A fault early warning method for DC charging piles based on soft classification

Abstract

This invention relates to a soft-classification-based fault early warning method for DC charging piles, comprising: acquiring and using current, voltage, and SOC sequence data of the DC charging pile during each charging process as independent sample information; calculating a power difference sequence based on the current and voltage sequences; and using the power difference sequence and the corresponding SOC sequence as the charging features of the independent sample information; using the charging features and basic information of the DC charging pile as input features, loading them into a fault identification model integrating random forest, extreme gradient boosting, and K-nearest neighbor algorithms; and using a soft-classification method to calculate the probability values ​​of the charging pile's charging features belonging to various fault states, thereby providing fault early warning. Compared with existing technologies, this invention can provide early warning for DC charging piles that are about to fail, with satisfactory accuracy and recall. This method has certain practical value for the operation and maintenance of DC charging piles.

Description

Technical Field

[0001] This invention relates to the field of DC charging pile fault early warning technology, and in particular to a DC charging pile fault early warning method based on soft classification. Background Technology

[0002] The development of electric vehicles is of great significance to promoting energy conservation, emission reduction and technological innovation in the industry. The development of electric vehicles is inseparable from the construction of infrastructure, and charging piles, as charging equipment for electric vehicles, have become a hot research topic.

[0003] Faults are inevitable after charging stations are deployed. Compared to AC charging stations, DC charging stations have a more complex structure and higher charging power, making them more prone to malfunctions, and the causes of these malfunctions are more complex. [2] Since the service rate of DC charging piles is much higher than that of AC charging piles, a malfunction of a DC charging pile will have a negative impact on a large number of users. Therefore, the service quality of DC charging piles has become a key concern for electric vehicle users and power grid companies.

[0004] An effective way to ensure the reliability of charging stations is to improve fault diagnosis technology, thereby quickly identifying faulty charging stations and locating the fault point. Currently, the focus is on how to plan and deploy charging stations. [3] Optimization of charge and discharge control [4] Charging load forecast [5] There is a lot of research on charging piles, but research on how to ensure the reliability of charging piles is relatively lacking. The operation and maintenance of charging piles still relies on the experience of experts, resulting in low operation and maintenance efficiency.

[0005] In the field of charging pile fault diagnosis, some literature has proposed models for fault diagnosis and location: The literature "Optimization Configuration Model of Electric Vehicle Charging Pile Fault Location System" (Yuan Yaning, Li Lin. Electrical Automation, 2021, 43(2):41-43,46.) constructs a charging pile fault location model based on integer linear programming. This model locates the branch where the faulty charging pile is located based on the disturbance power generated when the charging pile fails; The literature "Configuration and Location Method of Charging Pile Fault Location System Based on Binary Particle Swarm Optimization" (Yi Zhonglin, Ju Hanji, Wang Jie, et al. Electrical Measurement & Instrumentation, 2021, 58(2):139-145.) proposes a solution method based on binary particle swarm optimization algorithm based on the literature "Optimization Configuration Model of Electric Vehicle Charging Pile Fault Location System" (Yuan Yaning, Li Lin. Electrical Automation, 2021, 43(2):41-43,46.), which improves the model performance; The literature "A DC Charging Pile Fault Diagnosis Method Based on Deep Neural Network" (Zhao Xiang, Chen Liangliang, Li Mingzhen, The paper "Research on Fault Diagnosis of Charging Piles Based on Hidden Generation and Extinction Process Model" (Lin Yue, Liu Tingzhang, Pan Jian, et al. Journal of Hefei University of Technology (Natural Science Edition), 2019, 42(4): 486-492) uses the hidden generation and extinction process model and the hidden Markov model to identify charging pile faults. The simulation results show that the hidden generation and extinction process model has a faster fault identification speed. The paper "Fault Diagnosis of Charging Piles Based on AP-HMM Hybrid Model" (Lin Yue, Liu Tingzhang, Chen Yifan, et al. Journal of Guangxi Normal University (Natural Science Edition), 2018, 36(1): 25-33) combines similarity clustering with hidden Markov model to construct a fault diagnosis model for charging piles, which improves the diagnostic accuracy of the model to a certain extent. The aforementioned model uses artificial intelligence algorithms to locate faults, but its fault diagnosis types are relatively limited. The literature "Research on Fault Diagnosis Expert System for DC Charging Piles Based on Fault Tree" (Yang Shasha, Beijing Jiaotong University, 2019) provides a detailed classification of fault types for DC charging piles and establishes a fault tree-based expert system for DC charging pile fault diagnosis based on fault phenomena and causes. However, the construction and analysis of the fault tree still require manual intervention, and research on fault prediction is insufficient. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology, such as high frequency and variety of faults in DC charging piles, difficulty in manual inspection, and untimely maintenance, and to provide a fault early warning method for DC charging piles based on soft classification.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A fault early warning method for DC charging piles based on soft classification includes the following steps:

[0009] The current, voltage, and SOC sequence data of the DC charging pile during each charging process are acquired and used as independent sample information. The power difference sequence is calculated based on the current sequence and voltage sequence, and the power difference sequence and the corresponding SOC sequence are used as the charging features of the independent sample information.

[0010] The charging characteristics and basic information of the DC charging pile are used as input features and loaded into a pre-established and trained fault identification model to obtain the probability value of each type of fault of the DC charging pile, thereby performing fault warning. The fault identification model is an integrated model including multiple sub-classifiers.

[0011] Furthermore, the ensemble model includes a random forest sub-classifier, a KNN sub-classifier, and an XGBoost sub-classifier. Each sub-classifier calculates the probability of each type of fault occurring in the input feature. The ensemble model weights and sums the probability calculations output by each classifier, and uses the fault type with the highest summation result as the judgment result of the ensemble model.

[0012] Furthermore, by comparing the probability value of each type of fault occurring in the DC charging pile with a pre-set warning threshold, if the probability value is greater than the warning threshold, a fault warning is issued; otherwise, no fault warning is issued.

[0013] Furthermore, the process of obtaining the warning threshold involves calculating the probability values ​​of various faults by fitting normal samples of fault-free charging piles through a model, thereby solving for the fault threshold. The expression for solving the fault threshold is as follows:

[0014]

[0015] In the formula, P A To help the model identify the probability value of a sample experiencing a type A fault, T A α represents the warning threshold for Class A faults, and α represents the significance level.

[0016] Furthermore, different warning thresholds are set for different fault types of the DC charging pile.

[0017] Furthermore, the evaluation metrics for the fault identification model during the training process include accuracy, recall, and F1-Score.

[0018] Furthermore, during the training process, the fault identification model is evaluated by calculating the average value and weighted average value of each evaluation index to determine the training effect of the fault identification model.

[0019] Furthermore, the power difference sequence is calculated by extracting the maximum current and voltage values ​​from the current and voltage sequences.

[0020] Furthermore, the distance weights of each feature in the KNN sub-classifier are all equal.

[0021] Furthermore, the basic information of the DC charging pile as an input feature includes the rated power and maximum output voltage of the DC charging pile.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] This invention addresses the frequent faults in DC charging piles by exploring the impact of charging power at the charging pile end on the charging efficiency of electric vehicles. It proposes using power difference sequences and vehicle-end SOC sequences as sample features to construct a DC charging pile fault early warning model based on a soft classification algorithm. This model integrates three classification algorithms—RF, KNN, and XGBoost—to achieve complementary advantages and can identify 12 frequently occurring fault types in DC charging piles. The model achieves a weighted average of 95% in terms of recognition accuracy, recall, and F1-Score, demonstrating optimal performance.

