Charging operation management platform based on data analysis

Abstract

The invention relates to the technical field of charging pile monitoring management, and discloses a charging operation management platform based on data analysis, and the platform comprises a data collection module which obtains and uploads the multi-source data of a corresponding charging pile in real time; the data preprocessing module is used for carrying out normalization processing on the multi-source data and filling missing values; the feature extraction and analysis module is used for extracting and acquiring statistical features of the multi-source data according to a time sequence based on a sliding window algorithm; the AI analysis module is used for outputting a fault probability and a residual life prediction value according to statistical characteristics by adopting a multi-task deep learning model optimized by transfer learning; the early warning analysis module is used for triggering graded early warning according to the fault probability, and is linked with the digital twin platform to generate a processing plan; and the log management module is used for recording an alarm event, a processing state and an equipment historical health degree curve, multi-dimensional data is subjected to cross validation through a multi-task deep learning model, potential faults of the charging pile can be recognized in real time, active early warning can be performed, and the operation and maintenance efficiency and safety are improved.

Description

Technical Field

[0001] The present application relates to the technical field of charging pile monitoring and management, and in particular to a charging operation management platform based on data analysis. Background Art

[0002] Electric vehicles use electricity instead of oil, have the advantages of zero emissions and significantly reduce traffic noise, and are one of the effective means to solve traffic environmental pollution, energy conservation and emission reduction problems, and alleviate the oil crisis. In recent years, with the increasing popularity of new energy vehicles, charging piles, as supporting products for new energy vehicles, have also gradually received attention.

[0003] The function of the charging pile is similar to that of a gas pump in a gas station. It can be fixed on the ground or on a wall and installed in public buildings (public buildings, shopping malls, public parking lots, etc.) and residential parking lots or charging stations. It can charge various types of electric vehicles according to different voltage levels. The input end of the charging pile is directly connected to the AC power grid, and the output end is equipped with a charging plug for charging electric vehicles.

[0004] Charging piles generally provide two charging methods: conventional charging and fast charging. People can use a specific charging card to swipe the card on the human-computer interaction interface provided by the charging pile to perform corresponding operations such as charging method, charging time, and fee data printing. It is a high-frequency, high-safety demand product, so there is an urgent need for a platform that can uniformly, efficiently, and reliably operate and manage charging piles.

[0005] However, the existing charging pile management platform has the following problems: Most of them use single parameter threshold alarms, which only rely on static threshold judgments of single parameters such as voltage and current. They cannot predict hidden faults such as battery thermal runaway or insulation aging, and the alarms are redundant and delayed. Traditional methods are sensitive to occasional anomalies, prone to false alarms, and lack a graded response mechanism for fault severity. In addition, the data is highly isolated, and the equipment status data, environmental data (temperature, humidity), and user behavior data are not integrated for analysis, making it difficult to explore fault correlations, and therefore difficult to respond to faults accurately and promptly. Summary of the invention

[0006] In order to provide a platform that can operate and manage charging piles more accurately, timely and reliably, this application provides a charging operation management platform based on data analysis, which adopts the following technical solutions: A charging operation management platform based on data analysis, comprising: A data acquisition module is used to acquire and upload multi-source data of the corresponding charging pile in real time; the multi-source data includes electrical parameters, environmental parameters, battery status and user behavior data; Data preprocessing module, used to normalize multi-source data and fill missing values; A feature extraction and analysis module, which extracts and obtains statistical features of the multi-source data in time series based on a sliding window algorithm; The AI ​​analysis module uses a multi-task deep learning model optimized by transfer learning to output the failure probability and remaining life prediction value according to the statistical features; An early warning analysis module is used to trigger a graded early warning according to the failure probability and link the digital twin platform to generate a processing plan; The log management module is used to record alarm events, processing status and historical health curve of equipment.

[0007] By adopting the above technical solution, the present application can cross-validate through multi-dimensional data by a multi-task deep learning model, map the warning information to a three-dimensional visualization platform, identify potential faults of charging piles in real time and actively issue warnings, thereby improving operation and maintenance efficiency and safety.

