Charging station operation and maintenance data inconsistency detection and restoration method, system and device based on multi-source data fusion and medium

Abstract

The invention discloses a charging station operation and maintenance data inconsistency detection and restoration method, system and device based on multi-source data fusion and a medium, and relates to the field of intelligent maintenance of charging station operation data, and the method comprises the steps: collecting heterogeneous data sources in power station operation and maintenance; preprocessing the collected data to generate a consistent fusion data set; constructing an inconsistency detection model on the basis of fused data, extracting data implicit structural features based on a feature learning method of an auto-encoder, identifying potential abnormal points by using reconstruction errors, and performing consistency dynamic detection and classification labeling on multi-source data; and inputting the identified inconsistent data into a repair module, executing a repair mechanism, and complementing, correcting and reconstructing the abnormal data. According to the method, a multi-source heterogeneous data collaborative analysis framework integrating the operation state of the charging pile, the power grid interaction data and the environment perception data is constructed, the limitation of traditional single data source verification is broken through, and the cross-system data relevance and the anomaly detection coverage rate are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent maintenance of charging station operation data, and particularly relates to a charging station operation and maintenance data inconsistency detection and repair method, system, equipment and medium based on multi-source data fusion. BACKGROUND

[0002] When performing operation and maintenance management on the charging station, a large amount of data from different sources often needs to be processed. Multi-source heterogeneous data fusion and inconsistency detection and repair are important technologies for modern charging station operation and maintenance management, and their efficiency and accuracy directly affect the operation stability and service quality of the charging station. With the diversification of charging station data sources, problems such as format heterogeneity, semantic conflict and inconsistency may occur during data collection and analysis. Traditional methods are difficult to efficiently integrate these multi-source data, limiting the value mining and application of data in charging station operation and maintenance decision-making.

[0003] To solve the complexity and efficiency problems in charging station operation and maintenance data fusion, on the one hand, the research on data feature extraction and inconsistency detection technology needs to be strengthened. Advanced deep learning models are used to efficiently extract features from charging station operation and maintenance data from different devices, sensors and users, to mine deep semantic associations and improve the accuracy and comprehensiveness of data fusion. On the other hand, the repair strategy for inconsistent data needs to be optimized. Combined with the context information and data distribution characteristics of charging station operation and maintenance, high-quality repair values are generated to enhance the scientificity and accuracy of repair, ensuring the consistency and credibility of charging station operation and maintenance data, and thus improving the effectiveness and reliability of operation and maintenance decision-making.

[0004] Traditional data processing methods have deficiencies when processing multi-source heterogeneous data in charging station operation and maintenance, mainly in three aspects of data fusion, inconsistency detection and data repair. Traditional methods are difficult to deeply mine the potential relationship between multi-source data, limiting the ability of high-quality data fusion. In the inconsistency detection of high-dimensional and multi-modal data, relying on rules or simple models, it is difficult to fully capture conflict features, and the accuracy of the detection result is insufficient. Repair methods mostly use interpolation algorithms or fixed rules, ignoring context information and global distribution characteristics, resulting in low credibility of repair results and difficulty in ensuring data consistency and quality. These deficiencies limit the demand for real-time, accuracy and reliability of the charging station operation and maintenance system, affecting the deep mining and application of data.

[0005] To solve the above problems, the present application proposes a multi-source data fusion and inconsistency repair method based on deep learning. Through deep representation learning, efficient data fusion is realized, and combined with the generation of adversarial network (GAN), the data repair process is optimized, significantly improving the quality and efficiency of charging station operation and maintenance data processing, and providing reliable support for intelligent operation and maintenance of charging stations. SUMMARY

[0006] In view of the above problems, the present application is proposed.

[0007] Therefore, the problem to be solved by the present application is that the conventional method is difficult to deeply mine the potential relationship between multi-source data, limits the high-quality data fusion capability, and in the inconsistency detection of high-dimensional and multi-modal data, relies on rules or simple models, and is difficult to comprehensively capture conflict features.

