DESC:SYSTEM AND METHOD FOR DETERMINING BATTERY CAPACITY
CROSS REFERENCE TO RELATED APPLICTIONS
The present application claims priority from Indian Provisional Patent Application No. 202421014547 filed on 28-02-2024, the entirety of which is incorporated herein by a reference.
TECHNICAL FIELD
The present disclosure generally relates to battery capacity forecasting. Further, the present disclosure particularly relates to a system and method to predict battery capacity using hybrid forecasting models based on filtered datasets.
BACKGROUND
Battery packs are essential components of energy storage systems used in various applications, including electric vehicles. Such battery packs typically comprise multiple battery cells connected to provide higher energy capacity. Lithium-ion cells are commonly used due to their high energy density, fast charging capability and relatively long lifecycle. Such cells store energy chemically and release the stored chemical energy as electrical energy during operation.
Over time, the capacity of battery cells to store and deliver energy degrades due to factors such as calendar ageing, cycle ageing, side reactions, mechanical stress and environmental conditions like temperature variations. Such degradation can affect the efficiency, reliability and safety of the systems utilizing the battery cells. Further, parameters influencing the degradation include charge-discharge cycles, discharge end voltage and operating conditions.
Techniques to predict the remaining capacity of battery cells have traditionally relied on statistical models that use historical data trends. While effective for basic patterns, such models struggle to address the complex interdependencies between variables affecting degradation. As a result, their predictive accuracy is limited, particularly for systems with intricate degradation behaviour.
To address such shortcomings, data-driven techniques involving machine learning and deep learning have been employed to predict battery health and capacity. Such techniques use datasets comprising charge capacity, discharge voltage, cycle numbers and other parameters to train models for predicting capacity degradation. However, challenges persist in achieving high predictive accuracy due to factors such as limited availability of quality datasets, inadequate preprocessing methods and the complexity of the relationships between variables. Additionally, issues like inconsistent handling of data below thresholds, limited adaptability of predictive models to new data and ineffective combination of multiple prediction techniques hinder the accuracy and reliability of predictions.
Thus, there exists an urgent need for solutions capable of addressing the challenges associated with battery capacity degradation forecasting.
SUMMARY
An aim of the present disclosure is to provide a system to determine battery capacity. Another aim of the present disclosure is to provide a method for forecasting battery capacity.
The present disclosure provides a system to determine battery capacity. The system comprises a data acquisition unit to acquire a capacity dataset associated with a battery. The capacity dataset comprises total charge capacity, discharge voltage and cycle number. The system comprises a dataset filtration unit to filter the acquired capacity dataset to create a first dataset comprising capacity data values greater than a predetermined threshold and a second dataset comprising capacity data values equal to the predetermined threshold. The system further comprises a model generation unit to generate a first prediction model based on the first dataset using a first technique and a second prediction model based on the second dataset using a second technique. The system also comprises a hybrid forecasting model generation unit to generate a hybrid prediction model based on the first prediction model and the second prediction model to determine the battery capacity of the battery.
In an embodiment, the first technique and the second technique are selected from Support Vector Regression, Long Short-Term Memory, Random Forest Regressor, Gradient Boosting Machines, AdaBoost, Extreme Gradient Boosting and neural network-based techniques.
In another embodiment, the first technique and the second technique are different.
In yet another embodiment, the predetermined threshold for the dataset filtration unit is selected from a range of 3 to 15 volts.
In still another embodiment, the system comprises a data preprocessing module to preprocess the acquired capacity dataset. The preprocessing comprises discarding capacity data values that are under the predetermined threshold.
In another embodiment, the system comprises a validation module to compare the forecasted parameters with the acquired capacity dataset to evaluate accuracy.
In still another embodiment, the system comprises a denoising module to reduce noise in the acquired capacity dataset.
In yet another embodiment, the dataset filtration unit employs dynamic thresholding based on parameters such as the real-time degradation rate of the battery, the charging cycle number or the usage profile of the battery.
In another embodiment, hybrid forecasting model processes additional parameters associated with the battery, including temperature data, current load profiles and environmental conditions.
In still another embodiment, the hybrid forecasting model comprises a weighting module to assign different weights to the first prediction model and the second prediction model based on a confidence level derived from the respective datasets.
