Intelligent diagnosis and residual life prediction system for PACK battery pack

Abstract

The invention discloses an intelligent diagnosis and residual life prediction system for a PACK battery pack, and relates to the technical field of PACK battery packs. According to the invention, multi-dimensional information is fused through high-frequency and low-delay data capture of the data acquisition module and a real-time deep learning algorithm of the data processing and fusion module, so that adaptability improvement under a dynamic working condition is realized; through comparison of an anomaly detection algorithm and a historical data mode of the intelligent diagnosis module and combination of adaptive threshold optimization, the false alarm rate is remarkably reduced, and graded early warning is triggered; data cleaning, normalization and feature extraction are performed through a distributed computing architecture of the data processing and fusion module, and the adaptability of the prediction model is enhanced by fully utilizing real-time information; diagnostic reports and maintenance suggestions are automatically generated through the output and control module, real-time data are regularly utilized to optimize precision through an online model updating mechanism of the self-calibration module, manual dependence is reduced, and robustness and reliability within the extreme temperature range of-20 DEG C to 60 DEG C are ensured.

Description

Technical Field

[0001] This invention relates to the field of PACK battery pack technology, and specifically to an intelligent diagnostic and remaining life prediction system for PACK battery packs. Background Technology

[0002] The battery pack is the core energy storage system of new energy vehicles. It consists of multiple battery cells connected in series or parallel, and integrates a battery management system (BMS), a thermal management system, and structural components to form the "energy heart" that can directly power the entire vehicle.

[0003] To improve safety, optimize performance, and reduce maintenance costs, intelligent diagnostics and remaining life prediction for battery packs are necessary. Existing intelligent diagnostics and remaining life prediction systems for battery packs often rely on offline data analysis and static models, failing to capture the dynamic changes of batteries under complex operating conditions in real time. This leads to diagnostic delays, insufficient prediction accuracy, and a tendency for false alarms or missed alarms, especially under extreme conditions such as battery aging, temperature fluctuations, or overcharging / discharging. Furthermore, existing systems have weak capabilities for fusing and processing multi-source heterogeneous data, making it difficult to fully utilize real-time information from sensors in BMS and thermal management systems, thus limiting the adaptability of the prediction model. Simultaneously, the maintenance process requires frequent manual intervention and calibration, increasing maintenance costs and potentially introducing errors due to human factors, affecting overall reliability and lifespan. Therefore, we propose an intelligent diagnostics and remaining life prediction system for battery packs. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent diagnostic and remaining life prediction system for battery packs in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention specifically adopts the following technical solution: A smart diagnostic and remaining life prediction system for a battery pack includes: The data acquisition module is used to acquire multi-source heterogeneous data from the battery management system, thermal management system, current / voltage sensor and temperature sensor in real time, so as to achieve high-frequency and low-latency data acquisition. The data processing and fusion module is used to clean, normalize, and extract features from the collected data, and to fuse multi-dimensional information through deep learning algorithms to improve adaptability to dynamic changes in battery status. The intelligent diagnostic module is used to analyze battery health status, fault modes and abnormal behavior in real time, reducing the risk of false alarms and missed alarms. The remaining life prediction module is used to dynamically predict the remaining life and provide a confidence interval by combining battery aging history, operating load and environmental factors. The output and control module is used to generate diagnostic reports, prediction results and maintenance suggestions, and automatically outputs them through a human-machine interface or remote monitoring system, reducing the need for manual intervention. The self-calibration module has a built-in online model update mechanism that regularly uses real-time data to optimize prediction accuracy, ensuring the system's robustness and reliability under extreme conditions.

[0006] Furthermore, the data acquisition module further includes a high-precision Hall current sensor for accurately measuring the charging and discharging current, and incorporates an ambient temperature compensation algorithm to reduce measurement errors.

[0007] Furthermore, the data processing and fusion module adopts a Kubernetes-based distributed computing architecture, which supports real-time data cleaning and feature fusion at edge nodes, thereby improving system response speed.