[0008] To solve the above technical problems, the present application provides the following technical solutions: a charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, comprising: for heterogeneous data sources in the operation and maintenance of the charging station, formulating a hierarchical collection strategy according to the data source, the granularity characteristics and the update frequency, collecting multi-source original data including device state, operation and maintenance records and environmental information in a time window; based on the collected multi-source original data, constructing a pre-fusion processing module, cleaning and standardizing the data, and eliminating field naming ambiguity through an attribute alignment method, completing cross-source entity alignment through entity recognition, and generating a consistent fusion data set; constructing an inconsistency detection model based on the fusion data, wherein the inconsistency detection model extracts data implicit structure features based on a feature learning method of a self-encoder, identifies potential abnormal points using reconstruction error, and performs consistency dynamic detection and classification labeling on multi-source data; inputting the identified inconsistent data into a repair module to execute a repair mechanism to complete, correct and reconstruct abnormal data.

[0009] As a preferred scheme of the charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, the construction method of the attribute alignment method comprises: based on the data sources of the charging station equipment, the national network Internet of Vehicles and the energy service platform, analyzing the original entities in each data source, and through mapping rules, corresponding the original entities to aligned entities of a unified standard, to ensure the consistency and compatibility of multi-source data.

[0010] As a preferred scheme of the charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, the cross-source entity alignment through entity recognition comprises: for the multi-source heterogeneous entity names existing in the operation and maintenance data of the charging station, designing a similarity calculation mechanism of fusion text information and semantic information, establishing a similarity matching model taking entity similarity evaluation as the core, and supporting entity normalization processing; the similarity of the text information is represented as,

[0011]

[0012] Wherein, A and B are two text word sets, A text word set and B text word set are represented, |A∩B| represents the number of words appearing in common in the word sets of the two texts, |A∪B| represents the number of all non-repeated words in the two text word sets; the similarity of the semantic information is represented as,

[0013]

[0014] Wherein, And Respectively, the word vector representation of A text word set and B text word set, Indicates the vector dot product, A i Indicates the i-th word in the A text word set, B i Indicates the i-th word in the B text word set, ||A|| and ||B|| represent the modulus of the A text vector and the modulus of the B text vector respectively; the formula of the entity similarity is represented as,

[0015] Overall Similarity=0.4·Jaccard Similarity+0.6·Cosine Similarity

[0016] Wherein, Overall Similarity represents the overall similarity, Jaccard Similarity represents the text similarity, and Cosine Similarity represents the semantic similarity.

[0017] As a preferred scheme of the charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, wherein: the feature learning method based on the autoencoder extracts the data implicit structure feature, including constructing the autoencoder structure based on the multi-source fused charging station operation and maintenance data, extracting the potential feature representation of the data through the encoder, and high-dimensional reconstruction by the decoder; the formula of the potential feature z of the data is represented as,

[0018] z=f(x)=σ(W e ·x+b e )

[0019] Wherein, W e Indicates the weight matrix of the encoder, b e Indicates the encoder bias, and sigma indicates the activation function; the formula of the high-dimensional reconstruction is represented as,

[0020]

[0021] Wherein, Indicates the high-dimensional reconstructed data, W d Indicates the weight matrix of the decoder, b dThe decoder bias is represented.

[0022] As a preferred scheme of the charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, the consistency dynamic detection includes calculating reconstruction error and abnormal score based on the deviation between input data and reconstructed data to form an evaluation index for inconsistency point detection; the formula of the reconstruction error is,

[0023]

[0024] Wherein, e i represents the reconstruction error, x i represents the input data, represents the reconstructed data corresponding to the input data; the formula of the abnormal score is,

[0025]

[0026] Wherein, s i represents the abnormal score of the data, and e represents the set of all reconstruction errors.

[0027] As a preferred scheme of the charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, the completion includes completing through data source confidence, constructing a confidence weighted repair model based on the source reliability and historical consistency index of each data item in the multi-source data, and the formula is,

[0028] C(x i →v)=P(v|context(x i ))

[0029]

[0030] Wherein, v represents the repair value, C(x i →v) represents the confidence of repairing x i to v, and P(v|context(x i )) represents the conditional probability of v as the repair value under the current context condition, represents the repaired data.