The present disclosure also provides a method for forecasting battery capacity. The method comprises acquiring a capacity dataset associated with a battery. The capacity dataset comprises total charge capacity, discharge voltage and cycle number. The method comprises filtering the acquired capacity dataset to create a first dataset comprising capacity data values greater than a predetermined threshold and a second dataset comprising capacity data values equal to the predetermined threshold. The method further comprises generating a first prediction model based on the first dataset using a first technique. The method also comprises generating a second prediction model based on the second dataset using a second technique. The method additionally comprises generating a hybrid prediction model based on outputs from the first prediction model and the second prediction model to forecast the battery capacity of the battery.
In an embodiment, the first technique and the second technique are selected from Support Vector Regression, Long Short-Term Memory, Random Forest Regressor, Gradient Boosting Machines, AdaBoost, Extreme Gradient Boosting and neural network-based techniques.
In another embodiment, the first technique and the second technique are different.
In yet another embodiment, the method comprises preprocessing the acquired capacity dataset by discarding capacity data values that are under the predetermined threshold.
BRIEF DESCRIPTION OF DRAWINGS
The summary above, as well as the following detailed description of illustrative embodiments, is better understood when read in conjunction with the appended drawings. For the purpose of illustrating the present disclosure, exemplary constructions of the disclosure are shown in the drawings. However, the present disclosure is not limited to specific methods and instrumentalities disclosed herein. Moreover, those in the art will understand that the drawings are not to scale. Wherever possible, like elements have been indicated by identical numbers.
Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams wherein:
FIG. 1 illustrates a block diagram of a system to determine a battery capacity of a battery, in accordance with an embodiment of the present disclosure;
FIG. 2 illustrates schematic illustration of usage of the system of FIG. 1, in accordance with an embodiment of the present disclosure; and
FIG. 3 illustrates a a flowchart of a method for determining battery capacity, in accordance with an embodiment of the present disclosure.
In the accompanying drawings, an underlined number is employed to represent an item over which the underlined number is positioned or an item to which the underlined number is adjacent. A non-underlined number relates to an item identified by a line linking the non-underlined number to the item. When a number is non-underlined and accompanied by an associated arrow, the non-underlined number is used to identify a general item at which the arrow is pointing.
DETAILED DESCRIPTION
The following detailed description illustrates embodiments of the present disclosure and ways in which they can be implemented. Although some modes of carrying out the present disclosure have been disclosed, those skilled in the art would recognise that other embodiments for carrying out or practising the present disclosure are also possible.
The description set forth below in connection with the appended drawings is intended as a description of certain embodiments of a system to determine battery capacity and is not intended to represent the only forms that may be developed or utilised. The description sets forth the various structures and/or functions in connection with the illustrated embodiments; however, it is to be understood that the disclosed embodiments are merely exemplary of the disclosure that may be embodied in various and alternative forms. The figures are not necessarily to scale; some features may be exaggerated or minimised to show details of particular components. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a representative basis for teaching one skilled in the art to variously employ the present invention.
While the disclosure is susceptible to various modifications and alternative forms, specific embodiment thereof has been shown by way of example in the drawings and will be described in detail below. It should be understood, however, that it is not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure.
The terms “comprise”, “comprises”, “comprising”, “include(s)”, or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a setup, system that includes a list of components or steps does not comprise only those components or steps but may include other components or steps not expressly listed or inherent to such setup or system. In other words, one or more elements in a system or apparatus preceded by “comprises... a” does not, without more constraints, preclude the existence of other elements or additional elements in the system or apparatus.
In the following detailed description of the embodiments of the disclosure, reference is made to the accompanying drawings, and which are shown by way of illustration specific embodiments in which the disclosure may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the disclosure, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present disclosure. The following description is, therefore, not to be taken in a limiting sense.
The present disclosure will be described herein below with reference to the accompanying drawings. In the following description, well known functions or constructions are not described in detail since they would obscure the description with unnecessary detail.
Referring to FIG. 1, there is shown a block diagram of a system 100 to determine a battery capacity of a battery 102, in accordance with an embodiment of the present disclosure. The term “battery” as used throughout the present disclosure refers to an energy storage device comprising one or more battery cells to store and deliver energy in the form of electrical power. The battery 102 can be a lithium-ion battery, nickel-metal hydride battery or other types of rechargeable batteries commonly used in electric vehicles, portable devices and energy storage systems. The battery undergoes charge and discharge cycles during operation and a performance thereof deteriorates over time due to factors such as calendar ageing, cycle ageing and environmental influences. The term “system” as used throughout the present disclosure refers to an arrangement configured to determine the battery capacity of a battery 102. The term “battery capacity” relates to the amount of energy a battery can store and deliver, typically measured in milliampere-hours (mAh) or watt-hours (Wh). The battery capacity is a key parameter that reflects the health of the battery, enables predictions about remaining useful life and facilitates timely maintenance or replacement. Further, factors affecting battery capacity include the depth of discharge, operating temperature and cumulative usage cycles.