[0008] Furthermore, the intelligent diagnostic module integrates anomaly detection algorithms to identify potential thermal runaway risks by comparing historical data patterns in real time and triggering an early warning mechanism.

[0009] Furthermore, the anomaly detection algorithm is based on the following formula: Anomaly score = I - mu_I / sigma_I + ambda T - mu_T / sigma_T Where I is the real-time current measurement; mu_I is the historical average current; sigma_I is the current standard deviation; T is the real-time temperature measurement; mu_T is the historical average temperature; sigma_T is the temperature standard deviation; lambda is an adjustable weighting coefficient. Historical data statistics are updated through a sliding window mechanism, and anomaly scores are calculated by combining real-time input and compared with an adaptive threshold. When the anomaly score exceeds the threshold, a graded early warning mechanism is automatically triggered, prioritizing high-risk events. At the same time, the threshold parameters are optimized through feedback loops to improve detection accuracy and system robustness.

[0010] Furthermore, the remaining lifetime prediction module integrates an electrochemical impedance spectroscopy analysis algorithm, combines cyclic aging data, dynamically adjusts the prediction model parameters, and incorporates a reinforcement learning strategy to optimize the prediction cycle based on actual operating conditions and loads, providing personalized maintenance suggestions.

[0011] Furthermore, the prediction period is based on multi-source data such as real-time operating load, battery aging history, and ambient temperature. A reinforcement learning strategy is used to dynamically optimize the prediction time interval. For example, the short-term prediction is adjusted to the minute level and the long-term prediction to the week level according to the charging and discharging frequency. Combined with the confidence interval calculation algorithm, the accuracy and timeliness of the prediction results are ensured, while adaptively matching the needs of different application scenarios.

[0012] Furthermore, the output and control module supports integration with a remote monitoring platform via the MQTT protocol, automatically generating PDF format diagnostic reports and pushing them to mobile applications.

[0013] Furthermore, the self-calibration module incorporates a Bayesian optimization algorithm, automatically performing online model updates every 24 hours to adapt to performance drift in extreme temperature ranges (-20°C to 60°C).

[0014] Furthermore, the system also includes a security protection module that encrypts data transmission based on blockchain technology to prevent unauthorized access and ensure the integrity of diagnostic results.

[0015] The beneficial effects of this invention are as follows: This invention achieves enhanced adaptability under dynamic operating conditions by fusing multi-dimensional information through high-frequency, low-latency data acquisition in the data acquisition module and real-time deep learning algorithms in the data processing and fusion module. It significantly reduces false alarm rates and triggers tiered early warnings through anomaly detection algorithms and historical data pattern comparisons in the intelligent diagnosis module, combined with adaptive threshold optimization. The distributed computing architecture of the data processing and fusion module performs data cleaning, normalization, and feature extraction, fully utilizing real-time information to enhance the adaptability of the prediction model. The output and control module automatically generates diagnostic reports and maintenance suggestions, and the self-calibration module employs an online model update mechanism (such as Bayesian optimization algorithms) to periodically optimize accuracy using real-time data, reducing reliance on manual intervention and ensuring robustness and reliability within an extreme temperature range of -20°C to 60°C. Attached Figure Description

[0016] Figure 1 This is a block diagram of the present invention; Figure 2 This is a flowchart of the process of this invention; Figure 3 This is a diagram illustrating the method of using the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0018] Please see Figure 1 - Figure 3 This invention provides an intelligent diagnostic and remaining life prediction system for a battery pack, comprising: The data acquisition module is used to acquire multi-source heterogeneous data from the battery management system, thermal management system, current / voltage sensor, and temperature sensor in real time, achieving high-frequency, low-latency data capture. The data acquisition module enables real-time preprocessing and preliminary analysis of the data, using built-in advanced filtering algorithms (such as Kalman filters) to eliminate noise interference and improve data quality. It also supports a high sampling rate of 1000 times per second, ensuring the accuracy and stability of the data even under extreme temperature conditions, providing a reliable input basis for subsequent diagnostic modules.

