A system for realizing remaining life prediction of core components of charging piles based on big data
By constructing a hybrid model of XGBoost and LSTM using a big data system, the accuracy and adaptability issues of lifespan prediction for core components of charging piles were resolved. This enabled accurate lifespan prediction and maintenance guidance for core components of charging piles, improving maintenance efficiency and service quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANDONG ARTAPLAY INTELLIGENT TECH CO LTD
- Filing Date
- 2026-01-08
- Publication Date
- 2026-06-02
AI Technical Summary
In the current operation and maintenance management of charging piles, there is a lack of accurate assessment of the health status of core components, which leads to over-maintenance or sudden shutdowns. Furthermore, existing technologies cannot quantitatively predict the remaining lifespan and lack dynamic adaptability.
A big data-based system is adopted, which constructs a hybrid model combining XGBoost and LSTM through a data acquisition layer, a data preprocessing layer and a model training layer to predict the remaining life of the core components of charging piles. It covers multi-dimensional data acquisition and feature engineering, dynamically adjusts the acquisition frequency, and conducts model training and prediction applications.
It enables quantitative prediction of the remaining lifespan of core components of charging piles, improving prediction accuracy and applicability, reducing operation and maintenance costs, reducing charging pile downtime, and improving service quality.
Smart Images
Figure CN122134315A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of charging pile operation and maintenance technology, and in particular to a system for predicting the remaining lifespan of core components of charging piles based on big data. Background Technology
[0002] In today's era of booming electric vehicle development, charging stations, as crucial infrastructure, directly impact the operational efficiency of the entire charging network through their reliability and the lifespan of their core components. Core components in charging stations, such as filter capacitors and IGBTs, will gradually age due to electrical stress, thermal cycling, and other factors during long-term operation, inevitably leading to failure.
[0003] However, the current operation and maintenance management methods for these core components have many drawbacks. Most of them adopt the model of regular inspection or post-event maintenance. This model lacks accurate assessment of the health status of the components, which can easily lead to problems such as over-maintenance or sudden shutdown, causing great trouble to the normal operation of charging piles.
[0004] Specifically, existing technical solutions have a series of significant drawbacks: (1) Single data dimension: Only a few operating parameters are collected. Key factors affecting the lifespan of core components, such as the number of module switching, the number of temperature cycles, the cumulative value of output energy, and the frequency of voltage fluctuations, are not fully covered. This makes the description of the component status incomplete and greatly reduces the accuracy of the early warning.
[0005] (2) Simple analysis method: It is difficult to uncover the deep-seated correlation behind the data by using fixed thresholds or basic statistical analysis methods, and it is also difficult to quantify the intrinsic mapping relationship between the aging degree of components and the operating data, thus making it difficult to achieve quantitative prediction of the remaining life.
[0006] (3) Lack of dynamic adaptability: The aging patterns of core components of charging piles vary in different environments (such as high temperature and high humidity areas) and under different usage frequencies. However, the fixed thresholds and analysis logic in the existing solutions cannot adapt to diverse scenarios and have poor versatility.
[0007] (4) Early warning lag: It is often only when the component is close to failure or has already become abnormal that an early warning is issued. It is impossible to predict the remaining service life in advance, and therefore it is impossible to leave enough time window for the formulation of operation and maintenance plans.
[0008] To overcome the challenge of accurately predicting the remaining lifespan of core components and to promote the development of operation and maintenance work towards a more proactive and precise direction, this invention proposes a system for predicting the remaining lifespan of core components of charging piles based on big data. Summary of the Invention
[0009] To overcome the shortcomings of existing technologies, this invention provides a simple and efficient system for predicting the remaining lifespan of core components of charging piles based on big data.
[0010] This invention is achieved through the following technical solution: A system for predicting the remaining lifespan of core components of charging piles based on big data includes a data acquisition layer, a data preprocessing layer, a model training layer, and a prediction application layer. The data acquisition layer is responsible for comprehensively collecting multi-dimensional data related to the lifespan of core components during the operation of charging piles, including basic component information, operating status data, environmental data, and fault history data. The data acquisition layer collects real-time operating status data through the CAN bus and RS485 interface of the charging pile controller, collects environmental data through integrated environmental sensors, and enters basic component information and fault history data through the operation and maintenance management platform. The data collection frequency is dynamically adjusted according to the data type. The data collection frequency for operational status is once per minute, the data collection frequency for environmental data is once per hour, and the data collection frequency for basic component information and fault history is entered and updated as needed.