The system 100 comprises a data acquisition unit 104 to acquire a capacity dataset 102A associated with the battery 102. The term “data acquisition unit” as used throughout the present disclosure refers to a component of the system 100 to acquire the capacity dataset 102A associated with the battery. The data acquisition unit 104 may comprise sensors, communication modules or data logging devices that monitor parameters such as voltage, current and cycle count during the operation of the battery. The data acquisition unit 104 manages accurate and up-to-date representation of battery performance by continuously monitoring parameters such as discharge voltage during operation. Additionally, by capturing real-time cycle numbers, the data acquisition unit 104 supports detailed analysis of battery degradation patterns over time. The term “capacity dataset” as used throughout the present disclosure refers to a collection of data parameters associated with the battery 102. The capacity dataset 102A comprises total charge capacity, discharge voltage and cycle number. The total charge capacity reflects the maximum charge the battery can hold, discharge voltage indicates the voltage level during energy release and cycle number tracks the number of charge-discharge cycles completed by the battery. For example, the data acquisition unit 104 may monitor the battery 102 during an operation of the electric vehicle and record discharge voltage values at regular intervals, forming part of the capacity dataset 102A. In another example, the data acquisition unit 104 may collect charge capacity data during scheduled charging cycles in a stationary energy storage system.
The system 100 further comprises a dataset filtration unit 106 to filter the acquired capacity dataset 102A to create a first dataset 106A comprising capacity data values greater than a predetermined threshold and a second dataset 106B comprising capacity data values equal to the predetermined threshold. The term “dataset filtration unit” as used throughout the present disclosure refers to a component of the system 100 to categorize the acquired capacity dataset 102A into subsets based on a predetermined threshold. The first dataset 106A contains data values that exceed the threshold, often representing high-capacity battery states. The second dataset 106B includes data values equal to the threshold, which may indicate transitional battery performance states. The dataset filtration unit 106 enhances prediction accuracy by segmenting the data into distinct value ranges, allowing models to target specific performance scenarios. Furthermore, removing outlier data during filtration allows that subsequent prediction models are trained with consistent and meaningful inputs. For example, the dataset filtration unit 106 may filter capacity data where total charge capacity exceeds 15 volts and assign these values to the first dataset 106A. Alternatively, the unit may identify discharge voltage readings of exactly 3 volts and group them into the second dataset 106B for specific model training.
The system 100 also comprises a model generation unit 108 to generate a first prediction model 108A based on the first dataset 106A using a first technique and a second prediction model 108B based on the second dataset 106B using a second technique. The term “model generation unit” as used throughout the present disclosure refers to a component of the system 100 configured to create predictive models using the filtered datasets. The first prediction model 108A is trained using a first technique optimized for high-capacity battery states, while the second prediction model 108B employs a second technique customized for threshold-level battery performance data. The model generation unit 108 provides accurate and context-specific predictions by employing techniques suited to specific data subsets. Additionally, the use of distinct techniques for each dataset provides understanding of battery behaviour under varying conditions. For example, the model generation unit 108 may use Support Vector Regression to create the first prediction model 108A based on the first dataset 106A, enabling accurate estimation of remaining charge capacity. In another example, the unit may employ Long Short-Term Memory techniques to generate the second prediction model 108B using the second dataset 106B, capturing detailed battery degradation trends over time.
The system 100 further comprises a hybrid forecasting model generation unit 112 to generate a hybrid prediction model 112A based on the first prediction model 108A and the second prediction model 108B to determine the battery capacity of the battery 102. The term “hybrid forecasting model generation unit” as used throughout the present disclosure refers to a component of the system 100 configured to combine the outputs from multiple predictive models. The hybrid prediction model 112A integrates the strengths of the first prediction model 108A and the second prediction model 108B, resulting in enhanced prediction accuracy. The hybrid forecasting model generation unit 112 provides reliable predictions across various battery states by dynamically weighting the contributions of the individual models based on their confidence levels or historical accuracy. Furthermore, combining models allows for a robust system capable of handling a wide range of operating conditions. For example, the hybrid forecasting model generation unit 112 may combine predictions from the first and second prediction models, assigning greater weight to the model with more reliable past performance. In another example, the hybrid prediction model 112A may incorporate updated data from the dataset filtration unit 106, enabling continuous refinement of prediction accuracy.