[0019] The data processing and fusion module cleans, normalizes, and extracts features from the collected data. It then fuses multi-dimensional information using deep learning algorithms to improve adaptability to dynamic changes in battery status. This module enables efficient data cleaning and feature enhancement, utilizing convolutional neural networks (CNNs) or long short-term memory networks (LSTMs) for multi-dimensional information fusion. It accurately extracts key features such as battery voltage, current, and temperature, and optimizes model parameters through adaptive algorithms to improve robustness under dynamic operating conditions. Simultaneously, this module supports real-time feature dimensionality reduction and anomaly detection, reducing computational overhead and ensuring low-latency response when processing thousands of data points per second. This provides high-precision, fused, multi-dimensional data input for subsequent diagnostic modules, thereby enhancing the system's ability to capture battery degradation patterns and improve predictive reliability.

[0020] The intelligent diagnostic module analyzes battery health status, fault modes, and abnormal behaviors in real time, reducing the risk of false alarms and missed alarms. It achieves high-precision, real-time battery status assessment using deep neural networks (such as multilayer perceptrons or Transformer architectures) for health status quantification, fault classification, and anomaly detection. It further reduces the risk of false alarms and missed alarms by integrating time series analysis and pattern recognition algorithms (such as support vector machines or random forests). This module supports adaptive threshold adjustment and multi-source data fusion to ensure high diagnostic accuracy (>98%) under dynamic operating conditions. It also optimizes computing resources, achieving a low-latency response of 500 diagnostic requests per second, providing a reliable and robust input foundation for the remaining battery life prediction module.

[0021] The remaining life prediction module dynamically predicts remaining lifespan and provides confidence intervals by combining battery aging history, operating load, and environmental factors. It achieves high-precision, dynamic remaining life prediction by utilizing deep reinforcement learning (such as the DRL algorithm) combined with a Gaussian process regression (GPR) model. This integrates multi-dimensional data sources including battery aging history, operating load, and environmental factors, updating prediction results and generating probability confidence intervals in real time. The module supports adaptive model training and online calibration. By integrating time series analysis and Bayesian optimization algorithms, it reduces prediction errors (error rate <5%) while optimizing computational overhead, ensuring low latency (latency <50ms) when processing 300 prediction requests per second, providing reliable and robust input for battery maintenance decisions.

[0022] The output and control module generates diagnostic reports, prediction results, and maintenance recommendations, which are then automatically output through a human-machine interface or remote monitoring system, reducing the need for manual intervention. This module enables efficient and automated report generation and output control, utilizing template engines or natural language processing technologies (such as the BERT model) to generate structured diagnostic reports, prediction results, and maintenance recommendations. It seamlessly integrates with the human-machine interface or remote monitoring system via API, enabling real-time data push. This module supports adaptive output format adjustment and priority scheduling, ensuring low-latency response (latency <30ms) under high load conditions while optimizing computational overhead, processing over 400 output requests per second. This provides operators with immediate and reliable decision support, significantly reducing the frequency of manual intervention and improving overall system efficiency.

[0023] The self-calibration module features a built-in online model update mechanism that periodically optimizes prediction accuracy using real-time data, ensuring the system's robustness and reliability under extreme conditions. Through settings, it enables efficient online model calibration and accuracy optimization, utilizing real-time data streams for incremental learning (such as Bayesian updates or adaptive gradient descent algorithms) to periodically adjust model parameters and reduce prediction bias (error rate <3%). This module supports dynamic threshold settings and anomaly handling mechanisms, ensuring high reliability (accuracy >99%) under battery aging, temperature fluctuations, or high-load conditions. Simultaneously, it optimizes computing resources, achieving a low-latency response of 250 calibration requests per second (latency <40ms), providing continuous optimization input for the entire system and significantly improving the stability and reliability of long-term predictions.

