Charging equipment fault prediction system based on multi-model integrated learning

Abstract

The invention discloses a charging equipment fault prediction system based on multi-model ensemble learning, and relates to the technical field of fault prediction, and the system comprises the steps: building an equipment attribute classifier, carrying out the attribute classification of target charging equipment, and obtaining the attribute parameters of the target equipment; constructing a fault prediction task list, obtaining an equipment fault task model set based on fault prediction task list transfer learning, obtaining a charging equipment fault data set, carrying out feature cascade analysis on the charging equipment fault data set, and obtaining an N-level equipment fault data graph; and according to the attribute parameters of the target equipment and the fault prediction precision requirement, matching and associating with the N-level equipment fault data atlas to obtain an equipment associated fault data set, constructing an equipment fault integrated prediction model, and performing fault prediction and early warning on the target charging equipment. The technical problem that the accuracy of fault prediction of the charging equipment is insufficient in the prior art is solved, and the technical effect of improving the accuracy of fault prediction of the charging equipment is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a charging equipment fault prediction system based on multi-model integrated learning. Background Art

[0002] With the continuous expansion of equipment use and the diversification of fault types, existing charging equipment fault prediction technologies face significant challenges. Traditional fault prediction methods mostly rely on a single model or experience-based rules, and usually have the problem of insufficient prediction accuracy when dealing with complex fault scenarios. For example, when faced with different fault types of charging equipment (such as battery failure, charging module failure, line problems, etc.), these methods are often unable to simultaneously and efficiently identify, resulting in a high rate of missed reports and a high rate of false alarms in fault prediction. In addition, due to changes in environmental factors, equipment aging and other factors, traditional models often lack good adaptability and generalization capabilities, and the prediction results are unstable under different equipment and different usage environments, making it difficult to meet the needs of modern charging facilities for efficient, accurate, and real-time early warnings. Summary of the invention

[0003] The present application provides a charging equipment fault prediction system based on multi-model integrated learning, which is used to solve the technical problem of insufficient accuracy of charging equipment fault prediction in the prior art.

[0004] In view of the above problems, the present application provides a charging equipment fault prediction system based on multi-model ensemble learning.

[0005] The present application provides a charging equipment fault prediction system based on multi-model integrated learning, the system comprising: A classifier building module is used to build a device attribute classifier, classify the attributes of the target charging device based on the device attribute classifier, and obtain the target device attribute parameters; a task list building module is used to build a fault prediction task list according to the charging device fault prediction target, and obtain the device fault task model set based on the transfer learning of the fault prediction task list; an analysis module is used to mine and obtain the charging device fault data set, perform feature cascade analysis on the charging device fault data set, and obtain an N-level device fault data map; a matching and association module is used to match and associate with the N-level device fault data map according to the target device attribute parameters and fault prediction accuracy requirements, and obtain a device-associated fault data set; an early warning module is used to perform integrated learning and training on the device fault task model set based on the device-associated fault data set, build an equipment fault integrated prediction model, and perform fault prediction and early warning on the target charging device through the equipment fault integrated prediction model.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application builds a device attribute classifier, classifies the attributes of the target charging device based on the device attribute classifier, and obtains the target device attribute parameters; constructs a fault prediction task list according to the charging device fault prediction target, and obtains the device fault task model set based on the transfer learning of the fault prediction task list; mines and obtains the charging device fault data set, performs feature cascade analysis on the charging device fault data set, and obtains an N-level device fault data map; matches and associates with the N-level device fault data map according to the target device attribute parameters and the fault prediction accuracy requirements, and obtains a device-associated fault data set; performs integrated learning and training on the device fault task model set based on the device-associated fault data set, constructs an equipment fault integrated prediction model, and performs fault prediction and early warning on the target charging device through the equipment fault integrated prediction model. The present invention solves the technical problem of insufficient accuracy in the prior art in predicting charging equipment faults. It classifies the attributes of target equipment by constructing an equipment attribute classifier, obtains an equipment fault task model set by using a fault prediction task list and transfer learning, and constructs an equipment fault data map through feature cascade analysis. Combined with the target equipment attribute parameters and the fault prediction accuracy requirements, an equipment associated fault data set is obtained through association matching, and finally an equipment fault integrated prediction model is constructed through integrated learning to predict and warn charging equipment faults, thereby achieving the technical effect of improving the accuracy of charging equipment fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of the structure of a charging equipment fault prediction system based on multi-model integrated learning provided in an embodiment of the present application; Figure 2 A schematic diagram of the process of building a device attribute classifier for a charging equipment fault prediction system based on multi-model ensemble learning provided in an embodiment of the present application.

[0009] Explanation of the accompanying drawings: classifier building module 11, task list building module 12, analysis module 13, matching association module 14, early warning module 15. DETAILED DESCRIPTION

[0010] The present application provides a charging equipment fault prediction system based on multi-model integrated learning, aiming to solve the technical problem of insufficient accuracy of charging equipment fault prediction in the prior art. It classifies the attributes of target equipment by constructing a device attribute classifier, obtains a set of device fault task models by using a fault prediction task list and transfer learning, and constructs a device fault data map through feature cascade analysis. Combined with the target equipment attribute parameters and fault prediction accuracy requirements, a device associated fault data set is obtained through association matching. Finally, a device fault integrated prediction model is constructed through integrated learning, and fault prediction and early warning are performed on the charging equipment, thereby achieving the technical effect of improving the accuracy of charging equipment fault prediction.