In one exemplary scenario, during operation in an electric vehicle, the system 100 collects real-time data such as discharge voltage and cycle count using the data acquisition unit 104. The dataset filtration unit 106 processes this data to create filtered datasets, which the model generation unit 108 uses to train prediction models. The hybrid forecasting model generation unit 112 combines these models to predict the remaining lifespan of the battery 102 and identify when maintenance or replacement is required. In another exemplary scenario, the system 100 in a renewable energy storage system analyses historical capacity datasets collected over several months. The dataset filtration unit 106 segments such data and the model generation unit 108 creates customized prediction models. The hybrid forecasting model generation unit 112 integrates these models to recommend optimal battery replacement schedules, improving system efficiency and reliability.
In an embodiment, the first technique and the second technique used by the model generation unit 108 are selected from Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Random Forest Regressor, Gradient Boosting Machines, AdaBoost, Extreme Gradient Boosting (XGBoost) and neural network-based techniques. The term “technique” as used throughout the present disclosure refers to an algorithm or method employed to create predictive models for determining battery capacity. Each technique offers distinct advantages based on the characteristics of the datasets and the targeted prediction objectives. For example, SVR is particularly effective for regression problems involving smaller datasets with clear margins, while LSTM is suitable for sequential data and time-series forecasting, capturing long-term dependencies in battery behaviour. Random Forest Regressor and Gradient Boosting Machines utilize ensemble learning methods to aggregate predictions from multiple models, enhancing robustness and accuracy. Similarly, AdaBoost and XGBoost improve performance by focusing on data points with higher errors during successive iterations. Neural network-based techniques are versatile and capable of modelling complex, non-linear relationships, making them suitable for larger and more diverse datasets.
The system 100 achieves adaptability to different data characteristics and degradation patterns by selecting the first and second techniques from this range of options. The flexibility provided by these techniques enables that the first prediction model 108A and the second prediction model 108B are optimized for their respective datasets 106A and 106B. Such operation enhances the granularity and accuracy of predictions made by the hybrid prediction model 112A. Additionally, the variety of techniques enables the system 100 to adapt to evolving data trends and diverse battery usage scenarios, thereby reliable performance over time. For example, when processing datasets with strong temporal correlations, LSTM can effectively capture these patterns, while XGBoost is advantageous for managing highly non-linear relationships in complex datasets. Such versatility improves the predictive capability of the system 100 under various battery degradation conditions.
In an embodiment, the first technique used for the first prediction model 108A and the second technique used for the second prediction model 108B are different. The term “different” as used throughout the present disclosure refers to the use of distinct algorithms or methods for creating prediction models, which makes each model customized to the specific characteristics of its corresponding dataset. The use of different techniques allows the system 100 to address the unique requirements of the datasets 106A and 106B effectively. For example, the first dataset 106A, containing capacity data values greater than the predetermined threshold, may benefit from a technique like Gradient Boosting Machines, which is well-suited for non-linear relationships in large datasets. Conversely, the second dataset 106B, comprising values equal to the predetermined threshold, may require a technique like LSTM, which excels at sequential data and detecting subtle transitions in battery performance.
Employing distinct techniques for the first prediction model 108A and the second prediction model 108B optimizes each model for its respective dataset, enhancing the precision and reliability of predictions. Such a differentiation enables the hybrid forecasting model generation unit 112 to combine the complementary strengths of the prediction models. For example, integrating insights from a non-linear ensemble model such as Gradient Boosting Machines with sequential pattern detection from LSTM results in a hybrid prediction model 112A that is more robust and adaptable. The combination minimizes the limitations associated with relying on a single technique and provides understanding of battery behaviour across diverse operating conditions.
In an embodiment, the predetermined threshold used by the dataset filtration unit 106 is selected from a range of 3 to 15 volts. The term “predetermined threshold” as used throughout the present disclosure refers to a predefined value or range that determines how the dataset filtration unit 106 categorizes the capacity dataset 102A into the first dataset 106A and the second dataset 106B. Such a threshold distinguishes between different battery performance states, such as high-capacity and transitional conditions. The system 100 provides compatibility with typical battery operating voltages by selecting a threshold within the specified range of 3 to 15 volts, particularly for lithium-ion and other rechargeable batteries. Such a range encompasses common discharge voltage levels enables the filtration process to capture data relevant to battery performance without misclassifying key information.