[0008] Optionally, the data acquisition module includes: A multi-type sensor group for collecting the electrical parameters, environmental parameters, battery status and user behavior data; The electrical parameters include active power and apparent power and the corresponding voltage and current data; the environmental parameters include charging gun tip temperature, charging pile internal temperature, charging pile internal humidity and smoke data; the battery status includes state of charge SOC and health state SOH and internal resistance; the user behavior data includes charging behavior information and payment information; The charging behavior information includes charging start time, duration, charging amount requirement, and number of plugging and unplugging; the payment information includes payment method, rate package selection, and refund record; The multi-type sensor group includes a current sensor that supports dynamic sampling rate switching, which is used to switch to a high-speed sampling mode when a graded warning is triggered.

[0009] Optionally, the data preprocessing module includes: The missing value processing module is used to fill the short-term lost data of the sensor using the linear difference method; the long-term lost data or abnormal power-off data of the sensor is marked as invalid segments and does not participate in subsequent processing and analysis; for data with different sampling rates, linear resampling is used to unify them into 10Hz time series data; among them, long-term lost data refers to data lost for 1 second or more, and short-term lost data refers to data lost for less than 1 second; The data normalization module is used to perform Z-score normalization processing on the electrical parameters, and scale the environmental parameters to the [0, 1] interval using the Min-Max method to simultaneously eliminate dimensional differences.

[0010] Optionally, the method for acquiring the statistical features includes: Get a specified number m of data points in a sliding window of dynamic time length; Calculate the corresponding statistical features according to the data points; the statistical features at least include mean, variance, skewness, kurtosis and Hurst index and multidimensional statistics of FFT energy spectra; The configuration of the dynamic time length is related to the Hurst exponent and must satisfy ; in, , is the re-calibrated range, n is the dynamic time length, , and Specify the upper and lower limits of the Hurst exponent threshold range for the preset.

[0011] By adopting the above technical solution, the Hurst index is a method for analyzing the autocorrelation and memory of time series data. Its core is to dynamically evaluate the persistence, anti-persistence or randomness characteristics of the sequence through a local window. If the statistical characteristics develop towards persistence, they will also develop towards greater than , indicating that the regularity is increasing. At this time, the window needs to be enlarged. By increasing the dynamic time length and reducing the sensitivity, the analysis frequency is reduced and computing power is saved. If the statistical characteristics develop in the direction of anti-persistence, it means that the regularity is increasing. , indicating that the regularity is reduced and the probability of abnormality is greater. At this time, the window needs to be reduced, and the analysis frequency and sensitivity can be increased by reducing the dynamic time length, so as to detect abnormal problems more timely.

[0012] Optionally, the training method of the multi-task deep learning model includes: Build multi-task deep learning models that contain shared layers and task-specific layers; The shared layer includes a 3-layer bidirectional long short-term memory network BiLSTM, and the task-specific layer includes a fault classification branch and a RUL regression branch; The Softmax layer of the fault classification branch outputs the fault probability , the fully connected layer of the RUL regression branch outputs the remaining lifetime ; , ; Training is performed on the data training set established based on the statistical features. Through domain adaptation technology, the pre-trained model of the multi-task deep learning model is migrated to a small sample data set of the target charging pile brand to perform domain adversarial training and optimize the generalization ability of the pre-trained model on cross-brand data.

[0013] By adopting the above technical solution, the model failure problem caused by data differences of multi-brand devices is solved through transfer learning + multi-task optimization. This application realizes cross-device universal monitoring through transfer learning.

[0014] Optionally, the training method of the multi-task deep learning model includes: Obtain historical time series data including charging voltage, battery surface temperature and ambient humidity, perform three-dimensional thermal runaway feature extraction and calculation on the historical time series data based on a sliding window algorithm, and extract and calculate the charging voltage fluctuation variance and the temperature rise rate of the battery surface temperature and the maximum ambient humidity within 5 seconds; The pre-trained model is trained according to the three-dimensional thermal runaway characteristics to obtain the multi-task deep learning model.