[0031] As a preferred scheme of the charging station operation and maintenance data inconsistency detection and repair method based on multi-source data fusion, the correction includes correcting abnormal data through information entropy fluctuation characteristics, constructing an information entropy minimization model based on the distribution characteristics and uncertainty level of data attributes, and performing probability optimization repair on inconsistent data, and the formula is,

[0032]

[0033] where H(x i ) represents the information entropy of input data x i , V represents a set of possible repair values, and P(v) represents the probability distribution of v;the reconstruction comprises: constructing an abnormal data learning and reconstruction model based on a generative adversarial network structure, the reconstruction model being composed of a generator G and a discriminator D, modeling the relationship of the data and semantic reconstruction of the missing area through game optimization;constructing a target function The formula is expressed as,

[0034]

[0035] where x z represents real data, P data (x) represents the distribution of real data, z z represents noise, P Z (z) represents the distribution of input context features and noise, G(z) represents the output of the generator, i.e., the generated reconstructed value, and E represents mathematical expectation;the maximization loss function L D of the discriminator is expressed as,

[0036]

[0037] the minimization loss function L G of the generator is expressed as,

[0038]

[0039] To solve the above technical problems, the present application provides the following technical scheme: a charging station operation and maintenance data inconsistency detection and repair method system based on multi-source data fusion, comprising: a data acquisition module, a pre-fusion processing module, a detection module and a repair module;The data acquisition module formulates a hierarchical acquisition strategy according to data sources, granularity characteristics and update frequency for heterogeneous data sources in the operation and maintenance of the charging station, acquires multi-source original data including device status, operation and maintenance records and environmental information in a time window;The pre-fusion processing module cleans and standardizes the collected data, eliminates field naming ambiguity through attribute alignment method, completes cross-source entity alignment by entity recognition, and generates consistent fusion data set;The detection module constructs an inconsistency detection model based on the fusion data, the inconsistency detection model extracts data implicit structure features based on a feature learning method of a self-encoder, identifies potential abnormal points by reconstruction error, and performs consistency dynamic detection and classification labeling on multi-source data;The repair module completes, corrects and reconstructs the abnormal data.

[0040] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method for inconsistent detection and repair of charging station operation and maintenance data based on multi-source data fusion when executing the computer program.

[0041] A computer readable storage medium stores a computer program, and the computer program implements the steps of the method for inconsistent detection and repair of charging station operation and maintenance data based on multi-source data fusion when executed by a processor.

[0042] The present application has the following advantages: the present application constructs a multi-source heterogeneous data collaborative analysis framework that fuses charging pile operation state, power grid interaction data and environment perception data, breaks through the limitations of traditional single data source verification, significantly improves the cross-system data correlation and abnormal detection coverage; an inconsistent detection algorithm based on dynamic rule mining and spatio-temporal correlation analysis is proposed, which solves the pain point of insufficient adaptability of traditional static threshold rules under complex working conditions, and improves the accuracy of abnormal identification; an adaptive repair mechanism that fuses data confidence evaluation and repair path optimization is designed, which realizes intelligent repair decision-making in multiple conflict scenarios, and provides a high-precision and high-robustness data governance closed-loop solution for charging station operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0044] Figure 1 The flowchart of the method for inconsistent detection and repair of charging station operation and maintenance data based on multi-source data fusion in embodiment 1.

[0045] Figure 2 The flowchart of feature data discretization of the K-means clustering algorithm of the method for inconsistent detection and repair of charging station operation and maintenance data based on multi-source data fusion in embodiment 1.

[0046] Figure 3 The comparison result chart of different data inconsistent detection methods of the method for inconsistent detection and repair of charging station operation and maintenance data based on multi-source data fusion in embodiment 2. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for automatically comparing and discovering power standard difference information, including: Figure 1 As shown:

[0050] 1. For heterogeneous data sources in charging station operation and maintenance, a layered collection strategy is formulated based on data sources, granularity characteristics, and update frequency. Multi-source raw data, including equipment status, operation and maintenance records, and environmental information, are collected according to time windows.

[0051] This embodiment obtains data from charging pile equipment, the State Grid Internet of Vehicles, and provincial energy service platforms, and processes the obtained data.

[0052] 2. Based on the collected multi-source raw data, a fusion pre-processing module is constructed to clean and standardize the data, eliminate field naming ambiguity through attribute alignment, and use entity recognition to complete cross-source entity alignment to generate a consistent fusion data set.