The use of this threshold range enables the dataset filtration unit 106 to accurately segregate data into datasets 106A and 106B, optimizing the quality of input provided to the model generation unit 108. For example, a threshold at 3 volts focuses on capturing end-of-discharge data essential for analysing the degradation behaviour of the battery 102, while a threshold at 15 volts targets full-charge conditions, aiding predictions of peak battery performance. The dataset filtration unit 106 improves the accuracy of the first prediction model 108A and the second prediction model 108B by preventing the inclusion of irrelevant or anomalous data outside this range. Moreover, the threshold range enhances the system’s adaptability to varying battery chemistries and operational profiles.
In an embodiment, the system 100 comprises a data preprocessing module to preprocess the acquired capacity dataset 102A. The preprocessing includes discarding capacity data values that are under the predetermined threshold. The term “data preprocessing module” as used throughout the present disclosure refers to a component of the system 100 configured to refine the capacity dataset 102A before it is processed by the dataset filtration unit 106 or used for model generation. The data preprocessing involves systematically removing data points that fall below the predetermined threshold, as these values often represent noise, anomalies or states that do not contribute to meaningful analysis of battery performance. The model manages the dataset filtration unit 106 to receive high-quality, relevant data for further segmentation.
The preprocessing module reduces noise and prevents the inclusion of outliers by eliminating capacity data values under the predetermined threshold, which could otherwise distort model predictions. Such a step enhances the efficiency of the system 100 by reducing the computational load associated with processing unnecessary data. For example, removing discharge voltage readings below 3 volts avoids the inclusion of data caused by measurement errors or abnormal discharge conditions, which could compromise the reliability of the prediction models 108A and 108B. Additionally, the preprocessing step improves robustness of the hybrid prediction model 112A, resulting in more precise and actionable forecasts.
In an embodiment, the system 100 comprises a validation module to compare the obtained forecasted parameters with the acquired capacity dataset 102A for accuracy evaluation. The term “validation module” as used throughout the present disclosure refers to a component of the system 100 configured to verify the accuracy of predictions made by the hybrid prediction model 112A. The validation process involves comparing forecasted parameters, such as predicted charge capacity or discharge voltage, with actual values recorded in the capacity dataset 102A. This comparison enables the identification of deviations or errors, which the system 100 uses to refine the prediction models 108A and 108B or adjust the hybrid prediction model 112A.
The inclusion of the validation module improves reliability of the system 100 through close alignment between forecasted and observed parameters. For example, if the forecasted charge capacity significantly deviates from the actual recorded value, the validation module can prompt recalibration of weight assignments within the hybrid prediction model 112A or adjust the segmentation parameters in the dataset filtration unit 106. Such refinements enhance the precision of the system 100 in predicting battery performance under varying conditions. Additionally, the validation module enables continuous learning, allowing the system 100 to adapt dynamically to changing battery behaviours. Adaptability allows predictions for applications such as electric vehicles and energy storage systems, where forecasting battery health is essential for operational efficiency and maintenance planning.
In an embodiment, the system 100 comprises a denoising module to reduce noise in the acquired capacity dataset 102A. The term “denoising module” as used throughout the present disclosure refers to a component of the system 100 configured to process the capacity dataset 102A by removing noise or irrelevant fluctuations that could obscure meaningful patterns in the data. Noise can result from sensor inaccuracies, environmental disturbances or operational anomalies and if left unaddressed, this noise can compromise the training of prediction models 108A and 108B or distort forecasts made by the hybrid prediction model 112A.
The denoising module enhances the quality of the capacity dataset 102A by applying techniques such as median filtering, wavelet transforms or low pass filtering to eliminate unwanted variations while preserving trends. The dataset filtration unit 106, through denoising module, receives high-integrity data for segmentation into datasets 106A and 106B by smoothing fluctuations in parameters like discharge voltage or cycle count. This process improves the accuracy and reliability of the prediction models and reduces computational overhead by focusing on relevant data. For example, by removing spurious voltage readings caused by transient disturbances, true performance patterns are detected by the prediction models 108A and 108B, leading to more precise forecasts by the hybrid prediction model 112A. Additionally, by reducing the impact of irrelevant data points, the denoising module helps the system 100 produce robust and actionable predictions suitable for a wide range of operational scenarios.
In an embodiment, the dataset filtration unit 106 employs dynamic thresholding based on a real-time degradation rate of the battery 102, the charging cycle number or a usage profile of the battery. The term “dynamic thresholding” as used throughout the present disclosure refers to the process of adaptively adjusting the predetermined threshold used by the dataset filtration unit 106 based on current operating parameters and performance indicators of the battery 102. It will be appreciated that unlike static thresholding, which relies on fixed values, dynamic thresholding enables the system 100 to customize dataset segmentation to the evolving conditions of the battery.
