Lithium ion battery failure intelligent prediction system and method based on multi-source data fusion

Abstract

The invention discloses a lithium ion battery failure intelligent prediction system and method based on multi-source data fusion, and belongs to the technical field of battery health management. Aiming at the problems of data islands, feature processing defects, insufficient model generalization and the like in the prior art, the invention provides a three-level innovation system: 1, a multi-dimensional database architecture; 2, a mixed feature processor; and 3, a composite prediction model. Dynamic updating of the model is realized through an OTA upgrade interface, the failure prediction accuracy can be improved by 12-15% and the false alarm rate can be reduced to 3% or below through actual measurement verification, meanwhile, an interpretable failure root cause positioning function is provided, the problems of misjudgment and missed judgment in battery safety early warning are effectively solved, and the safety early warning effect is improved. The method is suitable for battery intelligent operation and maintenance management of electric vehicles, energy storage power stations and other scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery health management, and specifically relates to a lithium-ion battery failure prediction system that combines database architecture optimization with a machine learning model, and is particularly suitable for battery fault diagnosis in electric vehicles and energy storage systems. Background Art

[0002] In the field of lithium-ion battery failure prediction, with the widespread application of electric vehicles and energy storage systems, the requirements for battery safety and reliability have become increasingly stringent, and existing technologies have exposed many problems.

[0003] 1. Data island problem: For example, patent CN102842939B only monitors and analyzes the real-time voltage and current data during battery operation, but lacks the integration of process parameters during battery production, data under different test conditions, and environmental temperature and humidity. This makes it impossible to fully consider the factors that affect battery performance during the prediction process, resulting in one-sided prediction results.

[0004] 2. Feature processing defects: When processing battery data, patent CN115469227B mainly relies on the automatic feature extraction of neural networks for the processing of mixed data, and does not fully integrate the professional knowledge in the field of batteries. For example, when processing category features such as battery types (such as cylindrical batteries and square batteries), the differences between different types of batteries in terms of electrode structure, charging and discharging differences, etc. are not deeply explored.

[0005] 3. Insufficient model generalization: Taking patent CN119133660A as an example, although this technology realizes dynamic monitoring, it mainly focuses on the changes in electrolytes and does not involve a dynamic update mechanism for adaptability to changes in battery data distribution. As new battery materials and manufacturing processes continue to emerge, the prediction accuracy of the old model drops significantly when processing new data.

[0006] 4. Lack of explainability: The prediction model used in patent CN118734233B is similar to a black box operation. Although it can predict the abnormal state of lithium batteries through deep learning models, it lacks the ability to explain the prediction results and cannot provide engineers with the root cause of battery failure, making it difficult for engineers to take effective measures to optimize battery management strategies. Summary of the invention

[0007] The purpose of the present invention is to solve the problems of data islands, feature processing defects and insufficient model generalization in the prior art, and to provide a lithium-ion battery failure intelligent prediction system and method based on multi-source data fusion.

[0008] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0009] Intelligent Prediction System for Lithium-Ion Battery Failure Based on Multi-Source Data Fusion, the system includes the following parts:

[0010] Multi-Source Heterogeneous Data Fusion Module: Integrate battery data from multiple sources and different types, including electrochemical parameters, thermodynamic parameters, mechanical parameters, and environmental parameters, break data silos, and provide comprehensive data support for subsequent analysis;

[0011] Domain Knowledge-Enhanced Feature Processor: Based on battery domain knowledge, targetedly process numerical features, categorical features, and time series features to improve the quality of data features;

[0012] Composite Machine Learning Prediction Model: Integrate the GBDT feature screening module, the random forest main prediction model, and the Prophet time series verification module to achieve high-precision battery failure prediction;

[0013] Dynamic Update and Visualization Feedback System: Support the OTA upgrade interface, update model parameters in real time through the incremental learning mechanism, and present the prediction results in a three-dimensional visualization manner and feedback relevant information.

[0014] Furthermore, the multi-source heterogeneous data fusion module is constructed by the following methods:

[0015] Sensor Data Aggregation Method with Timestamp Alignment: Accurately align and aggregate battery data from different sensors according to timestamps to ensure the time consistency of data and provide accurate time series data for subsequent analysis;

[0016] Feature Association Technology Based on Knowledge Graph: Construct a knowledge graph related to batteries, use it to mine potential associations between battery data, establish an intelligent connection between experimental data and the historical knowledge base, and achieve in-depth fusion and utilization of data;

[0017] Abnormal Data Automatic Annotation Algorithm: Automatically identify and annotate outliers in battery data to facilitate subsequent processing and analysis of abnormal data and improve data quality.