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0012] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0013] Examples, such as Figure 1 As shown, the embodiment of the present application provides a charging equipment fault prediction system based on multi-model integrated learning, the system comprising: The classifier building module 11 is used to build a device attribute classifier, and classify the attributes of the target charging device based on the device attribute classifier to obtain the target device attribute parameters.

[0014] In an embodiment of the present application, the classifier building module 11 is used to build a device attribute classifier to classify the attributes of the charging device. Specifically, first define the key attribute factor information of the charging device, including application scenarios, device specifications, and charging methods; then, collect data from the charging device database, and classify the devices based on these defined attribute factors to obtain the attribute parameter set of the charging device. Next, build a coding system to encode the device attribute information and parameter sets, and convert them into digital data that can be processed by the machine learning algorithm. Finally, use the decision tree algorithm to train and tune the encoded data to build a device attribute classifier.

[0015] After the device attribute classifier is built, the device is automatically judged and classified according to the input data of the target charging device, and the attribute parameters of the target device are finally output. The input data of the target charging device includes the model, rated power, input voltage, current specification, charging interface type, etc. of the device. These data are input into the trained device attribute classifier. The classifier classifies the attributes of the target device according to the decision rules learned by the model and outputs the corresponding target device attribute parameters.

[0016] Further, such as Figure 2 As shown, in the system provided by the application embodiment, the classifier building module 11 is also used for: Define charging equipment attribute factor information, wherein the charging equipment attribute factor information includes application scenarios, equipment specifications, and charging methods; collect and obtain a charging equipment database, and classify and identify the attributes of each device data in the charging equipment database based on the charging equipment attribute factor information to obtain a charging equipment attribute parameter set; build a coding system, and based on the coding system, perform knowledge coding on the charging equipment attribute factor information and the charging equipment attribute parameter set to obtain a device attribute coding parameter set; use a decision tree algorithm to classify, train, and optimize the device attribute coding parameter set to build the device attribute classifier.

[0017] In the embodiment of the present application, the attribute factor information of the charging device is first defined, including key features such as application scenarios, device specifications and charging methods. The application scenario refers to the use environment of the charging device, such as a home charging station, a public charging pile or an electric bus charging station. This information is obtained from the charging station operation database or the device management system; the device specifications involve the rated power, voltage, current specifications, interface type and other parameters of the device, which are extracted from the device technical manual or the device management system; the charging method refers to the charging mode supported by the device, such as DC fast charging, AC slow charging or wireless charging, which can be distinguished by the device nameplate or standard technical protocol.

[0018] Next, collect and obtain data from the charging device database. The charging device database contains a large amount of technical attribute information of charging devices. Relevant data is obtained from the database through SQL query or other data extraction methods. For example, the power, interface type, charging method and other information of the device are extracted, and the devices are divided into different categories such as "DC fast charging" and "AC slow charging" by using rule matching methods, thereby forming a charging device attribute parameter set.

[0019] Then build a coding system to encode the attribute information of the device so that it is suitable for machine learning models. For discrete variables such as charging methods and application scenarios, the unique hot encoding method is used to convert features of different categories into binary vectors. For example, "DC fast charging" can be represented as [1,0,0], while "AC slow charging" is represented as [0,1,0]. For continuous variables such as device power and current, the Z-score standardization method is used to convert them into standardized data with a mean of 0 and a standard deviation of 1 to reduce the scale effect between different features. The encoded data is integrated to form a set of device attribute encoding parameters.

[0020] Next, the decision tree algorithm is used to classify and train the device attribute encoding parameter set and tune it. During the training process, the decision tree recursively divides the data set and selects an optimal feature for splitting. Here, information gain can be used as the criterion for selecting split features, and features that can maximize the classification effect, such as power, charging method, etc., are selected. The decision tree model will continue to split on the training data and build a tree structure until the stopping condition is met (such as the depth of the tree reaches a preset value or the samples in the node are of the same category). To prevent overfitting, the decision tree uses a post-pruning strategy to remove unimportant branches, simplify the model and improve its generalization ability. During the tuning process, the hyperparameters of the tree, such as the maximum depth and the minimum number of samples, are adjusted to ensure that the model achieves the best performance on the training data and can adapt well to unseen test data. Through this series of steps, a device attribute classifier is finally constructed. The classifier receives the attribute input of the target charging device (such as power, current, charging method, etc.), automatically classifies the device attributes, and outputs the device attribute parameters, such as device category (fast charging or slow charging), applicable scenario (public or home), etc.

[0021] The task list construction module 12 is used to construct a fault prediction task list according to the charging equipment fault prediction target, and obtain a device fault task model set based on the fault prediction task list through transfer learning.

[0022] In an embodiment of the present application, the task list construction module 12 first constructs a fault prediction task list based on the charging equipment fault prediction target. The charging equipment fault prediction target is pre-set by the equipment manufacturer or the operation and maintenance team, including abnormal charging power, excessive equipment temperature, and interface wear or damage. Next, based on these fault prediction targets, a fault prediction task list is constructed, which contains all fault types that need to be monitored and predicted. Each task will describe in detail the prediction target, fault type, fault level, and predicted probability of occurrence of the fault in the list. For example, for the power abnormality prediction task, the task goal is to predict whether the power fluctuation during charging exceeds the normal range, and set different levels for the fault according to the fluctuation amplitude, such as minor, medium, and severe; for the temperature over-high prediction task, it predicts whether the device temperature exceeds the safety threshold, and sets the fault level according to the temperature change, such as normal, warning, and critical; and the battery fault prediction task predicts the health of the battery through data such as battery voltage, current, and number of charging times, and outputs the probability of fault occurrence. Through this process, the construction of the fault prediction task list is completed.