[0053] This embodiment proposes a multivariate data fusion method, taking into account the normal operating state and fault state of the charging equipment, including:

[0054] First, the collected data is cleaned and preprocessed to deal with missing values, outliers and duplicate values.

[0055] Missing value processing: For samples with uniform distribution, the mean value can be used to fill the missing values, while for data with skewed distribution, the median can be used to fill the missing values.

[0056]

[0057] Among them, X i is the data value, n i is the number of data, α determines whether the data needs to be padded, α=1 indicates that data needs to be padded, and α=0 indicates that data does not need to be padded.

[0058] Outlier processing: Outliers refer to values ​​in the data that significantly deviate from the normal range and may have a negative impact on data analysis. When using the Z-score normalization method, the mean μ and standard deviation σ of the original data must first be calculated. The calculation formula is as follows:

[0059]

[0060] where x in represents the input operation and maintenance data, x out represents the output operation and maintenance data, μ represents the mean of the data, and σ represents the variance of the data.

[0061] Repeated value processing: The existence of repeated values can affect the accuracy and efficiency of data analysis. In terms of processing means, K-means clustering algorithm can be used to detect and remove repeated values, and the process is as shown in Figure 2 , which specifically includes:

[0062] Based on the given feature data set and the preset number of clusters K, first, K cluster centers are determined by random initialization; then each data point is classified to the nearest cluster center according to the distance measurement rule, and the preliminary division of data clusters is completed; then for each formed class cluster, the mean of all data points in it is recalculated as the updated cluster center; the data classification and center point updating operation are iteratively performed until the cluster center position converges stably or the preset termination condition is reached, and finally the K-means clustering division of the data set is realized. This process dynamically adjusts the class cluster center, gradually optimizes the balance between the intra-class compactness and the inter-class separation degree.

[0063] Secondly, after completing data preprocessing, in order to unify the format of multi-source data and ensure data consistency, attribute alignment and entity alignment are performed on the operation and maintenance data of charging stations to realize the fusion of multi-source data. Attribute alignment unifies the expression of attributes such as device state, sensor information and user behavior by constructing a mapping table, ensuring data consistency and comparability, and the specific implementation is shown in Table 1.

[0064] Table 1 Partial attribute alignment mapping of different data sources

[0065]

[0066] Finally, after completing attribute alignment, in order to further integrate the operation and maintenance data of charging stations, entity alignment needs to be performed to associate the records of the same device or user in different data sources. Entity alignment associates and integrates the records of the same device or user in different data sources, eliminates redundancy and conflicts, and provides a high-quality basis for data analysis and repair.

[0067] Through simple and efficient rules, the device and user records are preliminarily aligned; for the cases where the rules cannot accurately match, the similarity of the candidate entity pairs is calculated, and whether the alignment is performed is determined according to the preset threshold (such as 0.75). When multiple candidate pairs meet the threshold requirement, the pair with the highest similarity is selected as the final result, ensuring the accuracy and efficiency of entity alignment.

[0068] Aiming at the multi-source heterogeneous entity names existing in the operation and maintenance data of charging stations, a similarity calculation mechanism fusing text information and semantic information is designed, and a similarity matching model taking entity similarity evaluation as the core is established to support entity normalization processing.

[0069] The similarity of the text information is represented as,

[0070]

[0071] Wherein, A and B are the word sets of two texts, A and B represent the word sets of the A text and the B text, |A∩B| represents the number of common words in the word sets of the two texts, and |A∪B| represents the number of all non-repeated words in the word sets of the two texts.

[0072] The similarity of the semantic information is represented as,

[0073]

[0074] Wherein, And are the word vector representations of the word set A and the word set B of the A text and the B text respectively, represents the vector dot product, A i represents the i-th word in the word set A of the A text, B i represents the i-th word in the word set B of the B text, and ||A|| and ||B|| represent the modulus of the A text vector and the modulus of the B text vector respectively.

[0075] The formula of the entity similarity is represented as,

[0076] Overall Similarity=0.4·Jaccard Similarity+0.6·Cosine Similarity

[0077] Wherein, Overall Similarity represents the overall similarity, Jaccard Similarity represents the text similarity, and Cosine Similarity represents the semantic similarity.