[0018] Furthermore, the automatic annotation algorithm is implemented through the following steps:

[0019] (1) Dynamic Threshold Detection: Set dynamic statistical thresholds according to battery parameter types (such as voltage, temperature), and label data as abnormal when it exceeds the threshold range;

[0020] (2) Knowledge Graph Rule Matching: Associate and match abnormal data with failure mode nodes (such as "lithium plating", "SEI growth") in the knowledge graph to label the corresponding failure types;

[0021] (3) Clustering-assisted annotation: For abnormal data that fails to match the knowledge graph rules, outliers are separated through an unsupervised clustering algorithm (such as DBSCAN) and labeled as unknown abnormal categories.

[0022] Furthermore, the feature processor includes the following parts:

[0023] Channel-separated data processing architecture: For numerical features, categorical features, and time-series features, independent processing channels are designed respectively to achieve specialized processing of different types of features;

[0024] In the channel-separated data processing architecture, adaptive normalization processing is adopted for numerical features, and its formula is:

[0025]

[0026] where X is the original data feature, μ is the feature mean, and Q3 and Q1 are the third quartile and the first quartile respectively;

[0027] Adaptive sliding window mechanism: When processing time-series features, the length of the sliding window is automatically adjusted according to the data characteristics and change trends to flexibly capture the dynamic features in the data;

[0028] Expert rule-guided encoding strategy: When processing categorical features, expert rules in the battery field are introduced to perform enhanced encoding on the categorical features, fully reflecting the domain knowledge contained in the categorical features.

[0029] Furthermore, the expert rule-guided encoding strategy encodes categorical features (such as battery type, failure mode) into numerical values with domain significance based on the association relationship between the failure mode nodes and influence factor nodes in the battery field knowledge graph, specifically including: assigning codes to battery types according to battery differences (such as example battery 1 is coded as 100); adjusting the encoding weights in combination with the dynamic rules in the knowledge graph (such as when the temperature > 45°C, the capacity attenuation accelerates), and the encoding rules are as Figure 3 shown.

[0030] Furthermore, the composite machine learning prediction model specifically includes:

[0031] GBDT (Gradient Boosting Decision Tree) feature screening module: By calculating the feature importance, key features that have a significant impact on battery failure prediction are screened out, reducing the interference of redundant features and improving the model training efficiency and prediction accuracy;

[0032] Random forest main prediction model: Using the random forest algorithm as the main prediction model to perform the main prediction on the screened features, and using its powerful non-linear fitting ability and anti-overfitting characteristics to accurately predict the battery failure situation;

[0033] Prophet Time Series Verification Module: Based on the Prophet algorithm, it verifies the prediction results in the time dimension, corrects the deviation in the time series dimension, and ensures the rationality and accuracy of the prediction results in the time series.

[0034] The above three modules achieve integration through phased collaboration rather than parallel integration.

[0035] Further, the GBDT Feature Screening Module calculates the feature importance score based on the total information gain of the feature when splitting the decision tree nodes through the gradient boosting decision tree model, and normalizes the score to retain the features higher than the set threshold.

[0036] Further, the Dynamic Update and Visualization Feedback System includes:

[0037] Model Online Update Interface: Supports OTA (Over-The-Air) upgrade. Through the incremental learning mechanism, the model can obtain new data in real time and update the model parameters to maintain the adaptability of the model to new data and battery technology changes;

[0038] 3D Visualization Module for Prediction Results: Presents the battery failure prediction results in a 3D visualization manner, intuitively showing key information such as the health status, failure probability, and remaining service life of the battery;

[0039] Root Cause Analysis Matrix Module: Uses the root cause analysis matrix (RCA_Score = Σ(w_i * F_imp) / T_critical) to deeply analyze and locate the causes of battery failure, output accurate failure root cause information, and provide a basis for battery maintenance and improvement.