[0023] Then, according to the fault prediction task list, a source domain fault task analysis dataset is selected, which contains labeled equipment fault task data. Then, the model is pre-trained using the source domain dataset to generate a source domain fault task analysis model set. These models reflect the fault prediction capabilities learned from similar devices or environments. Then, the target domain for fault analysis of the charging equipment is determined according to the task list, and the source domain model is mapped and migrated to the target domain. By analyzing the characteristics and requirements of the target device, a corresponding migration fine-tuning strategy is formulated. Finally, the source domain model is fine-tuned for transfer learning according to the migration fine-tuning strategy, and its parameters and structure are adjusted to adapt to the fault prediction requirements of the target device, thereby generating an accurate set of equipment fault task models.

[0024] Furthermore, in the system provided in the application embodiment, the task list construction module 12 is also used for: According to the fault prediction task list, a source domain fault task analysis dataset is selected, and the source domain fault task analysis dataset includes a fault task analysis dataset of labeled devices; the source domain fault task analysis dataset is used to perform model pre-training to obtain a source domain device fault task analysis model set; according to the fault prediction task list, a charging device fault analysis target domain is determined, and the source domain device fault task analysis model set and the charging device fault analysis target domain are mapped and migrated to analyze to determine a migration fine-tuning strategy; based on the migration fine-tuning strategy, the source domain device fault task analysis model set is fine-tuned through transfer learning to obtain the device fault task model set.

[0025] In an embodiment of the present application, a source domain fault task analysis data set is first selected according to the fault prediction task list. The data set includes labeled device fault data, which comes from charging devices of similar type to the target device or devices of the same type. Specifically, through a database retrieval method, relevant fault data is extracted from the device operation management database, including sensor data (such as power, current, voltage, temperature), historical fault logs, and equipment maintenance records, and a data set that meets the requirements of the fault prediction task is screened out. Through this process, a labeled source domain fault task analysis data set is obtained.

[0026] After obtaining the source domain data set, supervised learning methods are used to pre-train the model to build a source domain equipment fault task analysis model set. The source domain data set is divided into a training set and a validation set, and the data is pre-processed, including normalization, voltage and current time series alignment, missing value filling, etc., to ensure the stability of the model input features. Then a deep neural network (DNN) is selected for model training, and the model parameters are updated through gradient descent optimization so that the model can learn the fault mode in the source domain data. For example, the system inputs the power, current, and temperature change data of the device during normal operation and failure, so that the model can learn how to distinguish between normal operation and failure. Through training, the source domain equipment fault task analysis model set is finally obtained, which can predict faults based on the data of the source device, but has not yet been optimized for the characteristics of the target device.

[0027] After obtaining the source domain model, the target domain of fault analysis of the target device is determined according to the fault prediction task list, that is, the fault prediction requirements and data characteristics of the target device. Specifically, a statistical analysis method is used to perform distribution calculations on the operating data of the target device, including calculating the mean, variance, maximum and minimum values ​​of the data, and analyzing the correlation between different sensor features to evaluate the data characteristics of the target device. For example, if the charging method of the target device is different from that of the source device, it may cause changes in voltage and current characteristics. Therefore, the correlation between the power and current characteristics of the target device and the source domain device is calculated to determine whether the data is transferable. At the same time, the fault history data of the target device is collected, and the frequency of occurrence of various types of faults is counted to confirm the distribution of different fault types on the target device. Finally, the characteristic distribution of the target domain of fault analysis of the target device is obtained to provide data support for subsequent transfer learning.

[0028] After determining the target domain for fault analysis of the target device, perform feature mapping and perform mapping migration analysis on the source domain device fault task analysis model set and the target domain for fault analysis of the target device. By using the cosine similarity calculation method, calculate the data distribution difference between the source domain device and the target device, and evaluate which source domain model knowledge can be effectively migrated to the target device. For example, vectorize the source domain data and the target device data, and calculate the cosine similarity to determine whether the source domain model can be effectively applied to the target device. If the feature difference is large, perform feature alignment during the transfer learning process to reduce inter-domain deviation. In addition, analyze the feature importance of different fault types, evaluate whether the target device has fault types that have not appeared in the source domain data set, and decide whether the input features of the model need to be adjusted. Finally, obtain the feature mapping relationship between the source domain and the target domain to ensure that transfer learning can be effectively performed.

[0029] Based on the results of the mapping migration analysis, a migration fine-tuning strategy is formulated to adjust the source domain model to adapt to the feature distribution of the target device. A layer freezing strategy is adopted to lock the underlying feature extraction layer of the source model, and only the high-level decision layer is retrained to maintain the common features learned by the source model, while optimizing the classification decision based on the data of the target device. In addition, the learning rate of the model is adjusted according to the data scale and distribution of the target device. If the target device has less data, a lower learning rate is used for fine-tuning to avoid drastic changes in model parameters and destroying the learning ability of the source model; if the target device has more data, a higher learning rate is used to speed up the model's adaptation to the target device data. Through this process, a migration fine-tuning strategy for the target device is obtained.

[0030] Finally, the source domain device failure task analysis model set is fine-tuned through transfer learning based on the transfer fine-tuning strategy. In this process, the operating data of the target device is used for further training, and small batch gradient descent is used for optimization to ensure that the model can gradually adapt to the data characteristics of the target device. During the training process, the loss change of the model on the target device data is monitored, and the number of training rounds is adjusted according to the loss convergence to ensure the stable performance of the model on the target device. In addition, the cross-validation technology is used to ensure the generalization ability of the model on different data subsets to avoid overfitting the model to the target device data. Through this process, a device failure task model set suitable for the target device is finally obtained.