[0011] The data preprocessing layer is responsible for cleaning, integrating, and performing feature engineering on the collected raw data to provide high-quality data for model training. The data preprocessing layer process is as follows: (1) Data cleaning First, outliers are removed, then missing values are filled using linear interpolation, and duplicate data are deleted. (2) Data integration Multi-dimensional data of the same core component belonging to the same charging pile are linked and integrated according to the time axis to form a structured dataset indexed by component-time. (3) Feature Engineering Calculate and extract statistical features, explore potential correlations between parameters, construct derived features, and map all feature data to the same scale to avoid the model training effect being affected by differences in feature dimensions.
[0012] The feature engineering implementation process is as follows: a) Basic Feature Extraction Calculate and extract statistical features, including the number of temperature cycles, the cumulative number of switching cycles, the cumulative output energy, the voltage fluctuation coefficient, and the average peak current. b) Derivative Feature Construction We construct temperature-on / off cycle interaction characteristics, output energy-temperature correlation characteristics, and ambient humidity-aging rate characteristics to explore potential correlations among parameters. c) Feature standardization The Z-score normalization method is used to map all feature data to the same scale, so as to avoid the model training effect being affected by the difference in feature dimensions.
[0013] The model training layer is responsible for training the remaining life prediction model of core components based on the preprocessed dataset using machine learning algorithms and optimizing the model parameters. The model training layer uses machine learning algorithms to train the remaining life prediction model for core components, and the steps are as follows: (1) Dataset partitioning Divide the structured dataset into training, validation, and test sets according to a custom ratio (e.g., 7:2:1); (2) Model selection and initialization A hybrid model combining Gradient Boosting Tree (XGBoost) and Long Short-Term Memory (LSTM) network was selected as the core prediction model, and the model hyperparameters were initialized. The model hyperparameters include the tree depth and learning rate of the gradient boosting tree XGBoost, and the number of hidden layer neurons and the number of iterations of the long short-term memory network LSTM.
[0014] Among them, Gradient Boosting Tree (XGBoost) is used to capture non-linear statistical associations between features, and Long Short-Term Memory Network (LSTM) is used to learn long-term dependencies in time series data. (3) Model training The training set is input into the hybrid model, the predicted value is calculated through forward propagation, the loss function is calculated based on the true remaining lifespan label, the model parameters are iteratively updated through the backpropagation algorithm, and the loss function is minimized. (4) Model evaluation The mean absolute error (MAE), root mean square error (RMSE), and prediction accuracy were used as evaluation metrics. The thresholds for these evaluation metrics were customized, and the model performance was evaluated using validation and test sets. (5) Parameter tuning If the model evaluation fails to meet the target, the hyperparameters are tuned using a grid search method, and the training-evaluation process is repeated until the model performance meets the target.
[0015] During model training, the model training layer uses the difference between the actual operating time of the component and its rated life as the label of the true remaining life, and uses the mean squared error (MSE) as the loss function.
[0016] The prediction application layer is responsible for deploying the trained model to the charging pile operation and maintenance management platform to achieve real-time prediction of remaining lifespan and operation and maintenance guidance.
[0017] The prediction application layer implements real-time prediction of remaining lifetime and operation and maintenance guidance, as follows: (1) Real-time data acquisition and input The data acquisition layer collects the latest operational and environmental data in real time and transmits it to the trained model; (2) Data preprocessing The data preprocessing layer is responsible for cleaning, integrating, and feature engineering the collected real-time data. (3) Model prediction The preprocessed real-time data is input into the deployed model, and the model outputs the remaining life prediction value (unit: months) and prediction confidence level of the core components; (4) Operation and maintenance decision generation Based on the remaining lifespan prediction results, a threshold is set to generate a reminder message, and combined with the charging pile's location and usage frequency information, a customized and optimized operation and maintenance resource scheduling plan is developed. When the remaining lifespan is ≤6 months, a priority maintenance reminder is generated; When the remaining lifespan is 6-12 months, generate a planned maintenance reminder; When the remaining life expectancy is greater than 12 months, generate routine monitoring recommendations; (5) Terminal push The forecast results and maintenance suggestions will be pushed to relevant maintenance personnel through the charging pile maintenance management platform or mobile APP terminal to guide them in carrying out targeted maintenance work.
[0018] The beneficial effects of this invention are: the system for predicting the remaining lifespan of core components of charging piles based on big data has achieved quantitative prediction of the remaining lifespan of core components, rather than just providing fault warnings. It can adapt to charging pile equipment with different operating environments and different usage frequencies, reduce operation and maintenance costs, reduce charging pile downtime, and improve the quality of charging services. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Appendix Figure 1 This is a schematic diagram of the system for predicting the remaining lifespan of core components of charging piles based on big data, as per the present invention.