The dataset filtration unit 106 improves the relevance and accuracy of the datasets 106A and 106B by aligning segmentation criteria with real-time battery behaviour by employing dynamic thresholding. For example, adjusting thresholds based on real-time degradation rates allows the system 100 to capture subtle declines in capacity or performance, such as rapid fading caused by high-discharge events. Incorporating the charging cycle number into the thresholding process helps the system 100 account for lifecycle-dependent degradation patterns, such as accelerated wear during the early cycles or stabilization over prolonged use. Additionally, using thresholds informed by the battery's usage profile ensures that the datasets reflect conditions specific to the application, such as heavy cycling in electric vehicles or sporadic usage in renewable energy systems. This adaptability enhances the training quality of the prediction models 108A and 108B, while also improving the robustness and reliability of the hybrid prediction model 112A provides accurate forecasts under diverse operational scenarios.
In an embodiment, the hybrid forecasting model generation unit 112 processes additional parameters associated with the battery 102. The additional parameters comprise one or more of temperature data, current load profiles, and environmental conditions. The term “additional parameters” as used throughout the present disclosure refers to supplemental data points beyond the capacity dataset 102A that influence battery performance and degradation. Temperature data captures both ambient and operating temperatures, which affect chemical reaction rates and overall battery efficiency. Current load profiles record the intensity and variability of electrical loads applied to the battery, while environmental conditions account for external factors such as humidity, pressure and other stressors that may accelerate degradation.
The hybrid forecasting model generation unit 112 significantly enhances the predictive accuracy of the hybrid prediction model 112A by incorporating these additional parameters. For example, including temperature data allows the system 100 to account for thermal effects on battery capacity, such as reduced performance at low temperatures or accelerated ageing under sustained high-temperature conditions. Analysing current load profiles helps the model detect and predict the impact of high-discharge or irregular load cycles, thereby predictions are customized to the operational demands of the battery. Environmental conditions such as humidity or atmospheric pressure allow the model to adapt forecasts to extreme or fluctuating environments, to allow accurate performance assessments even under challenging conditions. The integration of these parameters makes the hybrid prediction model 112A robust and adaptable, enabling the system 100 to deliver precise and actionable forecasts across a wide range of applications, including electric vehicles, portable devices and energy storage systems.
In an embodiment, the hybrid forecasting model generation unit 112 comprises a weighting module to assign different weights to the first prediction model 108A and the second prediction model 108B based on a confidence level derived from the respective datasets 106A and 106B. The term “weighting module” as used throughout the present disclosure refers to a component of the hybrid forecasting model generation unit 112 that dynamically adjusts the contributions of the first prediction model 108A and the second prediction model 108B to the hybrid prediction model 112A. The confidence level reflects the reliability and relevance of each prediction model and is determined by factors such as the quality of the dataset used for training, the historical accuracy of the model or its applicability to the current operating conditions of the battery 102.
The inclusion of the weighting module provides significant technical benefits by enhancing the adaptability and precision of the hybrid prediction model 112A. For example, if the first dataset 106A captures data from high-capacity battery states with minimal noise, the confidence level for the first prediction model 108A may be higher, prompting the weighting module to assign it greater influence. Conversely, if the second dataset 106B represents transitional states for understanding battery degradation, the second prediction model 108B may be weighted more heavily. This dynamic weighting mechanism utilizes the hybrid prediction model 112A to emphasize most reliable and contextually relevant data, resulting in accurate and actionable capacity forecasts under varying battery conditions.
The system 100 minimizes the impact of less reliable models or datasets on the final forecast by employing the weighting module, thereby reducing errors and improving the reliability of predictions. Additionally, through the weighting mechanism, the contributions of the prediction models 108A and 108B are balanced effectively, preventing over-reliance on any single model. This balance enhances the robustness of the hybrid prediction model 112A, allowing it to adapt to diverse operational scenarios, such as rapid capacity loss during heavy discharge or long-term degradation under sustained environmental stress. Furthermore, the weighting module broadens the applicability of the system 100, making it suitable for varied use cases, including electric vehicles requiring precise short-term capacity forecasts and renewable energy systems focused on long-term degradation analysis.