[0031] The analysis module 13 is used to mine and obtain a charging equipment fault data set, perform feature cascade analysis on the charging equipment fault data set, and obtain an N-level equipment fault data map.

[0032] In an embodiment of the present application, a charging equipment fault data set is first extracted from the operation record of the charging equipment. These data mainly come from the equipment's operation log, sensor monitoring data, maintenance records, and operation and maintenance system. A log analysis method is used to parse the operation log of the charging pile to identify key events when a fault occurs, such as abnormal current, increased temperature, battery overcharge, etc. At the same time, sensor data is extracted from the equipment monitoring system, including parameters such as voltage, current, power, temperature, and charging time, and records with fault characteristics are screened out in combination with anomaly detection algorithms. In addition, a historical maintenance record matching method is used to compare the time when the equipment failure occurred with the maintenance log to confirm the type of fault and the scope of impact. Through this series of data mining processes, a charging equipment fault data set is obtained.

[0033] Next, the charging equipment fault data set is subjected to feature cascade analysis. Specifically, the fault data set is first cleaned and converted, abnormal data is removed, missing values ​​are filled, and the data format is standardized to generate a usable charging equipment fault data set. Subsequently, fault-related features are extracted from the data set to screen out key features that affect equipment failures, and redundant or low-correlation features are removed to form a usable fault-related feature set. Then, logical cascades are performed according to the association levels between features to establish an N-level fault feature map (N≥2), that is, a multi-level relationship structure from basic sensor data to complex fault modes, which displays the correlation of different fault features in a hierarchical manner. Finally, the N-level fault feature map is associated with the available charging equipment fault data set through data mapping to ensure traceability and matching between data, and form an N-level equipment fault data map.

[0034] Furthermore, in the system provided in the application embodiment, the analysis module 13 is also used for: The charging equipment fault data set is cleaned and converted to obtain an available charging equipment fault data set; a fault correlation feature set of the available charging equipment fault data set is extracted, and an available fault correlation feature set is selected from the fault correlation feature set; the available fault correlation feature sets are logically cascaded according to the correlation level to construct an N-level fault feature map, where N≥2; the N-level fault feature map is data mapped and associated with the available charging equipment fault data set to obtain the N-level equipment fault data map.

[0035] In the embodiment of the present application, the charging equipment fault data set is first cleaned and converted to improve data quality and ensure data consistency. By using the Z-score method for outlier detection, abnormal data exceeding 3 times the standard deviation is eliminated. For example, when the device voltage is far beyond the normal operating range (such as more than 400V), the data is regarded as an outlier and removed. In addition, for missing data, the K nearest neighbor interpolation method is used to fill it. For example, if a device lacks temperature data in a certain period of time, it is supplemented based on the data at the adjacent time to ensure the integrity of the data. Then, in order to eliminate the influence of different data scales, the Min-Max normalization method is used to scale all numerical features (such as current, voltage, temperature) to the [0,1] interval. For example, the voltage range of a device is [200V, 400V], which is converted to [0,1] after normalization to ensure that different features have a unified scale. After this process, a usable charging equipment fault data set is obtained.

[0036] After obtaining the available charging equipment fault data set, fault-related feature extraction is performed to screen out the key factors affecting equipment failure and remove redundant or irrelevant features. Specifically, the information gain method is used to evaluate the contribution of each feature to fault prediction, such as calculating the impact of current fluctuations, battery temperature rise and other features on equipment failure, and screening out the features with the highest information gain. In addition, in order to identify the interaction between features, the Pearson correlation coefficient is used to analyze the relationship between different fault features, such as calculating the correlation between battery temperature and charging current. If the correlation coefficient is greater than 0.7, it indicates that the battery temperature rise may be closely related to current fluctuations and is a highly correlated feature. Subsequently, in order to reduce the data dimension and improve the computational efficiency, the principal component analysis method is used to merge or delete redundant features by calculating the importance of features. For example, among the 10 current monitoring features, only the most representative 2 features are retained. After feature extraction, an available fault-related feature set is obtained, which includes operation features (such as current, voltage, temperature, etc.), usage features (such as charging times, cumulative usage time, etc.) and environmental features (such as humidity, external temperature, etc.).

[0037] Then, according to the correlation between the features, the hierarchical clustering method is used for logical cascading to construct an N-level fault feature map, where N≥2, indicating that the fault feature contains at least two levels. First, the features are grouped according to the importance of the features and the scope of fault impact, such as primary features, secondary features, tertiary features, N-level features, etc. Among them, the primary features (basic monitoring data) include current, voltage, temperature, humidity, etc.; the secondary features (equipment status) include battery overheating, voltage anomaly, current fluctuation, etc.; the tertiary features (failure mode) include heat dissipation failure, power anomaly, short circuit failure, etc.; the N-level features (system-level failure) include charging pile failure, module damage, equipment failure, etc. After the hierarchical clustering is completed, the directed acyclic graph method is further used to establish the causal relationship between the fault features. For example, "current anomaly" → "charging power anomaly" → "charging module failure", "battery temperature increase" → "heat dissipation system load increase" → "heat dissipation failure", by constructing this acyclic path, the fault features can be logically connected in series according to different levels. In addition, the Markov chain method is used to calculate the conversion probability of different fault modes. For example, when the battery temperature rises above 50°C, the probability of power anomaly increases by 40%, and when the current fluctuation exceeds 20%, the probability of short circuit fault increases by 60%. Through hierarchical clustering and acyclic graph modeling, an N-level fault feature map is constructed, which can clearly show the fault evolution path of the equipment and organically integrate the fault features at different levels.