[0078] In this embodiment, it is assumed that:

[0079] Text A: "Intelligent charging pile voltage anomaly"

[0080] Text B: "Intelligent charging equipment voltage failure"

[0081] Intersection |A∩B| = {"intelligent", "voltage"}, size 2.

[0082] Union | A U B | = { "smart", "charging pile", "voltage", "abnormal", "charging", "equipment", "appear"}

[0083] , size 8.

[0084]

[0085] After vectorization:

[0086] Vector

[0087] Vector

[0088]

[0089] Overall Similarity = 0.4 Jaccard Similarity + 0.6 Cosine Similarity = 0.4 x 0.25 + 0.6 x 0.96 = 0.676.

[0090] Three, an inconsistency detection model is constructed on the basis of fused data, the inconsistency detection model extracts data implicit structure features based on a feature learning method of an autoencoder, and uses reconstruction error to identify potential abnormal points, so that multi-source data is dynamically detected and classified and labeled for consistency.

[0091] On the basis of completing multi-source data fusion, data needs to be detected for inconsistency, in order to solve the complexity of high-dimensional multi-modal data in inconsistency detection, an improved RIAD (Region Importance Aware Description) algorithm is provided, and an automatic encoder (AE) is used for deep feature learning of input data. The automatic encoder is composed of an encoder and a decoder, and by minimizing the reconstruction error, key features are extracted from high-dimensional input data and mapped to a low-dimensional latent space.

[0092] Based on the charging station operation and maintenance data after multi-source fusion, an autoencoder structure is constructed, potential feature representations of the data are extracted by the encoder, and high-dimensional reconstruction is performed by the decoder;

[0093] The formula representation of the potential feature z of the data is,

[0094] z = f(x) = sigma(W e x + b e )

[0095] Wherein, W e represents the weight matrix of the encoder, b e represents the encoder bias, and sigma represents the activation function.

[0096] The formula of high-dimensional reconstruction is,

[0097]

[0098] wherein, represents the data after high-dimensional reconstruction, W d represents the weight matrix of the decoder, b d represents the bias of the decoder.

[0099] Through the encoder, the original high-dimensional data is compressed into low-dimensional latent features, which retain the global patterns and key information of the data. The decoder reconstructs the original data based on these latent features, and further improves the accuracy and robustness of the feature representation by minimizing the reconstruction error, which can provide reliable feature support for subsequent inconsistency detection.

[0100] Based on the extracted latent features, the application calculates the reconstruction error and the anomaly score based on the deviation between the input data and the reconstructed data, forming the evaluation index of inconsistency point detection.

[0101] The formula of reconstruction error is,

[0102]

[0103] wherein, e i represents the reconstruction error, x i represents the input data, represents the reconstructed data corresponding to the input data.

[0104] The formula of anomaly score is,

[0105]

[0106] wherein, s i represents the anomaly score of the data, e represents the set of all reconstruction errors; in this embodiment, a threshold θ is set, when s i > θ, the data point is determined as an inconsistency point.

[0107] This example calculates an example, assuming that the input data point is: x = [0.8, 0.6, 0.2] T .

[0108] The encoder weight matrix W e and the bias b e are set as follows:

[0109]

[0110] The obtained latent feature is:

[0111]

[0112] Assume the decoder parameters as follows:

[0113]

[0114] Calculate the reconstruction error and anomaly score:

[0115]

[0116] Normalize the anomaly score (assuming the maximum error is 0.1):

[0117]

[0118] Set the threshold θ = 0.4, because s i > θ, it is determined that the data point is inconsistent.

[0119] Four, input the identified inconsistent data into the repair module, execute the repair mechanism, and complete, correct and reconstruct the abnormal data.

[0120] After the inconsistency detection is completed, the system will mark the inconsistent data, which needs to be further repaired to improve the overall data quality. The present application proposes a two-step repair method based on confidence and information entropy, introduces a generative adversarial network (GAN) for data repair, and through the adversarial optimization of the generator and the discriminator, generates high-quality repair data.