[0024] The combined collaboration of the above modules can address the shortcomings of existing systems in the background technology: By using the high-frequency, low-latency data acquisition module and the real-time deep learning algorithm of the data processing and fusion module to fuse multi-dimensional information, the adaptive improvement under dynamic working conditions can solve the problems of diagnostic delay and insufficient prediction accuracy caused by relying on offline data analysis and static models. By comparing anomaly detection algorithms and historical data patterns in the intelligent diagnostic module, combined with adaptive threshold optimization, the false alarm rate is significantly reduced and tiered early warnings are triggered, addressing the issue that false alarms or missed alarms perform worse under extreme conditions (such as battery aging, temperature fluctuations, or overcharging / discharging). The distributed computing architecture of the data processing and fusion module performs data cleaning, normalization, and feature extraction, fully utilizing real-time information to enhance the adaptability of the prediction model, thus addressing the weak fusion processing capability of multi-source heterogeneous data (such as BMS and thermal management system sensor information). The automatic generation of diagnostic reports and maintenance suggestions by the output and control module, along with the online model update mechanism of the self-calibration module (such as Bayesian optimization algorithms), periodically optimizes accuracy using real-time data, reducing manual intervention and ensuring robustness and reliability within the extreme temperature range of -20°C to 60°C, thus addressing the problem of frequent manual intervention and calibration increasing maintenance costs and introducing errors.

[0025] In this embodiment, preferably, the data acquisition module further includes a high-precision Hall current sensor for accurately measuring the charging and discharging current, and is combined with an ambient temperature compensation algorithm to reduce measurement errors.

[0026] In this embodiment, preferably, the data processing and fusion module adopts a Kubernetes-based distributed computing architecture, which supports real-time data cleaning and feature fusion at edge nodes, thereby improving system response speed.

[0027] In this embodiment, preferably, the intelligent diagnostic module further integrates anomaly detection algorithms to identify potential thermal runaway risks by comparing historical data patterns in real time and triggering an early warning mechanism.

[0028] In this embodiment, preferably, the anomaly detection algorithm is based on the following formula: Anomaly score = I - mu_I / sigma_I + ambda T - mu_T / sigma_T Where I is the real-time current measurement; mu_I is the historical average current; sigma_I is the current standard deviation; T is the real-time temperature measurement; mu_T is the historical average temperature; sigma_T is the temperature standard deviation; ambda is an adjustable weighting coefficient. Historical data statistics are updated through a sliding window mechanism, and anomaly scores are calculated by combining real-time input and compared with an adaptive threshold. When the anomaly score exceeds the threshold, a graded early warning mechanism is automatically triggered, prioritizing high-risk events. At the same time, the threshold parameters are optimized through feedback loops to improve detection accuracy and system robustness.

[0029] In this embodiment, preferably, the remaining lifetime prediction module integrates an electrochemical impedance spectroscopy analysis algorithm, combines cyclic aging data, dynamically adjusts the prediction model parameters, and combines a reinforcement learning strategy to optimize the prediction cycle based on actual operating conditions and load, providing personalized maintenance suggestions.

[0030] In this embodiment, preferably, the prediction period is based on multi-source data such as real-time operating load, battery aging history, and ambient temperature. A reinforcement learning strategy is used to dynamically optimize the prediction time interval. For example, the short-term prediction is adjusted to the minute level and the long-term prediction to the week level according to the charging and discharging frequency. Combined with the confidence interval calculation algorithm, the accuracy and timeliness of the prediction results are ensured, while adaptively matching the needs of different application scenarios.

[0031] In this embodiment, preferably, the output and control module supports integration with the remote monitoring platform via the MQTT protocol, automatically generating PDF format diagnostic reports and pushing them to the mobile application.

[0032] In this embodiment, preferably, the self-calibration module has a built-in Bayesian optimization algorithm that automatically performs online model updates every 24 hours to adapt to performance drift in extreme temperature ranges (-20°C to 60°C).