[0038] Finally, the N-level fault feature map is associated with the available charging equipment fault data set to ensure that the fault data can be correctly classified in the map structure and used for subsequent analysis. Specifically, the feature embedding method is first used to convert each feature in the data set into a vector and match it to the feature node in the map. For example, the "current anomaly" feature is embedded in the "first-level feature" layer of the map, and the "charging module failure" feature is embedded in the "third-level feature" layer. Then the Euclidean distance method is used to calculate the similarity between the data point and the map node to determine the attribution relationship of the data point. For example, if the Euclidean distance between the current fluctuation data of a certain device and the "power anomaly" feature node is less than 0.1, the data is mapped to the node. Subsequently, in order to ensure the matching of the data in the time dimension, the dynamic time warping method is used to ensure that the data in different time periods can be correctly matched to the map. For example, if the voltage anomaly occurs 15 minutes later and causes the charging pile failure, a time association is established in the map. Through this series of data mapping and association operations, the N-level equipment fault data map is finally generated.

[0039] The matching and associating module 14 is used to match and associate with the N-level equipment fault data map according to the target equipment attribute parameters and the fault prediction accuracy requirements to obtain an equipment associated fault data set.

[0040] In the embodiment of the present application, the matching and association module 14 obtains a device-associated fault data set by associating and matching the target device attribute parameters and the fault prediction accuracy requirements with the N-level device fault data maps.

[0041] Specifically, firstly, based on the target device attribute parameters, such as operating characteristics such as voltage, current, temperature, and usage characteristics such as the number of charging times and usage time, the associated features are matched with the N-level device fault data map to determine which fault characteristics are closely associated with the device attributes, and determine the association direction of the map, that is, the impact path between the features. Then, according to the requirements of fault prediction accuracy, the impact of different feature data on fault prediction accuracy is evaluated, and the data association depth of the fault map, that is, the hierarchical depth and complexity of the feature data, is determined to meet the accuracy requirements of the model. Finally, based on the above-mentioned association direction and data association depth, the data in the map is associated and divided, and the matched feature data is organized into a device-associated fault data set.

[0042] Furthermore, in the system provided in the embodiment of the application, the matching association module 14 is also used for: Based on the target device attribute parameters and the N-level device fault data map, correlation feature matching is performed to determine the fault map correlation direction; according to the fault prediction accuracy requirement, the fault map data correlation depth is determined; based on the fault map correlation direction and the fault map data correlation depth, the N-level device fault data map is data-correlatedly divided to obtain the device-correlated fault data set.

[0043] In an embodiment of the present application, firstly, the associated feature matching is performed based on the attribute parameters of the target device and the fault data map of the N-level device to determine the association direction of each feature in the fault map. The attribute parameters of the target device include the operation characteristics (such as voltage, current, temperature), usage characteristics (such as the number of charging times, usage time) and environmental characteristics (such as temperature and humidity) of the device. These attribute parameters are collected by the sensor data of the device and transmitted to the analysis system in real time. In order to determine the association direction of the fault map, the attribute parameters of the target device are converted into a vector representation using a feature embedding method, which is convenient for matching with the feature nodes in the fault data map of the N-level device. Feature embedding converts continuous numerical data (such as voltage, current, temperature) into a vector form that can be used for calculation, which is convenient for effective matching with the map nodes. Next, a similarity calculation method such as cosine similarity or Euclidean distance is used to quantify the similarity between the attribute vector of the target device and each feature node in the fault map. For example, if the cosine similarity between the abnormal current characteristics of the equipment and the "current fluctuation" characteristic node in the spectrum is greater than a preset threshold (the threshold is pre-set), it is determined that there is a close correlation between the two, and the current fluctuation characteristics have an important impact on the equipment failure, thereby determining the correlation direction of the fault spectrum.

[0044] Next, the data association depth of the fault map is determined according to the set fault prediction accuracy requirements. Fault prediction accuracy requirements usually include indicators such as accuracy, recall, and F1 score, which define the goals of fault prediction. For example, for some devices, more emphasis may be placed on reducing missed reports (high recall), while for other devices, more emphasis may be placed on reducing false positives (high accuracy). Based on these accuracy requirements, the Pearson correlation coefficient between each feature and the occurrence of the fault is calculated to directly determine its association depth in the map. Features with high correlation (such as current, voltage, temperature, etc.) are assigned to deep levels of the map, while features with low correlation (such as ambient humidity) are assigned to shallower levels. In this way, the data association depth of the fault map is determined to ensure that high-correlation features can be deeply associated, while low-correlation features are simplified.

[0045] Finally, according to the fault map association direction and data association depth determined previously, the N-level equipment fault data map is divided into data associations. The features are divided into levels through hierarchical clustering and other methods, and high-correlation features (such as current, voltage, etc.) are divided into deep levels, while low-correlation features (such as humidity, external temperature, etc.) are divided into shallow levels. Through this hierarchical structure, it is ensured that the map can accurately capture the failure mode of the equipment, and reasonable data division is performed according to the attributes of the equipment and the requirements for fault prediction accuracy. Finally, through these data divisions, the equipment-related fault data set is obtained.

