various processes as described in this specification. Model data in accordance with many embodiments of the invention can be updated through training on battery data captured on an SOC estimation element or can be trained remotely and updated at an SOC estimation element.

[0086] Although a specific example of an SOC estimation element 700 is illustrated in this figure, any of a variety of SOC estimation elements can be utilized to perform processes for estimating SOC similar to those described herein as appropriate to the requirements of specific applications in accordance with embodiments of the invention.

[0087] Although specific methods of estimating SOC are discussed above, many different methods of estimating SOC can be implemented in accordance with many different embodiments of the invention. It is therefore to be understood that the present invention may be practiced in ways other than specifically described, without departing from the scope and spirit of the present invention. Thus, embodiments of the present invention should be considered in all respects as illustrative and not restrictive. Accordingly, the scope of the invention should be determined not by the embodiments illustrated, but by the appended claims and their equivalents.

Claims

WHAT IS CLAIMED IS:1 . A method for estimating state of charge (SOC) in a battery, the method comprising: initiating a relaxation period in the battery during which no current flows through the battery; measuring battery data during the relaxation period at predetermined time intervals, wherein the battery data comprises measured voltage features and historical current data; extracting a feature vector from the battery data; predicting the SOC of the battery by applying a machine learning model to the extracted feature vector, wherein the machine learning model comprises three submodels where a first sub-model predicts voltage difference between measured voltage and equilibrium open circuit voltage (OCV), a second sub-model predicts SOC difference between measured SOC and a SOC from Coulomb counting, and a third sub-model predicts a final SOC; and generating an output comprising the predicted state of charge.

2. The method of claim 1 , wherein the relaxation period is a short-term rest period of less than 30 minutes.

3. The method of claim 2, wherein the relaxation period ranges from 30 to 600 seconds.

4. The method of claim 1 , wherein the voltage features are measured at intervals of 30 seconds during the relaxation period.

5. The method of claim 1 , wherein the feature vector comprises at least one selected from the group consisting of: measured voltage values; initial voltage measurements; mean current values; loading indications; resistance; environmental temperature measurements; and temperature difference calculations.

6. The method of claim 5, wherein the historical current data comprises at least one selected from the group consisting of : a mean current value characterizing average current demand during an operational period preceding the relaxation period; a current rate indicating magnitude and direction of current flow; and loading indications providing information about operational state prior to the relaxation period.

7. The method of claim 1 , wherein the machine learning model utilizes random forests.

8. The method of claim 7, wherein the first sub-model generates a first OCV calculated as a sum of measured voltage and predicted voltage difference, and generates a first SOC obtained through inversion of an OCV to SOC relationship using the first OCV.

9. The method of claim 8, wherein the second sub-model generates a second SOC calculated as a sum of measured SOC and predicted SOC difference, and generates a second OCV derived from the OCV to SOC relationship using the second SOC.

10. The method of claim 9, wherein the third sub-model estimates the final SOC based on the feature vector, the first and second OCVs, and the first and second SOCs.

11. The method of claim 1 , wherein generating the output comprises transmitting control signals to a battery management system.

12. The method of claim 1 , wherein generating the output comprises providing a visualization of the final SOC to a display.

13. The method of claim 1 , further comprising updating the machine learning model based on new voltage relaxation data collected during battery operation.

14. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of claims 1 to 13.

15. A system comprising means for carrying out the method of claims 1 to 13.

16. A battery management system for a lithium iron phosphate battery, comprising: a voltage measurement circuit configured to measure voltage of the lithium iron phosphate battery during a rest period; a processor configured to detect the rest period of the battery, collect voltage relaxation data during the rest period, extract features from the voltage relaxation data, and apply a trained machine learning model to estimate state of charge; and a memory storing the trained machine learning model, wherein the trained machine learning model comprises three sequential sub-models that process extracted features to generate state of charge predictions.

17. The battery management system of claim 16, wherein the voltage measurement circuit has a resolution of at least 1 millivolt.

18. The battery management system of claim 16, wherein the processor is configured to detect the rest period by monitoring current measurements and identifying when current flow drops to zero or below a predetermined threshold value.

19. The battery management system of claim 18, wherein the processor is configured to collect the voltage relaxation data at predetermined intervals during the rest period ranging from 30 to 600 seconds.

20. The battery management system of claim 19, wherein the processor is configured to extract the features comprising voltage differences between measured values and reference open circuit voltages, temperature variations during the rest period, and historical current data preceding the rest period.

21. The battery management system of claim 16, wherein the three sequential submodels comprise a first sub-model configured to predict voltage difference between measured voltage and equilibrium open circuit voltage, a second sub-model configured to predict state of charge difference based on outputs from the first sub-model, and a third sub-model configured to generate final state of charge predictions through model fusion.

Images