[0024] By fitting the model to predict the probability distribution function of the failure probability value of normal samples, a fault warning threshold with a significance level of 0.005 was designed. The case verified that the model can provide early warning of DC charging pile faults under the designed warning threshold and can accurately identify the specific fault type, giving maintenance personnel sufficient time to deal with the fault, thus demonstrating the early warning value of the model.

[0025] The proposed charging pile fault early warning method has theoretical value and practicality, and can provide a reference for the construction of intelligent charging pile maintenance platform. Attached Figure Description

[0026] Figure 1 This is a schematic diagram illustrating the changing trend of charging power and battery SOC over time, provided in an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the power curves of a DC charging pile charging batteries of different specifications provided in an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram illustrating the variation of charging power differential with SOC in an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of a DC charging pile fault identification model framework provided in an embodiment of the present invention;

[0030] Figure 5(a) is a first example diagram of an abnormal charging power differential curve provided in an embodiment of the present invention;

[0031] Figure 5(b) is a second example diagram of an abnormal charging power differential curve provided in an embodiment of the present invention;

[0032] Figure 5(c) is a third example of an abnormal charging power differential curve provided in an embodiment of the present invention;

[0033] Figure 6 This is a schematic diagram of the normal probability distribution and cumulative distribution of a normal sample provided in an embodiment of the present invention;

[0034] Figure 7 This is a schematic diagram of the cumulative failure probability distribution of a normal sample provided in an embodiment of the present invention;

[0035] Figure 8 This is a schematic diagram of a fault 1 early warning provided in an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0037] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0038] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0039] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed during use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0040] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0041] Furthermore, terms such as "horizontal" and "vertical" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0042] Example 1

[0043] This embodiment proposes a fault early warning method for DC charging piles based on soft classification, which includes the following steps:

[0044] The current, voltage, and SOC sequence data of the DC charging pile during each charging process are acquired and used as independent sample information. The power difference sequence is calculated based on the current sequence and voltage sequence, and the power difference sequence and the corresponding SOC sequence are used as the charging features of the independent sample information.

[0045] The charging characteristics and basic information of the DC charging pile are used as input features and loaded into a pre-established and trained fault identification model to obtain the probability value of each type of fault of the DC charging pile, thereby providing fault warning. The fault identification model is an integrated model that includes multiple sub-classifiers.

[0046] As an optional implementation, the ensemble model includes a random forest subclassifier, a KNN subclassifier, and an XGBoost subclassifier. Each subclassifier calculates the probability of each type of fault occurring in the input feature. The ensemble model weights and sums the probability calculations output by each classifier, and uses the fault type with the highest sum as the judgment result of the ensemble model.

[0047] As an optional implementation, the probability value of each type of fault occurring in the DC charging pile is compared with a pre-set warning threshold. If the probability value is greater than the warning threshold, a fault warning is issued; otherwise, no fault warning is issued.

[0048] Preferably, the warning threshold T AThe accuracy and lead time of early warnings are closely related: as the early warning threshold increases, the model's early warning accuracy increases, and the reliability of the early warning is higher; however, the lead time decreases as the early warning threshold increases, reducing the actual value of the early warning. Different types of faults have different characteristics, so early warning thresholds for different fault types should be set separately. This embodiment uses statistical methods to set the early warning threshold. By fitting normal samples of fault-free charging piles and calculating the probability values ​​of various faults through the model, the fault threshold is obtained. The expression for solving the fault threshold is:

[0049]

[0050] In the formula, PA is the probability value of the model identifying a sample as having a type A fault, TA is the warning threshold for type A faults, and α is the significance level.

[0051] As a preferred implementation, the evaluation metrics for the fault identification model during the training process include accuracy, recall, and F1-Score.

[0052] During the training process, the training effect of the fault identification model is judged by calculating the average value and weighted average value of each evaluation index.

[0053] More optimal implementation methods can be obtained by arbitrarily combining the above preferred implementation methods. The optimal implementation method obtained by combining all the implementation methods is described in detail below.