[0046] The early warning module 15 is used to perform integrated learning and training on the equipment failure task model set based on the equipment-associated failure data set, build an equipment failure integrated prediction model, and perform fault prediction and early warning on the target charging equipment through the equipment failure integrated prediction model.

[0047] In the embodiment of the present application, the early warning module 15 performs integrated learning and training on the equipment failure task model set based on the equipment-associated fault data set to construct an integrated prediction model for equipment failure. Specifically, the equipment failure task model set is first trained and updated in sequence, and an updated fault prediction task update model set is generated therefrom. Next, a model performance evaluation index set is constructed, and these indicators are used to perform performance evaluation on the updated model set to obtain a set of equipment failure model performance parameters. Based on these performance parameters, the fault model decision coefficient set of each model is further determined, and the model is integrated and learned based on these coefficients, and finally a comprehensive integrated prediction model for equipment failure is obtained.

[0048] Next, the target charging equipment is predicted and warned of failures through the integrated equipment failure prediction model. Specifically, the real-time data of the target charging equipment is input into the integrated model, and the model predicts the probability of failure based on the current state of the equipment and issues a warning when a potential failure is detected. The warning signal can inform the operation and maintenance personnel of possible equipment problems in advance and provide corresponding fault information, thereby helping to take appropriate maintenance measures to ensure the safe and stable operation of the equipment.

[0049] Furthermore, in the system provided in the application embodiment, the early warning module 15 is also used for: Based on the equipment-associated fault data set, the equipment fault task model set is trained and updated in sequence to obtain a fault prediction task update model set; a model performance evaluation index set is constructed, and the performance of the fault prediction task update model set is evaluated using the model performance evaluation index set to obtain an equipment fault model performance parameter set; a fault model decision coefficient set is determined based on the equipment fault model performance parameter set; based on the fault model decision coefficient set, a fusion integration learning is performed on the fault prediction task update model set to construct an equipment fault integrated prediction model.

[0050] In an embodiment of the present application, the equipment fault task model set is first trained and updated in sequence based on the equipment-associated fault data set. In this process, the equipment's operating data (such as voltage, current, temperature, usage time, etc.) and fault history data (such as fault type, occurrence time, repair record, etc.) are preprocessed, and machine learning methods such as random forest, support vector machine and gradient boosting decision tree are used to train different fault prediction models. Random forest uses multiple decision trees for integration to reduce the overfitting problem of a single tree model, SVM optimizes the fault classification boundary and improves classification accuracy, while GBDT reduces prediction errors and improves the advance amount of fault prediction through iterative optimization. In addition, hyperparameter tuning is used to optimize model parameters, and cross-validation is used to evaluate the generalization ability of the model to ensure the stability of each model on different fault tasks. Through this process, an optimized fault prediction task update model set is obtained.

[0051] Next, we construct a set of model performance evaluation indicators, and use the model performance evaluation indicator set to evaluate the performance of the updated model set for the fault prediction task. The model performance evaluation indicator set includes accuracy, recall, precision, F1 score, and AUC, among which accuracy measures the proportion of correct predictions, recall evaluates the model's ability to identify all fault instances, precision reflects the proportion of instances predicted to be faults that are actually faults, and F1 score is used to balance accuracy and recall, while AUC measures the model's ability to handle unbalanced data. By calculating these indicators for each fault prediction model in turn and using K-fold cross validation to reduce the error caused by data bias, we finally obtain the equipment fault model performance parameter set.

[0052] Then, based on the performance parameter set of the equipment fault model, the fault model decision coefficient set is calculated and determined, that is, weights are assigned to each model to ensure that the model with better performance in the integrated learning process can play a greater role. Specifically, the accuracy, recall, precision, F1 score and AUC of each model are first normalized to ensure that the dimensions of different indicators are consistent. Then weights are assigned to different evaluation indicators, and the comprehensive decision coefficient of each model is obtained based on the weighted calculation method. The allocation of weights is based on a preset strategy. For example, the recall rate has a higher weight to reduce missed faults, while the accuracy rate has a lower weight to improve the credibility of the overall prediction. After the calculation is completed, the comprehensive decision coefficients of all models are normalized, and finally the fault model decision coefficient set is obtained, which clarifies the relative contribution and influence of each model in the final prediction task.

[0053] After determining the weights of each model, the fault prediction task update model set is fused and integrated based on the fault model decision coefficient set to build an integrated prediction model for equipment faults. First, parallel ensemble learning is used to run all models simultaneously and build a preliminary initial fault integrated prediction model. Then, the impact model is parsed for each model to calculate the mutual influence relationship between different task models, thereby generating a set of affected task models and a set of associated impact models. The affected task model set contains models that may be affected by other tasks, while the associated impact model set describes the relationship between models. Subsequently, the impact decision coefficient set is calculated based on the fault model decision coefficient set to ensure that high-impact models have a greater weight in the final prediction. Finally, a weighted fusion integration is performed using a weighted voting method to vote on the prediction results of different models according to the calculated decision weights, and finally an integrated prediction model for equipment faults is generated.

[0054] Furthermore, in the system provided in the application embodiment, the early warning module 15 is also used for: The fault prediction task update model set is subjected to parallel integrated learning to construct an initial fault integrated prediction model; the impact model is analyzed for each task model in the fault prediction task update model set to obtain an affected task model set and a corresponding associated impact model set; according to the fault model decision coefficient set, an impact decision coefficient set of the associated impact model set is determined; according to the impact decision coefficient set, the initial fault integrated prediction model is weightedly integrated based on the affected task model set and the corresponding associated impact model set to obtain an equipment fault integrated prediction model.