[0121] Complete through the data source confidence, based on the source reliability and historical consistency index of each data item in the multi-source data, construct a confidence weighted repair model, the formula is,

[0122] C(x i →v) = P(v | context(x i ))

[0123]

[0124] Where v represents the repair value, C(x i →v) represents the confidence of repairing x i to v, P(v | context(x i )) represents the conditional probability of v as the repair value under the current context condition, represents the repaired data.

[0125] Correct the abnormal data through the information entropy fluctuation feature, based on the distribution characteristics and uncertainty level of the data attributes, construct an information entropy minimization model, and optimize the probability repair for inconsistent data, the formula is,

[0126]

[0127] where H(x i ) represents the information entropy of the input data x i , V represents a set of possible repair values, and P(v) represents the probability distribution of v.

[0128] This example calculates an example, assuming that the possible repair values in the current context are:

[0129] V = {v1 = 0.2, v2 = 0.3, v3 = 0.4}

[0130] The probability distribution under the context condition is:

[0131] P(v1) = 0.1, P(v2) = 0.7, P(v3) = 0.2

[0132] The confidence of repairing data point x i = 0.2 to v2 = 0.3 is:

[0133] C(x i → v2) = P(v2 | context(x i )) = 0.7

[0134] Using the same repair value set V and probability distribution:

[0135]

[0136] The higher the information entropy, the greater the uncertainty of the data point, and the repair strategy needs to be further optimized by referring to the context.

[0137] An abnormal data learning and reconstruction model is constructed based on a generative adversarial network structure, the reconstruction model is composed of a generator G and a discriminator D, and the relationship of the data is modeled and the semantic reconstruction of the missing area is performed through game optimization;

[0138] The objective function is constructed, which is expressed as,

[0139]

[0140] where x z represents real data, P data (x) represents the distribution of real data, z z represents noise, P Z (z)) represents the distribution of input context features and noise, G(z) represents the output of the generator, i.e. the generated reconstructed value, and E represents the mathematical expectation.

[0141] The maximization loss function L D of the discriminator is the same as the objective function, which is expressed as,

[0142]

[0143] The minimization loss function L of the generator G is represented as,

[0144]

[0145] By continuously optimizing the generator and the discriminator, the generator gradually generates repair values close to the real distribution.

[0146] Embodiment 2, with reference to Figure 2 , which is different from the first embodiment, a method for automatically comparing and discovering power standard deviation information further comprises, in order to verify the practicability of the method, in the case of different detection methods for the same detection data, comparative experiments are carried out on the conditional dependent algorithm (CIND), micro function dependent algorithm (MD) and improved RIAD algorithm. In the detection, the field values that do not meet the CIND, MD and improved RIAD algorithm rules are regarded as inconsistent data, and the precision (Precision), recall (Recall) and F1 score of the detection results of each algorithm are respectively calculated, and the formulas are as follows:

[0147]

[0148] The precision and recall are calculated by using the confusion matrix, as shown in Table 2. In the binary confusion matrix, the column represents the true class, and the row represents the predicted class. The intersection of the matrix is defined as follows: TP (true positive) is the number of samples that are both positive and actual, FP (false positive) is the number of samples that are predicted to be positive but are actually negative, TN (true negative) is the number of samples that are both negative and actual, and FN (false negative) is the number of samples that are predicted to be negative but are actually positive.

[0149] Table 2 Confusion matrix

[0150]

[0151]

[0152] The performance of the improved RIAD algorithm in the multi-table attribute inconsistency detection and repair is evaluated by the above-mentioned indexes. The accuracy is calculated as the proportion of correctly detected or repaired attributes to the total detected or repaired attributes, and the recall is measured as the proportion of correctly detected or repaired attributes to the total number of inconsistent attributes. The comparison results are shown in Table 3.

[0153] Table 3 Recall and precision results of different detection methods

[0154]

[0155] Figure 3 The comparative results of different data inconsistency detection methods are shown. From Figure 3 It can be found that the improved RIAD detection method has the highest recall rate and precision, mainly because CIND mainly verifies the tuple-level consistency across data tables, and its detection dimension is limited to the preset master data reference relationship, making it difficult to effectively identify complex association anomalies between multiple attributes. While the MD method based on local feature constraints can detect explicit rule conflicts between specific attributes, it lacks global data association analysis and has blind spots for cross-table or composite rule anomaly detection. In contrast, the improved RIAD can effectively identify overall and local contradiction information between multiple tables or multiple fields, with more detailed analysis granularity and more extensive rule adaptability. In terms of identification accuracy and recall ability, it has more obvious advantages compared to CIND and MD methods.