[0033] In this embodiment, preferably, the system also includes a security protection module that encrypts data transmission based on blockchain technology to prevent unauthorized access and ensure the integrity of diagnostic results.

[0034] The working principle and usage process of this invention include the following steps: 1. The data acquisition module starts up and acquires multi-source heterogeneous data in real time from the battery management system, thermal management system, current / voltage sensor and temperature sensor, including parameters such as current, voltage and temperature. It uses a high-precision Hall current sensor combined with an ambient temperature compensation algorithm to perform accurate measurements, ensuring a high sampling rate of 1000 times per second, and maintaining data accuracy and stability under extreme temperature conditions. Initially, noise interference is eliminated through a Kalman filter to improve data quality.

[0035] 2. The data processing and fusion module receives the collected data, performs data cleaning and normalization operations, and extracts features in real time at edge nodes through a Kubernetes-based distributed computing architecture. It uses convolutional neural networks (CNN) or long short-term memory networks (LSTM) to fuse multi-dimensional information, accurately extract key features (such as voltage anomalies and temperature fluctuations), and simultaneously executes adaptive algorithms to optimize model parameters and perform real-time feature dimensionality reduction. It supports processing thousands of data points per second, reduces computational overhead, ensures low-latency response, and outputs high-precision fused data to the diagnostic module.

[0036] 3. The intelligent diagnostic module uses fused data and deep neural networks (such as multilayer perceptron or Transformer architecture) to quantify the battery health status in real time. It combines time series analysis and pattern recognition algorithms (such as support vector machine or random forest) to classify faults and detect anomalies. The anomaly detection algorithm calculates a real-time anomaly score (based on current standard deviation, temperature standard deviation and adjustable weight coefficients), compares it with an adaptive threshold to identify potential thermal runaway risks, triggers a graded early warning mechanism, and optimizes computing resources to achieve 500 diagnostic requests per second, ensuring a diagnostic accuracy rate of over 98% and reducing the risk of false alarms and missed alarms.

[0037] 4. The remaining life prediction module integrates diagnostic results, battery aging history, operating load, and environmental factors. It uses deep reinforcement learning (DRL) combined with a Gaussian process regression (GPR) model to dynamically predict the remaining life. It integrates electrochemical impedance spectroscopy analysis algorithm and reinforcement learning strategy to optimize the prediction cycle (e.g., adjusting short-term prediction to minute level and long-term prediction to week level according to charge and discharge frequency). It generates probability confidence intervals in real time and reduces prediction error to less than 5% through Bayesian optimization algorithm. It supports processing 300 prediction requests per second with a latency of less than 50ms and provides personalized maintenance suggestions.

[0038] 5. The output and control module receives prediction results and diagnostic reports, and automatically generates structured PDF reports and maintenance suggestions using a template engine or natural language processing technology (such as the BERT model). It seamlessly integrates with remote monitoring platforms or mobile applications via the MQTT protocol to achieve real-time data push, supports adaptive output format adjustment and priority scheduling, processes more than 400 output requests per second, and has a latency of less than 30ms, significantly reducing the frequency of manual intervention.

[0039] 6. The self-calibration module performs online model updates periodically (every 24 hours). It uses real-time data streams to perform incremental learning through Bayesian optimization algorithms or adaptive gradient descent, dynamically adjusting model parameters to reduce prediction bias to an error rate of less than 3%. Combined with anomaly handling mechanisms and dynamic threshold settings, it ensures that the accuracy rate remains above 99% under battery aging, temperature fluctuations, or high load conditions. It supports 250 calibration requests per second with a latency of less than 40ms, improving the long-term stability and reliability of the system.

[0040] 7. The security protection module monitors data transmission throughout the process, encrypts data streams based on blockchain technology to prevent unauthorized access and ensure the integrity of diagnostic results, and optimizes system parameters through feedback loops to achieve end-to-end security protection.