[0015] Optionally, the training method of the multi-task deep learning model includes: When the data of new brand equipment reaches 1,000, incremental training is started; The incremental training includes: Freeze the shared layers of BiLSTM and only fine-tune the task-specific layer parameters; ; in, is the new total loss function, The loss of incremental training, is the balance control coefficient, is the model parameter The Fisher information value of is the optimal model parameter before starting the i-th incremental training.

[0016] Optionally, the early warning analysis module calculates the risk score based on the failure probability and the remaining life value ; The risk score The calculation methods include: ; in, , are the weighted coefficients of the corresponding items, is the maximum lifespan; If the risk score exceeds the first threshold, a local pop-up window will be displayed and a record will be sent to the log management module; If the risk score exceeds the second threshold, the device is triggered to limit power operation, the operation and maintenance personnel are notified, and the log management module is sent for record; If the risk score exceeds the third threshold, the power supply will be automatically cut off and the emergency handling process will be pushed to the digital twin platform, and the log management module records will be sent simultaneously.

[0017] By adopting the above technical solutions, human intervention can be reduced and the emergency response speed can be improved.

[0018] In summary, the present application includes at least one of the following beneficial technical effects: 1. This application can be cross-validated by a multi-task deep learning model through multi-dimensional data, and the warning information can be mapped to a three-dimensional visualization platform. It can identify potential faults of charging piles in real time and actively warn, thereby improving operation and maintenance efficiency and safety.

[0019] 2. This application is trained based on the data training set established based on the statistical features. Through domain adaptation technology, the pre-trained model of the multi-task deep learning model is migrated to a small sample data set of the target charging pile brand for domain adversarial training. The model failure problem caused by data differences among multi-brand devices is solved through transfer learning + multi-task optimization, and the generalization ability of the pre-trained model on cross-brand data is optimized. The sliding window is dynamically adjusted through the design of dynamic time length, taking into account both computing power energy saving and warning timeliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a module connection diagram of the charging operation management platform in this application. DETAILED DESCRIPTION

[0021] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings.

[0022] In the description of this specification, the description with reference to the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0023] The present application embodiment discloses a charging operation management platform based on data analysis, referring to Figure 1 ,include: A data acquisition module is used to acquire and upload multi-source data of the corresponding charging pile in real time; the multi-source data includes electrical parameters, environmental parameters, battery status and user behavior data; Data preprocessing module, used to normalize multi-source data and fill missing values; A feature extraction and analysis module, which extracts and obtains statistical features of the multi-source data in time series based on a sliding window algorithm; The AI ​​analysis module uses a multi-task deep learning model optimized by transfer learning to output the failure probability and remaining life prediction value according to the statistical features; An early warning analysis module is used to trigger a graded early warning according to the failure probability and link the digital twin platform to generate a processing plan; The log management module is used to record alarm events, processing status and historical health curve of equipment.

[0024] The multi-task deep learning model can be cross-validated through multi-dimensional data, and the warning information can be mapped to a three-dimensional visualization platform. It can identify potential faults of charging piles in real time and actively warn, thereby improving operation and maintenance efficiency and safety.

[0025] In this embodiment of the present application, the data acquisition module includes a multi-type sensor group, a communication unit and a time synchronization unit; the communication unit supports multi-mode communications such as 4G / 5G, LoRa, Ethernet, etc., and the time synchronization unit is based on the nanosecond clock synchronization module of GPS / Beidou. The edge gateway can be configured in the data preprocessing module. The edge gateway has a built-in FPGA chip, which is responsible for data preprocessing and protocol conversion, realizes local data preprocessing (such as filtering, feature extraction), reduces the pressure on cloud transmission bandwidth, and ensures the spatiotemporal consistency of multi-source data.