[0055] In an embodiment of the present application, the fault prediction task update model set is firstly parallel integrated learning to construct an initial fault integrated prediction model. In this process, a stacked generalization method is adopted to train multiple basic models (such as decision trees, random forests, support vector machines, etc.), and their prediction results are input as new features into the meta-model for training. The meta-model usually adopts simple linear regression or logistic regression to reduce the deviation of a single model and improve the overall prediction performance. Finally, the initial fault integrated prediction model is obtained through this method.

[0056] Subsequently, the influence model is parsed for each task model in the fault prediction task update model set to identify the association between models and form an affected task model set and an associated influence model set. In this process, the Pearson correlation coefficient analysis is used to calculate the linear correlation between the prediction results of different task models, and a threshold (such as 0.6) is set. When the correlation between the prediction results of two task models exceeds the threshold, it is considered that there is an influence relationship between them. Based on the calculation results, an affected task model set is constructed, which contains models that are greatly affected by other task models, and an associated influence model set is formed, which contains task models that have a strong influence on other models.

[0057] Based on the results of the impact model analysis, the impact decision coefficient set of the associated impact model set is further calculated according to the fault model decision coefficient set. In this process, the analytic hierarchy process (AHP) is used to calculate the impact weights of each task model. First, a comparison matrix between task models is constructed, including evaluation indicators such as accuracy, recall, F1 score, AUC, and the importance weight of each indicator is determined using the pairwise comparison method. Then, the eigenvector is calculated to obtain the final impact decision coefficient of each model, and a consistency check is performed to ensure the rationality of the calculation. Finally, the impact decision coefficient set is obtained to adjust the contribution of each task model in the final prediction.

[0058] Finally, according to the set of influencing decision coefficients, based on the set of affected task models and the set of associated impact models, the initial fault integrated prediction model is weighted and integrated, and finally the equipment fault integrated prediction model is obtained. In this process, the weighted mean square error method is used to calculate the weighted average of the prediction results of each task model, so that the model with a higher influencing decision coefficient accounts for a larger proportion in the final prediction, while the model with a smaller influence contributes less. Through this process, the construction of the equipment fault integrated prediction model is completed.

[0059] Furthermore, in the system provided in the application embodiment, the early warning module 15 is also used for: The associated influence model set is weightedly fused based on the influence decision coefficient set to obtain the fault influence model set; the output layer of the fault influence model set is linked to the corresponding affected task model set, and the initial fault integrated prediction model is integrated, optimized and updated through the linked affected task model set to obtain the equipment fault integrated prediction model.

[0060] In an embodiment of the present application, firstly, based on the obtained set of influence decision coefficients, the associated influence model set is weightedly fused to obtain a fault influence model set. Specifically, the output results of each associated influence model are weighted according to its decision coefficient to ensure that the model that contributes more to the final prediction result occupies a larger proportion in the fusion process. Weighted fusion adopts the weighted average method, that is, the weighted average of the output results is calculated according to the decision coefficient of each model. For example, if the decision coefficient of a fault prediction model is high, the output result of the model will occupy a greater weight in the weighted average calculation, reflecting its importance to the final prediction result. Through weighted fusion, the prediction information of multiple associated influence models is effectively integrated to generate a fault influence model set that comprehensively considers the influence of each model.

[0061] The output layer of the fault impact model set is then linked to the corresponding affected task model set. This process achieves interaction and information sharing between task models by combining the prediction results of the fault impact model set with the output results of the affected task model set. Each model in the affected task model set represents the prediction of a specific fault type, while the fault impact model set integrates multiple influencing factors, such as the prediction output of other task models and the influence of associated models. By linking the outputs of these models, the prediction results of the fault impact model are combined with the prediction results of the affected task model to improve the accuracy and reliability of the overall prediction. For example, the mutual influence between the battery fault model and the charging module fault model can be reflected through this link, making the final prediction result more comprehensive, taking into account all relevant failure modes.

[0062] Finally, based on the linked set of affected task models, the initial fault integrated prediction model is integrated, optimized and updated. By combining the output results of the fault impact model set and the affected task model set, the initial fault integrated prediction model is optimized. This optimization process usually adopts a stacking learning method, in which the prediction outputs of multiple models are input as new features into a new meta-learning model (such as linear regression, support vector machine, etc.). The meta-model will further learn how to optimally fuse the outputs of each model to optimize the prediction results. Through this process, it is ensured that the equipment fault prediction model can combine the advantages of all models, reduce deviations and improve the accuracy of predictions. Finally, through these steps, the equipment fault integrated prediction model is obtained.

[0063] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects: The present application builds a device attribute classifier, classifies the attributes of the target charging device based on the device attribute classifier, and obtains the target device attribute parameters; constructs a fault prediction task list according to the charging device fault prediction target, and obtains the device fault task model set based on the transfer learning of the fault prediction task list; mines and obtains the charging device fault data set, performs feature cascade analysis on the charging device fault data set, and obtains an N-level device fault data map; matches and associates with the N-level device fault data map according to the target device attribute parameters and the fault prediction accuracy requirements, and obtains a device-associated fault data set; performs integrated learning and training on the device fault task model set based on the device-associated fault data set, constructs an equipment fault integrated prediction model, and performs fault prediction and early warning on the target charging device through the equipment fault integrated prediction model. The present invention solves the technical problem of insufficient accuracy in the prior art in predicting charging equipment faults. It classifies the attributes of target equipment by constructing an equipment attribute classifier, obtains an equipment fault task model set by using a fault prediction task list and transfer learning, and constructs an equipment fault data map through feature cascade analysis. Combined with the target equipment attribute parameters and the fault prediction accuracy requirements, an equipment associated fault data set is obtained through association matching, and finally an equipment fault integrated prediction model is constructed through integrated learning to predict and warn charging equipment faults, thereby achieving the technical effect of improving the accuracy of charging equipment fault prediction.