[0054] The DC charging pile fault early warning method based on soft classification in this embodiment includes: First, based on the power change of the charging pile during a single charging process, the stage characteristics of the charging process are mined. Then, a power difference sequence is constructed, and combined with the SOC sequence of the electric vehicle as the charging characteristics of the charging pile, a soft classification fault identification model based on "random forest + XGBoost + KNN" is adopted. This model calculates the probability value of the charging pile's charging characteristics before the fault occurs belonging to various fault states. When the probability of the fault occurs exceeds the early warning threshold, an early warning message is issued. Finally, a case study is conducted using vehicle network data to verify the effectiveness of the model, which is described in detail below.

[0055] 1. Analysis of charging power characteristics

[0056] The DC charging pile enters the charging state after receiving the charging request message sent by the electric vehicle. To ensure the normal progress of the entire charging process, the charging pile and the vehicle will monitor each other's status before the charging ends, and continuously send real-time status information such as charging current, voltage and current battery level to each other. The charging pile will adjust the current, voltage and other parameters in real time according to the interactive information.

[0057] Real-time charging current and voltage information from the charging pile can be used to calculate the real-time charging power, thus reflecting the charging pile's power output per unit time. Real-time SOC information from the vehicle reflects the electric vehicle's power input. During a normal charging process, the trends of the DC charging pile's charging power and battery SOC over time are as follows: Figure 1 As shown, the charging pile has a rated power of 60kW, a maximum output voltage of 500V, and a charging time from "2021-08-28 10:08:39" to "2021-08-28 12:28:02".

[0058] according to Figure 1 The charging process of DC charging piles has three different stages: Stage 1 corresponds to a vehicle-side SOC value of about 0% to 90%, during which the charging power at the pile end remains at a relatively high level and increases slowly; Stage 2 corresponds to a vehicle-side SOC value of generally between 90% and 95%, during which the charging power at the pile end drops rapidly from the highest value to a lower value; Stage 3 corresponds to a vehicle-side SOC approaching 100%, during which the charging power at the pile end remains at the lowest level.

[0059] For electric vehicles with different battery specifications, DC charging stations will provide different charging powers, such as... Figure 2 As shown, it is not very meaningful to directly use the charging power as the charging characteristic of DC charging piles. New characteristics need to be built on the basis of power.

[0060] Based on the three-stage characteristics of normal charging, the charging power curves of fault-free charging piles exhibit the same trend. A first-order difference is applied to the power value at each data acquisition time point to represent the rate of power change within a unit data acquisition interval. The charging power difference curve of a DC charging pile as a function of battery SOC is shown below. Figure 3 As shown: when the battery SOC is below 90%, the difference in charging power is not lower than 0. When the SOC exceeds about 90%, the difference in charging power decreases rapidly to a negative value. When the SOC approaches 100%, the difference in charging power fluctuates slightly around 0.

[0061] 2. Principle of soft classification algorithm

[0062] For each input sample, the soft-classification fault identification model outputs a set of values ​​corresponding to the probability of each type of fault occurring in that sample. This paper employs an ensemble method combining random forest, KNN, and XGBoost from the soft-classification model to construct a fault identification model for DC charging piles.

[0063] (1) Random Forest (RF)

[0064] Random Forest (RF) employs the Bagging technique, using only a subset of features from the samples to build each classifier during training, thus achieving classifier diversity. This method allows for the acquisition of classifiers with different focuses. In RF, each classifier is a single decision tree. By training multiple decision trees and summing their decisions through a voting process, the overall forest's results are obtained. Through Bagging, the classifiers in a Random Forest exhibit diverse performance structures, often resulting in a forest that outperforms a single decision tree, exhibiting lower overfitting risk and stronger model generalization ability.

[12] .