[0026] A multi-type sensor group for collecting the electrical parameters (voltage, current, power), environmental parameters (temperature, humidity, smoke), battery status (SOC, SOH) and user behavior data (charging time, frequency); The electrical parameters include active power and apparent power and the corresponding voltage and current data; the environmental parameters include charging gun tip temperature, charging pile internal temperature, charging pile internal humidity and smoke data; the battery status includes state of charge SOC and health state SOH and internal resistance; the user behavior data includes charging behavior information and payment information; The charging behavior information includes charging start time, duration, charging amount requirement, and number of plugging and unplugging; the payment information includes payment method, rate package selection, and refund record; The multi-type sensor group includes a current sensor that supports dynamic sampling rate switching, which is used to switch to a high-speed sampling mode when a graded warning is triggered. The abnormal trigger high-speed sampling mechanism takes into account both normal low power consumption and high-precision capture of emergency conditions.

[0027] Scenario: The cable of a DC fast charging pile (power 120kW) overheats during the charging process.

[0028] Sensor trigger: The temperature sensor detected that the temperature of the gun tip suddenly increased from 30°C to 85°C (sampling rate 10Hz); The current sensor synchronously captures the current fluctuation (1kHz sampling, fluctuation range ±20A).

[0029] Edge preprocessing: The edge gateway calculated that the temperature rise rate within the 5-second window was 11°C / s, exceeding the threshold of 5°C / s; Trigger the current sensor to switch to 10kHz high-speed sampling mode.

[0030] Data encapsulation and transmission: Package the temperature and current data into JSON format, append timestamp and hash value; Prioritize uploading to the cloud alarm queue via the 5G network.

[0031] Optionally, the data preprocessing module includes: The missing value processing module is used to fill the short-term lost data of the sensor using the linear difference method; the long-term lost data or abnormal power-off data of the sensor is marked as invalid segments and does not participate in subsequent processing and analysis; for data with different sampling rates, linear resampling is used to unify them into 10Hz time series data; among them, long-term lost data refers to data lost for 1 second or more, and short-term lost data refers to data lost for less than 1 second; The data normalization module is used to perform Z-score normalization processing on the electrical parameters, and scale the environmental parameters to the [0, 1] interval using the Min-Max method to simultaneously eliminate dimensional differences.

[0032] Optionally, the method for acquiring the statistical features includes: Get a specified number m of data points in a sliding window of dynamic time length; m can be 100, and the initial value of n can be 10s; Calculate the corresponding statistical features according to the data points; the statistical features at least include mean, variance, skewness, kurtosis and Hurst index and multidimensional statistics of FFT energy spectra; The configuration of the dynamic time length is related to the Hurst exponent and must satisfy ; in, , the base can be set to 10 by default, is the re-standard range, For the very poor, is the standard deviation, n is the dynamic time length, , and The upper and lower limits of the Hurst index threshold range are preset. In this embodiment of the present application, Configurable to 0.46, Configurable to 0.64.

[0033] In this embodiment of the present application, the Hurst index is a method for analyzing the autocorrelation and memory of time series data. The core of the method is to dynamically evaluate the persistence, anti-persistence or randomness characteristics of the sequence through a local window. If the statistical characteristics develop towards persistence, they will also develop towards greater than , indicating that the regularity is increasing. At this time, the window needs to be enlarged. By increasing the dynamic time length and reducing the sensitivity, the analysis frequency is reduced and computing power is saved. If the statistical characteristics develop in the direction of anti-persistence, it means that the regularity is increasing. , indicating that the regularity is reduced and the probability of abnormality is greater. At this time, the window needs to be reduced, and the analysis frequency and sensitivity can be increased by reducing the dynamic time length, so as to detect abnormal problems more timely.

[0034] Optionally, the training method of the multi-task deep learning model includes: Build multi-task deep learning models that contain shared layers and task-specific layers; The shared layer includes a 3-layer bidirectional long short-term memory network BiLSTM, and the task-specific layer includes a fault classification branch and a RUL regression branch; The Softmax layer of the fault classification branch outputs the fault probability , the fully connected layer of the RUL regression branch outputs the remaining lifetime ; , ; Training is performed on the data training set established based on the statistical features. Through domain adaptation technology, the pre-trained model of the multi-task deep learning model is migrated to a small sample data set of the target charging pile brand to perform domain adversarial training and optimize the generalization ability of the pre-trained model on cross-brand data.