Referring to FIG. 2, there is shown a schematic illustration of usage of the system 100 of FIG. 1, in accordance with an embodiment of the present disclosure. As shown, the battery 102 is first used within an electric vehicle 200, such as an electric motorcycle, where high-performance requirements necessitate optimal capacity levels for efficient operation. Upon degradation of the battery capacity, the battery 102 is subsequently repurposed for use within a home inverter 202, where lower capacity requirements are sufficient for backup power applications.
This repurposing of the battery 102 provides significant advantages. Extending the lifecycle of the battery 102 across multiple applications reduces the cost burden on users by deferring the purchase of new batteries for secondary uses such as home inverters. Additionally, this reuse minimizes the production of e-waste, addressing environmental concerns associated with the disposal of degraded batteries. The reduction in e-waste also contributes to environmental sustainability by lowering the demand for raw materials needed for the production of new batteries, thereby conserving resources and reducing the ecological footprint of battery manufacturing processes.
The system 100 plays a key role in enabling this transition between applications by accurately forecasting the capacity of the battery 102 through the hybrid prediction model 112A. In other words, the battery 102 is utilized optimally during its lifecycle. The battery 102 remains in vehicular use until its capacity drops below the performance threshold required for an electric vehicle 200. The system 100 then enables the efficient redeployment of the battery 102 for use in a home inverter 202, where its remaining capacity is adequate for power backup applications. This sustainable battery management approach maximizes the utility of the battery 102, reduces waste, and supports both economic and environmental benefits.
Referring to FIG. 3, there is shown a flowchart of a method 300 for determining battery capacity, in accordance with an embodiment of the present disclosure. At a step 302, a capacity dataset associated with a battery is acquired. The capacity dataset comprises total charge capacity, discharge voltage, and cycle number. The acquisition of this dataset provides input parameters reflecting the operational history and performance of the battery, which are essential for accurate capacity forecasting. For example, total charge capacity indicates the maximum energy the battery can store, discharge voltage captures energy output during usage and cycle number tracks the number of charge-discharge cycles completed, offering a direct measure of battery degradation.
At a step 304, the acquired capacity dataset is filtered to create a first dataset comprising capacity data values greater than a predetermined threshold and a second dataset comprising capacity data values equal to the predetermined threshold. This filtering process segments the data into subsets customized to specific performance conditions of the battery. The method eliminates irrelevant or non-representative data by categorizing data in this manner, improving the quality of the subsequent prediction models. Predetermined threshold allows high-capacity and transitional battery states to be effectively captured in the first and second datasets, respectively, enhancing the accuracy of the model training process.
At a step 306, a first prediction model is generated based on the first dataset using a first technique and at a step 308, a second prediction model is generated based on the second dataset using a second technique. These prediction models are customized to the specific characteristics of the datasets, results in precise modelling of battery performance trends.
At a step 310, a hybrid prediction model is generated based on outputs from the first prediction model and the second prediction model to forecast the battery capacity of the battery. This hybrid prediction model integrates the strengths of both models, dynamically balancing their contributions based on the confidence levels associated with their respective datasets. This step enhances the adaptability and precision of the forecasting process and allow accurate predictions across a range of battery conditions.
In an embodiment, the first technique and the second technique are selected from: Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Random Forest Regressor, Gradient Boosting Machines, AdaBoost, Extreme Gradient Boosting (XGBoost), and neural network-based techniques. Each technique offers distinct advantages based on the dataset's structure and prediction requirements. For example, SVR is effective for regression problems with smaller datasets, while LSTM excels at capturing temporal dependencies in sequential data.
In another embodiment, the first technique and the second technique are different, allowing the method to leverage complementary strengths of diverse algorithms. The models address specific nuances of their respective datasets by employing different techniques, improving the robustness and reliability of the final predictions.
In yet another embodiment, the method further comprises preprocessing the acquired capacity dataset by discarding capacity data values that are under the predetermined threshold. The preprocessing removes noise, anomalies or irrelevant data points from the dataset, improving its quality before filtering. This step reduces computational overhead, allowing the filtering process at step 304 and the subsequent model training steps to operate on meaningful data. Preprocessing enhances the accuracy and efficiency of the method by only high-quality data utilization, leading to reliable capacity forecasts suitable for diverse applications, such as electric vehicles or renewable energy storage systems.
In the description of the present invention, it is also to be noted that, unless otherwise explicitly specified or limited, the terms “disposed,” “mounted,” and “connected” are to be construed broadly, and may for example be fixedly connected, detachably connected, or integrally connected, either mechanically or electrically. They may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meaning of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Modifications to embodiments and combination of different embodiments of the present disclosure described in the foregoing are possible without departing from the scope of the present disclosure as defined by the accompanying claims. Expressions such as “comprising”, “comprising”, “incorporating”, “have”, “is” used to describe and claim the present disclosure are intended to be construed in a non- exclusive manner, namely allowing for items, components or elements not explicitly described also to be present. Reference to the singular is also to be construed to relate to the plural where appropriate.