[0065] (2) K-Nearest Neighbor (KNN)

[0066] The KNN model divides a sample point into its K nearest neighbors by calculating the distance between them (usually Euclidean distance). While highly sensitive to population distribution, KNN's accuracy is limited when distinguishing between mixed populations. However, its simplicity and fast training speed make it perform well in simple, routine classification problems.

[12] However, the KNN algorithm requires manual setting of distance weights for different features, which is quite cumbersome. In this study, the distance weights for each feature are equal to simplify the model.

[0067] (3) Extreme Gradient Boosting (XGBoost)

[0068] XGBoost is an improvement upon the Gradient Boosting Decision Tree (GBDT) principle. GBDT trains multiple classifiers sequentially, each learning from and correcting errors made by the previous classifier. The final results are summed to form the overall model's decision. Unlike traditional GBDT, XGBoost expands the objective function to a second-order Taylor expansion instead of a first-order one. Furthermore, while GBDT finds new fitting labels for the next classifier layer, XGBoost finds a new objective function. Finally, XGBoost incorporates L2 regularization (squared regularization) into the sub-leaf weights. These improvements have resulted in XGBoost achieving superior performance compared to traditional GBDT in recent machine learning competitions.

[13] .

[0069] (4) Integration Model

[0070] Different classifier models process data differently, resulting in varying information gleaned from sample features. Ensemble models strategically combine the results of sub-classifiers as their output, integrating information from each classifier with their judgment criteria to achieve more accurate and reliable results. This paper employs an ensemble model based on a soft voting mechanism for DC charging pile fault identification: Random Forest, KNN, and XGBoost are used as sub-classifiers. Each sub-classifier calculates the probability of each type of fault occurring in the input sample. The ensemble model weights and sums the calculated probabilities of each classifier, using the highest sum as the fault type as the ensemble model's judgment.

[0071] 3. Model Framework and Evaluation Metrics

[0072] 3.1 Model Framework and Threshold Settings

[0073] The current, voltage, and SOC data recorded by the DC charging pile during each charging process are treated as independent sample information. The maximum current and voltage values ​​are extracted from the current and voltage sequences, and the power difference sequence is calculated. This sequence, along with the SOC sequence of that charging process, serves as the charging feature of the sample. This, combined with the rated power and maximum output voltage of the DC charging pile, are used as input features for the fault identification model. The model kernel employs a fault identification model based on an ensemble of "RF+KNN+XGBoost," which outputs the probability value for each type of fault. The framework of the DC charging pile fault identification model is as follows: Figure 4 As shown.

[0074] Warning threshold T A The accuracy and lead time of early warnings are closely related: as the early warning threshold increases, the model's early warning accuracy increases, and the reliability of the early warning is higher; however, the lead time decreases as the early warning threshold increases, reducing the actual value of the early warning. Different types of faults have different characteristics, so the early warning thresholds for different fault types should be set separately. This paper uses statistical methods to set the early warning thresholds. By fitting normal samples of fault-free charging piles and calculating the probability values ​​of various faults through the model, taking fault A as an example, the probability density function f(P) is solved. A The warning threshold is calculated using the following formula:

[0075]

[0076] In the formula: P A To help the model identify the probability value of a sample experiencing a type A fault, T A α is the warning threshold for Class A faults, and α is the significance level. Considering that the sample size is not large, this paper selects α as 0.005.

[0077] By comparing P A and T A Based on the size, the early warning criteria are as follows:

[0078]

[0079] 3.2 Model Evaluation Indicators

[0080] A confusion matrix is ​​a matrix used to statistically analyze the number of correct and incorrect results in a classification model test. Taking a binary classification task (containing two labels A and B) as an example, the confusion matrix structure is as follows:

[0081] Table 1. Binary Classification Confusion Matrix

[0082]

[0083] In the table, 'a' represents the number of samples whose true label is A and whose label is identified as A by the model.

[0084] In the table, b represents the number of samples whose true label is B and whose label is A as identified by the model;

[0085] In the table, 'c' represents the number of samples whose true label is A and whose model identifies the label as B.