[0035] The multi-task deep learning model is a technology that improves the generalization ability and training efficiency of the model by sharing knowledge between different tasks. Its core principles are the knowledge sharing mechanism and the regularization effect. The knowledge sharing mechanism refers to multi-task learning (MTL) that enables multiple tasks to share part of the network structure (such as the underlying feature extraction layer) so that the model can learn common representations from related tasks. This sharing mechanism is equivalent to imposing soft constraints on model parameters, which can effectively prevent overfitting and improve data utilization efficiency. The joint training of multiple tasks is equivalent to introducing additional regularization terms. The regularization effect indicates that multi-task learning can help the model better understand the data distribution, thereby alleviating the common problem of insufficient generalization ability in traditional deep learning.

[0036] In this embodiment of the present application, multiple tasks can share the same hidden layer (such as a convolutional layer / LSTM layer), and the output layer can be forked to process different tasks. Different tasks can also use independent networks, but similarity is encouraged through parameter distance constraints (such as L2 regularization). It is mentioned that the loss weight of each task can be automatically adjusted through homoscedastic uncertainty.

[0037] Through transfer learning + multi-task optimization, a multi-task deep learning model optimized by transfer learning is adopted. By combining the knowledge reuse capability of transfer learning and the joint training advantage of multi-task learning, the generalization performance of the model across tasks and fields can be significantly improved, and the model failure problem caused by data differences of multi-brand equipment can be solved.

[0038] Optionally, the training method of the multi-task deep learning model includes: Obtain historical time series data including charging voltage, battery surface temperature and ambient humidity, perform three-dimensional thermal runaway feature extraction and calculation on the historical time series data based on a sliding window algorithm, and extract and calculate the charging voltage fluctuation variance and the temperature rise rate of the battery surface temperature and the maximum ambient humidity within 5 seconds; The pre-trained model is trained according to the three-dimensional thermal runaway characteristics to obtain the multi-task deep learning model.

[0039] Optionally, the training method of the multi-task deep learning model includes: When the data of new brand equipment reaches 1,000, incremental training is started; The incremental training includes: Freeze the shared layers of BiLSTM and only fine-tune the task-specific layer parameters; ; in, is the new total loss function, The loss of incremental training, is the balance control coefficient, is the model parameter The Fisher information value of is the optimal model parameter before starting the i-th incremental training.

[0040] Optionally, the early warning analysis module calculates the risk score based on the failure probability and the remaining life value ; The risk score The calculation methods include: ; in, , are the weighted coefficients of the corresponding items, is the maximum lifespan; If the risk score exceeds the first threshold, a local pop-up window will be displayed and a record will be sent to the log management module; If the risk score exceeds the second threshold, the device is triggered to limit power operation, the operation and maintenance personnel are notified, and the log management module is sent for record; If the risk score exceeds the third threshold, the power supply will be automatically cut off and the emergency handling process will be pushed to the digital twin platform, and the log management module records will be sent simultaneously, which can fully reduce manual intervention and improve the emergency response speed.

[0041] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A charging operation management platform based on data analysis, characterized in that: include: A data acquisition module is used to acquire and upload multi-source data of the corresponding charging pile in real time; the multi-source data includes electrical parameters, environmental parameters, battery status and user behavior data; Data preprocessing module, used to normalize multi-source data and fill missing values; A feature extraction and analysis module, which extracts and obtains statistical features of the multi-source data in time series based on a sliding window algorithm; The AI ​​analysis module uses a multi-task deep learning model optimized by transfer learning to output the failure probability and remaining life prediction value according to the statistical features; An early warning analysis module is used to trigger a graded early warning according to the failure probability and link the digital twin platform to generate a processing plan; The log management module is used to record alarm events, processing status and historical health curve of equipment.

2. The charging operation management platform based on data analysis according to claim 1 is characterized in that: The data acquisition module comprises: A multi-type sensor group for collecting the electrical parameters, environmental parameters, battery status and user behavior data; The electrical parameters include active power and apparent power and the corresponding voltage and current data; the environmental parameters include charging gun tip temperature, charging pile internal temperature, charging pile internal humidity and smoke data; the battery status includes state of charge SOC and health state SOH and internal resistance; the user behavior data includes charging behavior information and payment information; The charging behavior information includes charging start time, duration, charging amount requirement, and number of plugging and unplugging; the payment information includes payment method, rate package selection, and refund record; The multi-type sensor group includes a current sensor that supports dynamic sampling rate switching, which is used to switch to a high-speed sampling mode when a graded warning is triggered.