[0040] A prediction method based on the above system, the method is:

[0041] Data Collection and Fusion: Utilize the multi-source heterogeneous data fusion module to collect multi-source data such as electrochemical parameters, thermodynamic parameters, and mechanical parameters during the entire life cycle of the battery, and perform data fusion processing through the sensor data aggregation method with timestamp alignment, the feature association technology based on the knowledge graph, and the abnormal data automatic annotation algorithm;

[0042] Feature Processing: Use the feature processor enhanced by domain knowledge to perform feature extraction and encoding processing on the fused data through the multi-channel data processing architecture, the adaptive sliding window mechanism, and the encoding strategy guided by expert rules;

[0043] Model Prediction: Input the processed feature data into the composite machine learning prediction model, and successively go through GBDT feature screening, random forest main prediction, and Prophet time series verification to obtain the battery failure prediction results;

[0044] Result feedback and update: Through the dynamic update and visualization feedback system, the prediction results are displayed in a three-dimensional visualization form, and the root cause analysis matrix is used for root cause analysis of failures. At the same time, with the help of the model online update interface, the model is dynamically updated according to the newly collected data.

[0045] The beneficial effects of the present invention compared with the prior art are as follows:

[0046] 1. By constructing a multi-dimensional database architecture and a dynamic data fusion mechanism, the present invention integrates electrochemical, thermodynamic, mechanical parameters and environmental data during the entire life cycle of the battery, and solves the problem of data islands.

[0047] 2. The present invention adopts a hybrid feature processor. Through a multi-channel processing architecture, numerical features, categorical features and temporal features are processed respectively. The categorical features adopt domain knowledge enhanced coding, which fully reflects the internal differences of different types of batteries and effectively improves the quality of data features.

[0048] 3. The present invention sets up a model online update interface (supporting OTA upgrade). With the help of the incremental learning mechanism, the model can adjust parameters in a timely manner according to new data, and the update efficiency is increased by 40%, so as to better adapt to the iteration of battery technology and maintain a high prediction performance.

[0049] 4. The present invention uses the root cause analysis matrix (RCA_Score = Σ(w_i * F_imp) / T_critical) to accurately locate the cause of battery failure, and the root cause location accuracy reaches 89%, providing strong support for battery fault troubleshooting and technical improvement.

[0050] 5. The prediction accuracy is increased by 12 - 15% (compared with traditional single models); the false alarm rate is reduced to less than 3%. Description of the Drawings

[0051] Figure 1 is the system architecture diagram of the present invention. It shows the overall architecture of the system, including a multi-source heterogeneous data fusion module, a feature processing module, a composite prediction model module and a dynamic update and visualization feedback system.

[0052] Figure 2 is the working process of the hybrid feature processor described in the present invention. It shows the detailed working process of the hybrid feature processor, including numerical feature normalization, categorical feature encoding and temporal feature extraction.

[0053] Figure 3 is the schematic diagram of the domain knowledge coding matrix described in the present invention. It shows the specific structure and coding rules of the domain knowledge coding matrix.

[0054] Figure 4This is an example of the three-dimensional visualization interface described in the present invention. It shows the specific display content of the three-dimensional visualization interface, including the battery health status, failure probability, and remaining service life.

[0055] Figure 5 This is the model performance comparison curve described in the present invention. It shows the comparison curves between the present invention and the prior art in terms of prediction accuracy, false alarm rate, model update efficiency, etc. Specific Embodiments

[0056] The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments, but are not limited thereto. Any modification or equivalent replacement of the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention shall be covered by the protection scope of the present invention.

[0057] The present invention proposes a three-level innovation system:

[0058] 1. Innovation at the data layer

[0059] Adopt a multi-dimensional database architecture. The following is the data table structure design:

[0060] Create a database table named "failure_signature" to store feature information related to battery failure; the table structure includes the following fields: "signature_id": the primary key, integer type, uniquely identifying each record; "voltage_curve": binary large object (BLOB) type, storing data of the voltage change curve; "thermal_profile": JSON format, storing thermal distribution characteristics; "impedance_spectrum": text type, storing the alternating current impedance spectrum; "degradation_pattern": integer type, representing the association index with the knowledge base.

[0061] Dynamic data fusion mechanism: Realize the association between experimental data and the historical knowledge base through the feature fingerprint matching algorithm: SimScore = α * V_corr + β * T_diff + γ * C_loss;

[0062] Among them: ScoreSim represents the similarity score, used to measure the matching degree between data. α, β, and γ are domain knowledge weight coefficients, used to adjust the importance of different features in the matching process. V_corr represents voltage correlation, used to measure the similarity of voltage data. T_diff represents temperature difference, used to measure the difference degree of temperature data. C_loss represents capacity loss, used to measure the attenuation of battery capacity.