[0086] In the table, d represents the number of samples whose true label is B and whose label is identified as B by the model;

[0087] To evaluate the performance of the DC charging pile fault identification model, accuracy, recall, and F1-score are used as evaluation metrics. Taking label A as an example, the calculation methods for each metric are as follows:

[0088]

[0089]

[0090]

[0091] For a fault label, Precision represents the reliability of the model's identification result, Recall refers to the proportion of times the model correctly identifies that type of label, and F1-Score integrates information from both Precision and Recall, representing the model's overall performance. The higher the values ​​of these three metrics, the better the model's performance.

[0092] Considering that the model identifies multiple fault labels, the Macro Average (numerical average) and Micro Average (weighted average) values ​​of each evaluation index are calculated as evaluation indicators of the fault identification model performance. The macro average can reflect the average performance of various faults on the same index, while the weighted average, because it combines the proportion information of various samples, can comprehensively reflect the model's performance on the entire dataset.

[0093] 4. Case Analysis

[0094] The experimental data came from historical information on 6,506 DC charging piles in Shanghai from the vehicle-to-everything (V2X) platform, including real-time data collection, fault information, and equipment information, covering the period from August 24, 2021 to November 16, 2021. The data collection recorded real-time voltage, real-time current, and real-time state of charge (SOC) data from the DC charging piles every 3 minutes. The fault information recorded the specific fault type, start time, and end time for each fault. The equipment information recorded the location, rated power, maximum output voltage, and manufacturer of the DC charging pile. These three types of information were linked together by the charging pile's serial number.

[0095] From the data sampled at three-minute intervals at the DC charging pile, Python programming was used to extract the data from each charging process. Real-time power was calculated based on real-time current and voltage, and then first-order differential processing was performed on the power to analyze the charging power of the DC charging pile. Figure 3 Comparison of normal charging pattern changes. Figures 5(a)-5(c) The abnormal charging power differential curve is shown.

[0096] During the charging process of a DC charging pile, the power difference value and SOC value corresponding to 30 sampling time points are uniformly selected as the power difference sequence and SOC sequence of the charging process, respectively. The maximum current, maximum voltage, power difference sequence, SOC sequence, rated power, and maximum allowable output voltage information of the DC charging pile in each charging process are used as the input features of a sample. The samples located within the fault start and end time periods are marked by fault information. When there is no relevant charging record within the fault time period, that is, the charging pile stops providing charging services due to fault, the charging process sample closest to the fault start time is marked as the fault sample.

[0097] According to statistics, more than 40 types of faults appeared in the data. After discussions among the staff of the power company, the fault types with higher frequency and greater importance were selected, resulting in twelve common fault types of DC charging piles as shown in Table 2.

[0098] Table 2 Fault Type Statistics

[0099]

[0100]

[0101] 8000 normal samples and the aforementioned faulty samples were randomly selected as training and testing data for the model. The training and testing sets were divided in a 7:3 ratio. To obtain unbiased model testing results, stratified sampling was used to ensure that the proportion of each class of samples in the training and testing sets was equal. Five-fold cross-validation was used during training, and the model parameters were tuned using a grid system. Under the premise of using the same dataset, the ensemble model proposed in this paper was compared with KNN, XGBoost, and RF. The training and testing results of each model are shown in Table 3.

[0102] Table 3 Model Performance Comparison

[0103]

[0104] The results show that the integrated model has higher accuracy and better performance than the fault identification model based on a single algorithm, verifying the effectiveness of the model in the fault diagnosis of charging piles.

[0105] Using an ensemble model, predictions were made for charging pile samples that had not experienced any faults. The model predicted the probability distribution and cumulative probability distribution of these normal samples not experiencing faults, as follows: Figure 6 As shown, the model predicts that 90% of normal samples have a probability of no fault occurrence exceeding 0.8, indicating that the model has high recognition accuracy.