Although embodiments have been described with reference to a number of illustrative embodiments thereof, it should be understood that numerous other modifications and embodiments can be devised by those skilled in the art that will fall within the spirit and scope of the principles of this disclosure. More particularly, various variations and modifications are possible in the component parts and/or arrangements of the subject combination arrangement within the scope of the present disclosure, the drawings and the appended claims. In addition to variations and modifications in the component parts and/or arrangements, alternative uses will also be apparent to those skilled in the art.
,CLAIMS:WE CLAIM:
1. A system (100) to determine a battery capacity, the system (100) comprising:
- a data acquisition unit (104) to acquire a capacity dataset (102A) associated with a battery (102), wherein the capacity dataset (102A) comprises total charge capacity, discharge voltage and cycle number;
- a dataset filtration unit (106) to filter the acquired capacity dataset (102A) to create:
- a first dataset (108A) comprising capacity data values greater than a predetermined threshold; and
- a second dataset (108B) comprising capacity data values equal to the predetermined threshold;
- a model generation unit (110) to generate:
- a first prediction model (110A) based on the first dataset (108A) using a first technique; and
- a second prediction model (110B) based on the second dataset (108B) using a second technique; and
- a hybrid forecasting model generation unit (112) to generate a hybrid prediction model (112A) based on the first prediction model (110A) and the second prediction model (110B) to determine the battery capacity of the battery (102).
2. The system (100) as claimed in claim 1, wherein the first technique and the second technique are selected from: Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Random Forest Regressor, Gradient Boosting Machines, AdaBoost, Extreme Gradient Boosting (XGBoost) and neural network-based techniques.
3. The system (100) as claimed in claim 1, wherein the first technique and the second technique are different.
4. The system (100) as claimed in claim 1, wherein the predetermined threshold for the dataset filtration unit (106) is selected from a range of 3 to 15 Volts.
5. The system (100) as claimed in claim 1, wherein the system (100) comprises a data preprocessing module to preprocess the acquired capacity dataset (102A) and wherein the preprocessing comprises discarding capacity data values that are under the predetermined threshold.
6. The system (100) as claimed in claim 1, wherein the system (100) comprises a validation module to compare the obtained forecasted parameters with the acquired capacity dataset (102A) for accuracy evaluation.
7. The system (100) as claimed in claim 1, wherein the system (100) comprises a denoising module to reduce noise in the acquired capacity dataset (102A).
8. The system (100) as claimed in claim 1, wherein the dataset filtration unit (106) employs dynamic thresholding based on a real-time degradation rate of the battery (102), the charging cycle number or a usage profile of the battery (102).
9. The system (100) as claimed in claim 1, wherein the hybrid forecasting model processes additional parameters associated with the battery (102) and wherein the additional parameters comprise one or more of: temperature data, current load profiles and environmental conditions associated with the battery (102).
10. The system (100) as claimed in claim 1, wherein the hybrid forecasting model comprises a weighting module to assign different weights to the first prediction model (110A) and the second prediction model (110B) based on a confidence level derived from the respective datasets.
11. A method for determining battery capacity, the method comprising:
- acquiring a capacity dataset (102A) associated with a battery (102), wherein the capacity dataset (102A) comprises total charge capacity, discharge voltage and cycle number;
- filtering the acquired capacity dataset (102A) to create:
- a first dataset (108A) comprising capacity data values greater than a predetermined threshold; and
- a second dataset (108B) comprising capacity data values equal to the predetermined threshold;
- generating a first prediction model (110A) based on the first dataset (108A) using a first technique;
- generating a second prediction model (110B) based on the second dataset (108B) using a second technique; and
- generating a hybrid prediction model (112A) based on outputs from the first prediction model (110A) and the second prediction model (110B) to forecast the battery capacity of the battery (102).
12. The method as claimed in claim 1, wherein the first technique and the second technique are selected from: Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Random Forest Regressor, Gradient Boosting Machines, AdaBoost, Extreme Gradient Boosting (XGBoost) and neural network-based techniques.
13. The method as claimed in claim 1, wherein the first technique and the second technique are different.
14. The method as claimed in claim 1, further comprising preprocessing the acquired capacity dataset (102A) by discarding capacity data values that are under the predetermined threshold.