[0063] 2. Innovation at the algorithm layer

[0064] Hybrid Feature Processor

[0065] Adopt a multi-channel processing architecture:

[0066] For numerical features, adaptive normalization is used, and the formula is: X_norm = (X - μ) / (Q3 - Q1), where: X represents the original numerical feature. μ represents the mean of the feature. Q3 and Q1 represent the third quartile and the first quartile of the feature respectively, which are used to calculate the distribution range of the feature. Through normalization processing, this formula converts numerical features into a standardized range, facilitating subsequent model processing.

[0067] Category feature codes: Adopt domain knowledge enhanced coding, combine category features (such as battery types, failure modes, etc.) with professional knowledge in the battery field, and assign coding values with domain significance to each category feature. For example, different types of batteries (cylindrical batteries, square batteries, etc.) are assigned different codes according to their differences in electrode structure, charge and discharge characteristics, etc., to fully reflect their internal differences.

[0068] Time series features: On the one hand, use the sliding window feature extraction method, and the window length can be adaptively adjusted according to the characteristics and change trends of the data. On the other hand, through the sliding window, local features are extracted from the time series data to capture the dynamic change rules in the data, providing richer time series information for the model.

[0069] Composite Prediction Model

[0070] Construct a triple model architecture of GBDT - RF - Prophet:

[0071] GBDT (Gradient Boosting Decision Tree): Feature importance screening. It is used for feature importance screening. By calculating the importance scores of features, key features that have a significant impact on battery failure prediction are screened out, reducing the interference of redundant features, thereby improving the model training efficiency and prediction accuracy.

[0072] Random Forest: The main prediction model of random forest, using its powerful non - linear fitting ability and anti - overfitting characteristics to accurately predict the battery failure situation.

[0073] Prophet (Time Series Decomposition Algorithm): Based on the Prophet algorithm, the prediction results are verified in the time dimension to ensure the rationality and accuracy of the prediction results in the time series, further improving the reliability of the prediction.

[0074] 3. Application Layer Innovation

[0075] Present the battery failure prediction results in a three-dimensional visualization manner, intuitively showing key information such as the health status, failure probability, and remaining service life of the battery. Through the three-dimensional visualization interface, users can more clearly understand the current status and future trends of the battery, facilitating battery management decisions.

[0076] Root cause analysis matrix: Use the formula RCA_Score = Σ(w_i * F_imp) / T_critical to deeply analyze and locate the causes of battery failure, where: RCA_Score represents the root cause analysis score, used to measure the severity of the failure cause. w_i represents the weight of the i-th feature, reflecting the importance of this feature in failure analysis. F_imp represents the impact value of the feature, indicating the actual impact degree of this feature on battery failure. T_critical represents the critical threshold, used to determine whether the failure cause reaches the severity level.

[0077] Through this formula, the root cause of battery failure can be accurately located, providing a basis for battery maintenance and improvement.

[0078] Model online update interface: Support OTA (Over-The-Air) upgrade. Through the incremental learning mechanism, the model can obtain new data in real time and update the model parameters. When new battery data or technologies appear, the model can automatically adjust the parameters to maintain adaptability to new data and battery technology changes, thereby improving the generalization ability and prediction performance of the model.

[0079] Example 1

[0080] 1. Deploy a multi-source sensor network to collect the following information:

[0081] Electrochemical parameters: voltage, current, internal resistance;

[0082] Thermodynamic parameters: surface temperature, heat distribution;

[0083] Mechanical parameters: expansion coefficient, deformation data.

[0084] 2. Construct a knowledge graph

[0085] Function description: Implement a simple knowledge graph class "KnowledgeGraph" through Python code to store and associate battery failure modes and their influencing factors.

[0086] Specific implementation: Define a class "KnowledgeGraph" that contains an initialization method "__init__" for initializing the node information of the graph.

[0087] In the initialization method, two main node sets are defined: the "failure mode" nodes, which include "lithium plating", "SEI growth", "micro short circuit", etc.; and the "influence factors" nodes, which are stored in dictionary form. For example, "temperature > 45°C: accelerated capacity fade" means that when the temperature exceeds 45°C, the battery capacity fades more quickly.

[0088] A method "link_nodes" is provided to achieve pattern matching and association between nodes.

[0089] Example 2

[0090] A two-stage training strategy is adopted:

[0091] Function description: Implement a two-stage hybrid training strategy using Python code to train a lithium-ion battery failure prediction model.