[0041] 8. The entire system maintains robustness and reliability in an extreme temperature range of -20°C to 60°C through the coordinated operation of its modules. Users can view real-time reports through the human-machine interface or remote monitoring system and implement preventive measures based on maintenance recommendations, thereby effectively reducing operation and maintenance costs and extending battery pack life.

[0042] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart diagnostic and remaining life prediction system for a battery pack, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous data from the battery management system, thermal management system, current / voltage sensor and temperature sensor in real time, so as to achieve high-frequency and low-latency data acquisition. The data processing and fusion module is used to clean, normalize, and extract features from the collected data, and to fuse multi-dimensional information through deep learning algorithms to improve adaptability to dynamic changes in battery status. The intelligent diagnostic module is used to analyze battery health status, fault modes and abnormal behavior in real time, reducing the risk of false alarms and missed alarms. The remaining life prediction module is used to dynamically predict the remaining life and provide a confidence interval by combining battery aging history, operating load and environmental factors. The output and control module is used to generate diagnostic reports, prediction results and maintenance suggestions, and automatically outputs them through a human-machine interface or remote monitoring system, reducing the need for manual intervention. The self-calibration module has a built-in online model update mechanism that regularly uses real-time data to optimize prediction accuracy, ensuring the system's robustness and reliability under extreme conditions.

2. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The data acquisition module further includes a high-precision Hall current sensor for accurately measuring charging and discharging current, and incorporates an ambient temperature compensation algorithm to reduce measurement errors.

3. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The data processing and fusion module adopts a Kubernetes-based distributed computing architecture, which supports real-time data cleaning and feature fusion at edge nodes, improving system response speed.

4. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The intelligent diagnostic module further integrates anomaly detection algorithms, which identify potential thermal runaway risks by comparing historical data patterns in real time and trigger an early warning mechanism.

5. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 4, characterized in that, The anomaly detection algorithm is based on the following formula: Anomaly score = I - mu_I / sigma_I + ambda_T - mu_T / sigma_T Where I is the real-time current measurement; mu_I is the historical average current; sigma_I is the current standard deviation; T is the real-time temperature measurement; mu_T is the historical average temperature; sigma_T is the temperature standard deviation; lambda is an adjustable weighting coefficient. Historical data statistics are updated through a sliding window mechanism, and anomaly scores are calculated by combining real-time input and compared with an adaptive threshold. When the anomaly score exceeds the threshold, a graded early warning mechanism is automatically triggered, prioritizing high-risk events. At the same time, the threshold parameters are optimized through feedback loops to improve detection accuracy and system robustness.

6. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The remaining lifetime prediction module integrates an electrochemical impedance spectroscopy analysis algorithm, combines cyclic aging data, dynamically adjusts the prediction model parameters, and incorporates a reinforcement learning strategy to optimize the prediction cycle based on actual operating conditions and loads, providing personalized maintenance suggestions.

7. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The prediction period is based on multi-source data such as real-time operating load, battery aging history, and ambient temperature. A reinforcement learning strategy is used to dynamically optimize the prediction time interval. For example, the short-term prediction is adjusted to the minute level and the long-term prediction to the week level according to the charging and discharging frequency. Combined with the confidence interval calculation algorithm, the accuracy and timeliness of the prediction results are ensured, while adaptively matching the needs of different application scenarios.

8. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The output and control module supports integration with a remote monitoring platform via the MQTT protocol, automatically generating PDF format diagnostic reports and pushing them to mobile applications.

9. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The self-calibration module incorporates a Bayesian optimization algorithm and automatically performs online model updates every 24 hours to adapt to performance drift under extreme temperature ranges.

10. The intelligent diagnostic and remaining life prediction system for a battery pack according to claim 1, characterized in that, The system also includes a security module that encrypts data transmission based on blockchain technology to prevent unauthorized access and ensure the integrity of diagnostic results.

Images