3. The charging operation management platform based on data analysis according to claim 1, characterized in that: The data preprocessing module comprises: The missing value processing module is used to fill the short-term lost data of the sensor using the linear difference method; the long-term lost data or abnormal power-off data of the sensor is marked as invalid segments and does not participate in subsequent processing and analysis; for data with different sampling rates, linear resampling is used to unify them into 10Hz time series data; among them, long-term lost data refers to data lost for 1 second or more, and short-term lost data refers to data lost for less than 1 second; The data normalization module is used to perform Z-score normalization processing on the electrical parameters, and scale the environmental parameters to the [0, 1] interval using the Min-Max method to simultaneously eliminate dimensional differences.

4. The charging operation management platform based on data analysis according to claim 1, characterized in that: The method for obtaining the statistical features includes: Get a specified number m of data points in a sliding window of dynamic time length; Calculate the corresponding statistical features according to the data points; the statistical features at least include mean, variance, skewness, kurtosis and Hurst index and multidimensional statistics of FFT energy spectra; The configuration of the dynamic time length is related to the Hurst exponent and must satisfy ; in, , is the re-standard range, For the very poor, is the standard deviation, n is the dynamic time length, , and Specify the upper and lower limits of the Hurst exponent threshold range for the preset.

5. The charging operation management platform based on data analysis according to claim 4 is characterized in that: The training method of the multi-task deep learning model includes: Build multi-task deep learning models that contain shared layers and task-specific layers; The shared layer includes a 3-layer bidirectional long short-term memory network BiLSTM, and the task-specific layer includes a fault classification branch and a RUL regression branch; The Softmax layer of the fault classification branch outputs the fault probability , the fully connected layer of the RUL regression branch outputs the remaining lifetime ; , ; Training is performed on the data training set established based on the statistical features. Through domain adaptation technology, the pre-trained model of the multi-task deep learning model is migrated to a small sample data set of the target charging pile brand to perform domain adversarial training and optimize the generalization ability of the pre-trained model on cross-brand data.

6. The charging operation management platform based on data analysis according to claim 5, characterized in that: The training method of the multi-task deep learning model includes: Obtain historical time series data including charging voltage, battery surface temperature and ambient humidity, perform three-dimensional thermal runaway feature extraction and calculation on the historical time series data based on a sliding window algorithm, and extract and calculate the charging voltage fluctuation variance and the temperature rise rate of the battery surface temperature and the maximum ambient humidity within 5 seconds; The pre-trained model is trained according to the three-dimensional thermal runaway characteristics to obtain the multi-task deep learning model.

7. The charging operation management platform based on data analysis according to claim 6, characterized in that: The training method of the multi-task deep learning model includes: When the data of new brand equipment reaches 1,000, incremental training is started; The incremental training includes: Freeze the shared layers of BiLSTM and only fine-tune the task-specific layer parameters; ； in, is the new total loss function, The loss of incremental training, is the balance control coefficient, is the model parameter The Fisher information value of is the optimal model parameter before starting the i-th incremental training.

8. The charging operation management platform based on data analysis according to claim 5, characterized in that: The early warning analysis module calculates the risk score based on the failure probability and the remaining life value ; The risk score The calculation methods include: ； in, , are the weighted coefficients of the corresponding items, is the maximum lifespan; If the risk score exceeds the first threshold, a local pop-up window will be displayed and a record will be sent to the log management module; If the risk score exceeds the second threshold, the device is triggered to limit power operation, the operation and maintenance personnel are notified, and the log management module is sent for record; If the risk score exceeds the third threshold, the power supply will be automatically cut off and the emergency handling process will be pushed to the digital twin platform, and the log management module records will be sent simultaneously.

Images