[0021] Appendix Figure 2 This is a schematic diagram of the data preprocessing method of the present invention.
[0022] Appendix Figure 3 This is a schematic diagram of the model training method of the present invention.
[0023] Appendix Figure 4 This is a schematic diagram illustrating the application process of the model trained in this invention. Detailed Implementation
[0024] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly and completely described below in conjunction with the embodiments of this invention. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.
[0025] This system, which uses big data to predict the remaining lifespan of core components of charging piles, includes a data acquisition layer, a data preprocessing layer, a model training layer, and a prediction application layer. The data acquisition layer is responsible for comprehensively collecting multi-dimensional data related to the lifespan of core components during the operation of charging piles, including basic component information, operating status data, environmental data, and fault history data. Basic component information: Model, rated parameters (rated voltage, rated power, rated temperature range), production batch, and installation date of core components such as filter capacitors and power devices; Operating status data: Module switching count (cumulative value), temperature cycle data (number of complete cycles from low temperature to high temperature and back to low temperature each time, and the temperature extreme value of each cycle), output energy data (cumulative output power, output energy fluctuation value per unit time), voltage data (average input / output voltage, voltage fluctuation frequency, number of voltage surges), and current data (average input / output current, peak current, and current duration). Environmental data: ambient temperature, humidity, altitude, and dust concentration at the charging pile installation location (collected through integrated environmental sensors). Fault history data: past fault types of core components, operating parameters at the time of the fault, and maintenance / replacement records.
[0026] The data acquisition layer collects real-time operating status data through the CAN bus and RS485 interface of the charging pile controller, collects environmental data through integrated environmental sensors, and enters basic component information and fault history data through the operation and maintenance management platform. The data collection frequency is dynamically adjusted according to the data type. The data collection frequency for operational status is once per minute, the data collection frequency for environmental data is once per hour, and the data collection frequency for basic component information and fault history is entered and updated as needed.
[0027] The data preprocessing layer is responsible for cleaning, integrating, and performing feature engineering on the collected raw data to provide high-quality data for model training. The data preprocessing layer process is as follows: (1) Data cleaning First, outliers (such as data that exceeds the reasonable range due to sensor failure) are removed. Then, linear interpolation is used to fill in the missing values (a small number of missing data are filled using linear interpolation, while a large number of missing outlier samples are directly removed), and duplicate data are deleted. (2) Data integration Multi-dimensional data of the same core component belonging to the same charging pile are linked and integrated according to the time axis to form a structured dataset indexed by component-time. (3) Feature Engineering Calculate and extract statistical features, explore potential correlations between parameters, construct derived features, and map all feature data to the same scale to avoid the model training effect being affected by differences in feature dimensions.
[0028] The feature engineering implementation process is as follows: a) Basic Feature Extraction Calculate and extract statistical features, including the number of temperature cycles, the cumulative number of switching cycles, the cumulative output energy, the voltage fluctuation coefficient, and the average peak current. b) Derivative Feature Construction We construct temperature-on / off cycle interaction characteristics, output energy-temperature correlation characteristics, and ambient humidity-aging rate characteristics to explore potential correlations among parameters. c) Feature standardization The Z-score normalization method is used to map all feature data to the same scale, so as to avoid the model training effect being affected by the difference in feature dimensions.
[0029] The model training layer is responsible for training the remaining life prediction model of core components based on the preprocessed dataset using machine learning algorithms and optimizing the model parameters. The model training layer uses machine learning algorithms to train the remaining life prediction model for core components, and the steps are as follows: (1) Dataset partitioning Divide the structured dataset into a training set (for model training), a validation set (for model parameter tuning), and a test set (for final model evaluation) in a custom ratio (e.g., 7:2:1). (2) Model selection and initialization We selected a hybrid model combining Gradient Boosting Tree (XGBoost) and Long Short-Term Memory (LSTM) as the core prediction model and initialized the model hyperparameters (such as the tree depth and learning rate of XGBoost, and the number of hidden layer neurons and iterations of LSTM). Among them, Gradient Boosting Tree (XGBoost) is used to capture nonlinear statistical associations between features, and Long Short-Term Memory (LSTM) network is used to learn long-term dependencies in time series data (such as the cumulative effect of component aging). (3) Model training The training set is input into the hybrid model, the predicted value is calculated through forward propagation, the loss function is calculated based on the true remaining lifespan label, the model parameters are iteratively updated through the backpropagation algorithm, and the loss function is minimized. (4) Model evaluation The mean absolute error (MAE), root mean square error (RMSE), and prediction accuracy were used as evaluation metrics. Custom thresholds for the evaluation metrics were defined (e.g., the error between the predicted remaining lifetime and the actual value was considered to be within ±10% as accurate prediction). The model performance was evaluated using validation and test sets. (5) Parameter tuning If the model evaluation fails to meet the target (e.g., prediction accuracy is below 85%), the hyperparameters are tuned using a grid search method, and the training-evaluation process is repeated until the model performance meets the target.