[0106] The model predicts the cumulative probability distribution of various failures occurring in the above normal samples as follows: Figure 7 As shown, at positions where the model's predicted probability values ​​are low, the cumulative distribution of various faults approaches 100%, further validating the model's accuracy.

[0107] Taking fault 1 (charging module over-temperature alarm) as an example, the ensemble model based on soft classification predicts the probability value P1 of normal samples experiencing fault 1, fits the probability distribution of the probability value P1, and according to equation (1), the warning threshold T1 of fault 1 can be obtained as 0.429342638751654. The warning performance of the model under this warning threshold is tested using charging pile samples that have experienced fault 1: the charging pile experienced fault 1 at "2021-10-01 12:54:59", and had not experienced any faults before. The process of using the model to issue a warning for the charging pile is as follows. Figure 8 As shown.

[0108] Based on the integrated model prediction using a soft classification algorithm, the charging sample at "2021-09-26 01:06:04" was predicted by the model to have a probability value P1 of fault 1 exceeding the warning threshold T1, which was 5 days earlier than the actual fault occurred. Although there were several charging sessions in between that did not trigger the warning, the fluctuating upward trend of the fault probability value was obvious, which indicated that the fault had occurred. Therefore, it can be considered that the warning threshold setting was reasonable and had early warning capability.

[0109] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A soft classification-based direct current charging pile fault early warning method, characterized in that, The method comprises the following steps: obtaining and taking the current, voltage and SOC sequence data of the direct-current charging pile in each charging process as independent sample information, and calculating a power difference sequence according to the current sequence and the voltage sequence, and taking the power difference sequence and the corresponding SOC sequence as charging features of the independent sample information; taking the charging features and basic information of the direct-current charging pile as input features, and loading the input features into a pre-established and trained fault identification model to obtain a probability value of each type of fault of the direct-current charging pile, thereby performing fault early warning, wherein the fault identification model is an ensemble model comprising multiple sub-classifiers; the ensemble model comprises a random forest sub-classifier, a KNN sub-classifier and an XGBoost sub-classifier, each of which calculates the probability of each type of fault of the input features, and the ensemble model calculates the probability values output by each classifier by weighted summation, and takes the fault type with the highest summation result as the judgment result of the ensemble model; the probability value of each type of fault of the direct-current charging pile is compared with a pre-set early warning threshold value, if the probability value is greater than the early warning threshold value, fault early warning is performed, otherwise, no fault early warning is performed; the early warning threshold value is obtained by fitting the probability values of each type of fault calculated by the model from normal samples of a fault-free charging pile, thereby obtaining a fault threshold value, and the expression for solving the fault threshold value is: In the formula, P A is the probability value of the model identifying that the sample has a class A fault, T A is a warning threshold of the class A fault, and α is a significance level, is a probability density function.

2. The soft classification-based DC charging pile fault early warning method according to claim 1, characterized in that, different early warning threshold values corresponding to different fault types of the direct-current charging pile are provided.

3. The soft classification-based DC charging pile fault early warning method according to claim 1, characterized in that, the evaluation indexes of the fault identification model in the training process include accuracy, recall rate and F1-Score.

4. The soft classification-based DC charging pile fault early warning method according to claim 3, characterized in that, in the training process of the fault identification model, the training effect of the fault identification model is judged by calculating the average value and weighted average value of each evaluation index.

5. The soft classification-based DC charging pile fault early warning method according to claim 1, characterized in that, the maximum current and voltage values in the current sequence and the voltage sequence are extracted to calculate the power difference sequence.

6. The soft classification-based DC charging pile fault early warning method according to claim 1, characterized in that, the feature distance weights in the KNN sub-classifier are equal.

7. The soft classification-based DC charging pile fault early warning method according to claim 1, characterized in that, the basic information of the direct-current charging pile as the input features comprises the rated power and the maximum output voltage of the direct-current charging pile.

Images