[0064] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0066] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A charging equipment fault prediction system based on multi-model ensemble learning, characterized in that: The system comprises: A classifier building module, used to build a device attribute classifier, classify the attributes of the target charging device based on the device attribute classifier, and obtain the target device attribute parameters; A task list construction module is used to construct a fault prediction task list according to the charging equipment fault prediction target, and obtain a device fault task model set based on the fault prediction task list transfer learning; An analysis module is used to mine and obtain a charging equipment fault data set, perform feature cascade analysis on the charging equipment fault data set, and obtain an N-level equipment fault data map; A matching and association module, used for matching and associating with the N-level equipment fault data map according to the target equipment attribute parameters and the fault prediction accuracy requirements, to obtain an equipment associated fault data set; The early warning module is used to perform integrated learning and training on the equipment failure task model set based on the equipment-associated fault data set, build an equipment failure integrated prediction model, and perform fault prediction and early warning on the target charging equipment through the equipment failure integrated prediction model.

2. The charging equipment fault prediction system based on multi-model ensemble learning according to claim 1, characterized in that: The classifier building module is used to: Defining charging equipment attribute factor information, wherein the charging equipment attribute factor information includes application scenarios, equipment specifications, and charging methods; Acquire a charging device database, and classify and identify the attributes of each device data in the charging device database based on the attribute factor information of the charging device to obtain a charging device attribute parameter set; Building a coding system, and performing knowledge coding on the charging device attribute factor information and the charging device attribute parameter set based on the coding system to obtain a device attribute coding parameter set; A decision tree algorithm is used to perform classification training and optimization on the device attribute encoding parameter set to build the device attribute classifier.

3. The charging equipment fault prediction system based on multi-model ensemble learning according to claim 1, characterized in that: The task list building module is used to: According to the fault prediction task list, a source domain fault task analysis dataset is selected, wherein the source domain fault task analysis dataset includes a fault task analysis dataset of annotated devices; Use the source domain fault task analysis data set to perform model pre-training to obtain a source domain device fault task analysis model set; According to the fault prediction task list, determine the charging equipment fault analysis target domain, perform mapping migration analysis on the source domain equipment fault task analysis model set and the charging equipment fault analysis target domain, and determine the migration fine-tuning strategy; Based on the migration fine-tuning strategy, the source domain device failure task analysis model set is subjected to transfer learning fine-tuning to obtain the device failure task model set.

4. The charging equipment fault prediction system based on multi-model ensemble learning according to claim 1, characterized in that: The analysis module is used to: Performing data cleaning and data conversion on the charging equipment fault data set to obtain a usable charging equipment fault data set; Extracting a fault-related feature set of the available charging equipment fault data set, and selecting the fault-related feature set to obtain an available fault-related feature set; Logically cascading the available fault correlation feature sets according to the correlation levels to construct an N-level fault feature map, where N≥2; The N-level fault feature map is associated with the available charging device fault data set through data mapping to obtain the N-level device fault data map.

5. The charging equipment fault prediction system based on multi-model ensemble learning according to claim 1, characterized in that: The matching association module is used to: Based on the target device attribute parameters and the N-level device fault data map, correlation feature matching is performed to determine the fault map correlation direction; Determine the fault map data association depth according to the fault prediction accuracy requirement; Based on the fault map association direction and the fault map data association depth, the N-level device fault data map is divided into data associations to obtain the device association fault data set.

6. The charging equipment fault prediction system based on multi-model ensemble learning as claimed in claim 3, characterized in that: The early warning module is used to: Based on the equipment-associated fault data set, the equipment fault task model set is trained and updated in sequence to obtain a fault prediction task update model set; Constructing a model performance evaluation index set, using the model performance evaluation index set to perform performance evaluation on the fault prediction task update model set respectively, and obtaining a device fault model performance parameter set; Determining a fault model decision coefficient set according to the equipment fault model performance parameter set; Based on the fault model decision coefficient set, the fault prediction task update model set is subjected to fusion ensemble learning to construct an equipment fault ensemble prediction model.

7. The charging equipment fault prediction system based on multi-model ensemble learning according to claim 6, characterized in that: The early warning module is used to: Performing parallel ensemble learning on the fault prediction task update model set to construct an initial fault ensemble prediction model; Performing impact model analysis on each task model in the fault prediction task update model set respectively to obtain an affected task model set and a corresponding associated impact model set; Determining the impact decision coefficient set of the associated impact model set according to the fault model decision coefficient set; According to the impact decision coefficient set, based on the affected task model set and the corresponding associated impact model set, the initial fault integrated prediction model is weightedly integrated to obtain the equipment fault integrated prediction model.

8. The charging equipment fault prediction system based on multi-model ensemble learning according to claim 7, characterized in that: The early warning module is used to: Performing weighted fusion on the associated impact model set based on the impact decision coefficient set to obtain a fault impact model set; The output layer of the fault impact model set is linked to the corresponding affected task model set, and the initial fault integrated prediction model is integrated, optimized and updated through the linked affected task model set to obtain the equipment fault integrated prediction model.

Images