[0030] During model training, the model training layer uses the difference between the actual operating time of the component and its rated life as the label of the true remaining life, and uses the mean squared error (MSE) as the loss function.
[0031] The prediction application layer is responsible for deploying the trained model to the charging pile operation and maintenance management platform to achieve real-time prediction of remaining lifespan and operation and maintenance guidance.
[0032] The prediction application layer implements real-time prediction of remaining lifetime and operation and maintenance guidance, as follows: (1) Real-time data acquisition and input The data acquisition layer collects the latest operational and environmental data in real time and transmits it to the trained model; (2) Data preprocessing The data preprocessing layer is responsible for cleaning, integrating, and feature engineering the collected real-time data. (3) Model prediction The preprocessed real-time data is input into the deployed model, and the model outputs the remaining life prediction value (unit: months) and prediction confidence level (e.g., 92%) of the core components. (4) Operation and maintenance decision generation Based on the remaining lifespan prediction results, a threshold is set to generate a reminder message, and combined with the charging pile's location and usage frequency information, a customized and optimized operation and maintenance resource scheduling plan is developed. When the remaining lifespan is ≤6 months, a priority maintenance reminder is generated; When the remaining lifespan is 6-12 months, generate a planned maintenance reminder; When the remaining life expectancy is greater than 12 months, generate routine monitoring recommendations; (5) Terminal push The forecast results and maintenance suggestions will be pushed to relevant maintenance personnel through the charging pile maintenance management platform or mobile APP terminal to guide them in carrying out targeted maintenance work.
[0033] Compared with existing technologies, this system for predicting the remaining lifespan of core components of charging piles based on big data has the following characteristics: (1) More comprehensive data dimensions Existing technologies typically collect only one or a few operating parameters, while this invention constructs a multi-dimensional data collection system that covers basic component information, operating status data, environmental data, and fault history data. This system can more comprehensively reflect the aging status of core components and lay the foundation for accurate prediction.
[0034] (2) Higher prediction accuracy Existing technologies use fixed thresholds or simple statistical analysis, which cannot achieve quantitative prediction. However, this invention uses a hybrid machine learning model that combines XGBoost and LSTM, which can uncover nonlinear correlations and time series dependencies between data. The remaining life prediction accuracy can reach more than 85%, and the prediction error is controlled within ±10%, which is far higher than the early warning accuracy of existing technologies.
[0035] (3) More versatile The fixed thresholds and analysis logic of existing technologies cannot adapt to different scenarios. This invention, through big data training and feature engineering, enables the model to learn the aging patterns of components under different environments and usage frequencies, making it widely adaptable to various charging pile devices and significantly improving its versatility.
[0036] (4) Better operation and maintenance guidance Existing technologies can only provide fault warnings but cannot provide advance planning time windows. This invention can quantitatively predict the remaining lifespan and generate targeted maintenance suggestions such as "priority maintenance" and "planned maintenance" to help maintenance personnel rationally allocate resources, reduce maintenance costs, reduce charging pile downtime, and improve service quality.
[0037] The embodiments described above are merely one specific implementation of the present invention. Ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.
Claims
1. A system for predicting the remaining lifespan of core components of charging piles based on big data, characterized in that: It includes a data acquisition layer, a data preprocessing layer, a model training layer, and a prediction application layer; The data acquisition layer is responsible for comprehensively collecting multi-dimensional data related to the lifespan of core components during the operation of charging piles, including basic component information, operating status data, environmental data, and fault history data. The data preprocessing layer is responsible for cleaning, integrating, and performing feature engineering on the collected raw data to provide high-quality data for model training. The model training layer is responsible for training the remaining life prediction model of core components based on the preprocessed dataset using machine learning algorithms and optimizing the model parameters. The prediction application layer is responsible for deploying the trained model to the charging pile operation and maintenance management platform to achieve real-time prediction of remaining lifespan and operation and maintenance guidance.
2. The system for predicting the remaining lifespan of core components of charging piles based on big data, as described in claim 1, is characterized in that: The data acquisition layer collects real-time operating status data through the CAN bus and RS485 interface of the charging pile controller, collects environmental data through integrated environmental sensors, and enters basic component information and fault history data through the operation and maintenance management platform. The data collection frequency is dynamically adjusted according to the data type. The data collection frequency for operational status is once per minute, the data collection frequency for environmental data is once per hour, and the data collection frequency for basic component information and fault history is entered and updated as needed.