[0092] Specific implementation: Define a function "hybrid_training(x,y)", where the input is feature data "x" and target label "y".

[0093] In the first stage, feature screening is performed. Instantiate a feature selector "selector" using the "GradientBoostingSelector()" class. Call the "selector.fit_transform(x,y)" method to screen out the key features that have a significant impact on the target "y" from the feature data "x", and store the screened features in the variable "selected_features".

[0094] In the second stage, the main model is trained. Define an ensemble model "EnsembleModel", which includes multiple base learners and a meta-learner: The base learners include: a random forest classifier "RandomForestClassifier()", denoted as "rf"; an optimized SVM classifier "OptimizedSVM()", denoted as "svm"; The meta-learner is a neural network "NeuralNetwork()". Call the "model.fit(x[:,selected_features],y)" method to train the model using the screened feature data "x[:,selected_features]" and the target label "y".

[0095] Example 3

[0096] Model dynamic update verification:

[0097] Function description: Implement the incremental learning and dynamic update functions of the model through Python code to verify the improvement of the model update efficiency.

[0098] Specific implementation:

[0099] Define a function "incremental_learning(new_data, model)", where the inputs are the newly collected data "new_data" and the model to be updated "model".

[0100] In the function, only perform incremental learning on the random forest part in the model: "Call the method "model.rf.partial_fit(new_data.x, new_data.y)", and use the new data "new_data.x" and the corresponding labels "new_data.y" to perform partial fitting on the random forest model "model.rf" to achieve dynamic update of the model parameters.

[0101] Upload new data (such as the charge and discharge curves of solid-state batteries) through the OTA interface. The model update time is shortened from 2 hours in the traditional method to 48 minutes, verifying the actual measured effect of a 40% improvement in update efficiency.

[0102] The above-described embodiments only represent one implementation manner of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

[0103] Example 4

[0104] Function description: This example aims at the failure prediction of lithium-ion batteries in industrial and commercial energy storage devices, demonstrating the application of the present invention in the field of industrial and commercial energy storage.

[0105] Specific implementation (performed according to Figure 1 the shown architecture):

[0106] (1) Data collection: Deploy a high-precision sensor network to collect parameters such as the voltage, current, temperature, and mechanical vibration of the battery.

[0107] (2) Feature processing: In view of the high load and complex working conditions of industrial equipment, adjust the parameters of the feature processing module and increase the weight of the mechanical vibration feature.

[0108] (3) Model training: Use the same two-stage training strategy as in Example 2, but perform model training on the battery data of industrial equipment.

[0109] (4) Result verification: Verify the advantages of the present invention in terms of prediction accuracy and false alarm rate by comparing with the traditional method.Figure 5 This is the difference between the model of the present invention and the traditional model in terms of accuracy prediction and false alarm rate.

Claims

1. An intelligent prediction system for the failure of lithium-ion batteries based on multi-source data fusion, characterized in that: The system comprises the following parts: Multi-source heterogeneous data fusion module: Integrates battery data from multiple sources and different types, including electrochemical parameters, thermodynamic parameters, mechanical parameters, and environmental parameters; Domain knowledge-enhanced feature processor: Based on battery domain knowledge, it processes numerical features, category features, and time series features in a targeted manner; Composite machine learning prediction model: Integrates the GBDT feature screening module, the random forest main prediction model, and the Prophet timing verification module to achieve high-precision battery failure prediction; Dynamic update and visual feedback system: supports OTA upgrade interface, updates model parameters in real time through incremental learning mechanism, presents prediction results in 3D visualization and provides feedback on relevant information.

2. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 1, wherein: The multi-source heterogeneous data fusion module is constructed using the following method: Sensor data aggregation method with timestamp alignment: accurately aligns and aggregates battery data from different sensors based on timestamps to ensure the temporal consistency of data and provide accurate time series data for subsequent analysis; Feature association technology based on knowledge graph: Build a battery-related knowledge graph, use it to mine potential associations between battery data, establish intelligent connections between experimental data and historical knowledge bases, and achieve deep integration and utilization of data; Abnormal data automatic labeling algorithm: Automatically identify abnormal values in battery data and label them to facilitate subsequent processing and analysis of abnormal data and improve data quality.

3. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 2, wherein: The automatic annotation algorithm is implemented by the following steps: (1) Dynamic threshold detection: Dynamic statistical thresholds are set according to the battery parameter type (such as voltage and temperature). When the data exceeds the threshold range, it is marked as abnormal. (2) Knowledge graph rule matching: associate and match abnormal data with failure mode nodes in the knowledge graph (such as "lithium deposition" and "SEI proliferation"), and mark the corresponding failure type; (3) Clustering-assisted labeling: For abnormal data that does not match the knowledge graph rules, an unsupervised clustering algorithm (such as DBSCAN) is used to separate cluster points and label them as unknown abnormal categories.

4. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 1, wherein: The feature processor includes the following parts: Channel-based data processing architecture: Independent processing channels are designed for numerical features, category features, and time series features to achieve specialized processing of different types of features; In the channel-wise data processing architecture, adaptive normalization is used for numerical features, and the formula is: Among them, X is the original data feature, μ is the feature mean, Q3 and Q1 are the third quartile and the first quartile respectively; Adaptive sliding window mechanism: When processing time series features, the length of the sliding window is automatically adjusted according to the data characteristics and change trends to flexibly capture the dynamic features in the data; Expert rule-guided coding strategy: When processing category features, expert rules in the battery field are introduced to perform enhanced coding on the category features to fully reflect the domain knowledge contained in the category features.

5. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 4, wherein: The expert rule-guided coding strategy encodes categorical features (such as battery type, failure mode) into domain-meaningful numerical values based on the association relationship between the failure mode nodes and influencing factor nodes in the battery domain knowledge graph. Specifically, it includes: assigning codes to battery types according to battery differences (such as coding example battery 1 as 100); adjusting the coding weights in combination with the dynamic rules in the knowledge graph (such as when the temperature > 45°C, the capacity attenuation accelerates).

6. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 1, wherein: The composite machine learning prediction model specifically includes: GBDT (Gradient Boosting Decision Tree) feature screening module: By calculating feature importance, it screens out the key features that have a significant impact on battery failure prediction; Random forest main prediction model: Using the random forest algorithm as the main prediction model to perform the main prediction on the screened features, and leveraging its powerful non-linear fitting ability and anti-overfitting characteristics to accurately predict the battery failure situation; Prophet time series verification module: Based on the Prophet algorithm, it verifies the prediction results in the time dimension to ensure the rationality and accuracy of the prediction results in the time series.

7. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 6, wherein: The GBDT feature screening module calculates the feature importance score through the gradient boosting decision tree model based on the total information gain of the feature when splitting the decision tree nodes, and normalizes the score to retain the features higher than the set threshold.

8. The intelligent prediction system for lithium-ion battery failure based on multi-source data fusion according to claim 1, characterized in that: The dynamic update and visualization feedback system includes: Model online update interface: Supporting OTA (Over-The-Air) upgrade, through the incremental learning mechanism, enabling the model to obtain new data in real time and update the model parameters to maintain the adaptability of the model to new data and battery technology changes; Prediction result three-dimensional visualization module: Presenting the battery failure prediction results in a three-dimensional visualization manner to intuitively display key information such as the health status, failure probability, and remaining service life of the battery; Root cause analysis matrix module: Using the root cause analysis matrix (RCA_Score = Σ(w_i * F_imp) / T_critical) to deeply analyze and locate the causes of battery failure, and output accurate failure root cause information to provide a basis for battery maintenance and improvement.

9. A prediction method based on the system according to any one of claims 1-8, characterized in that: The method is as follows: Data collection and fusion: Using the multi-source heterogeneous data fusion module, collecting multi-source data such as electrochemical parameters, thermodynamic parameters, and mechanical parameters during the entire life cycle of the battery, and performing data fusion processing through the sensor data aggregation method with timestamp alignment, the feature association technology based on the knowledge graph, and the abnormal data automatic annotation algorithm; Feature processing: Using the feature processor enhanced by domain knowledge, through the multi-channel data processing architecture, the adaptive sliding window mechanism, and the expert rule-guided coding strategy, performing feature extraction and coding processing on the fused data; Model prediction: Inputting the processed feature data into the composite machine learning prediction model, and successively passing through GBDT feature screening, random forest main prediction, and Prophet time series verification to obtain the battery failure prediction result; Result feedback and update: Through the dynamic update and visualization feedback system, the prediction results are presented in a three-dimensional visualization form, and the root cause analysis matrix is used for root cause analysis of failures; at the same time, with the help of the model online update interface, the model is dynamically updated according to the newly collected data.

Images