3. The system for predicting the remaining lifespan of core components of charging piles based on big data as described in claim 1, characterized in that: The data preprocessing layer process is as follows: (1) Data cleaning First, outliers are removed, then missing values are filled using linear interpolation, and duplicate data are deleted. (2) Data integration Multi-dimensional data of the same core component belonging to the same charging pile are linked and integrated according to the time axis to form a structured dataset indexed by component-time. (3) Feature Engineering Calculate and extract statistical features, explore potential correlations between parameters, construct derived features, and map all feature data to the same scale to avoid the model training effect being affected by differences in feature dimensions.
4. The system for predicting the remaining lifespan of core components of charging piles based on big data, as described in claim 3, is characterized in that: The feature engineering implementation process is as follows: a) Basic Feature Extraction Calculate and extract statistical features, including the number of temperature cycles, the cumulative number of switching cycles, the cumulative output energy, the voltage fluctuation coefficient, and the average peak current. b) Derivative Feature Construction We construct temperature-on / off cycle interaction characteristics, output energy-temperature correlation characteristics, and ambient humidity-aging rate characteristics to explore potential correlations among parameters. c) Feature standardization The Z-score normalization method is used to map all feature data to the same scale, so as to avoid the model training effect being affected by the difference in feature dimensions.
5. The system for predicting the remaining lifespan of core components of charging piles based on big data as described in claim 1, characterized in that: The model training layer uses machine learning algorithms to train the remaining life prediction model for core components, and the steps are as follows: (1) Dataset partitioning Divide the structured dataset into training, validation, and test sets according to a custom ratio; (2) Model selection and initialization A hybrid model combining Gradient Boosting Tree (XGBoost) and Long Short-Term Memory (LSTM) network was selected as the core prediction model, and the model hyperparameters were initialized. Among them, Gradient Boosting Tree (XGBoost) is used to capture non-linear statistical associations between features, and Long Short-Term Memory Network (LSTM) is used to learn long-term dependencies in time series data. (3) Model training The training set is input into the hybrid model, the predicted value is calculated through forward propagation, the loss function is calculated based on the true remaining lifespan label, the model parameters are iteratively updated through the backpropagation algorithm, and the loss function is minimized. (4) Model evaluation The mean absolute error (MAE), root mean square error (RMSE), and prediction accuracy were used as evaluation metrics. The thresholds for these evaluation metrics were customized, and the model performance was evaluated using validation and test sets. (5) Parameter tuning If the model evaluation fails to meet the target, the hyperparameters are tuned using a grid search method, and the training-evaluation process is repeated until the model performance meets the target.
6. The system for predicting the remaining lifespan of core components of charging piles based on big data, as described in claim 5, is characterized in that: The model hyperparameters include the tree depth and learning rate of the gradient boosting tree XGBoost, and the number of hidden layer neurons and the number of iterations of the long short-term memory network LSTM.
7. The system for predicting the remaining lifespan of core components of charging piles based on big data, as described in claim 5, is characterized in that: During model training, the model training layer uses the difference between the actual operating time of the component and its rated life as the label of the true remaining life, and uses the mean squared error (MSE) as the loss function.
8. The system for predicting the remaining lifespan of core components of charging piles based on big data according to claim 1, characterized in that: The prediction application layer implements real-time prediction of remaining lifetime and operation and maintenance guidance, as follows: (1) Real-time data acquisition and input The data acquisition layer collects the latest operational and environmental data in real time and transmits it to the trained model; (2) Data preprocessing The data preprocessing layer is responsible for cleaning, integrating, and feature engineering the collected real-time data. (3) Model prediction The preprocessed real-time data is input into the deployed model, and the model outputs the remaining lifetime prediction value and prediction confidence level of the core components. (4) Operation and maintenance decision generation Based on the remaining lifespan prediction results, a threshold is set to generate a reminder message, and combined with the charging pile's location and usage frequency information, a customized and optimized operation and maintenance resource scheduling plan is developed. When the remaining lifespan is ≤6 months, a priority maintenance reminder is generated; When the remaining lifespan is 6-12 months, generate a planned maintenance reminder; When the remaining life expectancy is greater than 12 months, generate routine monitoring recommendations; (5) Terminal push The forecast results and maintenance suggestions will be pushed to relevant maintenance personnel through the charging pile maintenance management platform or mobile APP terminal to guide them in carrying out targeted maintenance work.