[0156] Finally, the repair method of the generative adversarial network (GAN) combined with confidence and information entropy is used to repair the inconsistent points, and the repair results are shown in Table 4.

[0157] Table 4 Repair results

[0158]

[0159]

[0160] As can be seen from the repair results, the generative adversarial network (GAN) repair method based on the fusion of confidence and information entropy performs extremely high recall rate and precision rate when dealing with complex data anomalies. This method first uses confidence to measure the trustworthiness of data in the business context, and combines the distribution uncertainty of information entropy evaluation value to accurately identify high-risk inconsistent data points. On this basis, with the help of the generation ability of GAN, it can generate new values that meet the overall statistical characteristics in the case of missing master data or rule inapplicability, significantly improving the coverage and accuracy of data repair, and making up for the shortcomings of traditional repair methods in flexibility and intelligence.

[0161] Embodiment 3, which is different from the previous two embodiments, is a charging station operation and maintenance data inconsistency detection and repair system based on multi-source data fusion, comprising a data acquisition module, a pre-fusion processing module, a detection module and a repair module; the data acquisition module formulates a hierarchical acquisition strategy according to data sources, granularity characteristics and update frequency for heterogeneous data sources in the operation and maintenance of the charging station, and acquires multi-source raw data including device status, operation and maintenance records and environmental information in time windows; the pre-fusion processing module cleans and standardizes the acquired data, eliminates field naming ambiguity through attribute alignment method, completes cross-source entity alignment through entity recognition, and generates a consistent fusion data set; the detection module constructs an inconsistency detection model based on the fusion data, the inconsistency detection model extracts data implicit structure features based on the feature learning method of the autoencoder, identifies potential abnormal points using reconstruction error, and performs consistency dynamic detection and classification labeling on multi-source data; the repair module completes, corrects and reconstructs the abnormal data.

[0162] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the prior art that essentially contribute or the parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes instructions for making a computer device (which can be a personal computer, a server or a network device, etc.) execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0163] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, device or apparatus, or in conjunction with these instructions. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, device or apparatus, or in conjunction with these instructions.

[0164] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer readable medium can be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, as necessary, and stored in a computer memory.

[0165] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any forms of hardware, or combinations of hardware and software can be used, such as: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so on.

[0166] It should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application but not to limit the present application, and although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion, characterized in that: include, For heterogeneous data sources in charging station operation and maintenance, a layered collection strategy is developed based on data sources, granularity characteristics, and update frequency. Multi-source raw data, including equipment status, operation and maintenance records, and environmental information, is collected according to time windows. Based on the collected multi-source raw data, a pre-fusion processing module is built to clean and standardize the data. Attribute alignment is used to eliminate field naming ambiguity. Entity recognition is used to complete cross-source entity alignment to generate a consistent fusion dataset. An inconsistency detection model is constructed based on the fused data. The inconsistency detection model extracts implicit structural features of the data based on the feature learning method of the autoencoder, uses reconstruction errors to identify potential anomalies, and performs dynamic consistency detection and classification annotation on multi-source data. The identified inconsistent data is input into the repair module, and the repair mechanism is executed to complete, correct and reconstruct the abnormal data.

2. The method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to claim 1, characterized in that: The attribute alignment method is constructed by analyzing the original entities in each data source based on the data sources of charging station equipment, State Grid's Internet of Vehicles, and energy service platforms, and mapping the original entities to aligned entities of unified standards through mapping rules to ensure the consistency and compatibility of multi-source data.

3. The method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to claim 2, characterized in that: The cross-source entity alignment using entity recognition includes designing a similarity calculation mechanism that integrates textual information and semantic information for the multi-source heterogeneous entity names in the charging station operation and maintenance data, and establishing a similarity matching model with entity similarity evaluation as the core to support entity normalization processing; The similarity of the text information is expressed as, Where A and B are word sets of two texts, represents the word set of text A and the word set of text B, |A∩B| represents the number of words that appear together in the word sets of the two texts, and |A∪B| represents the number of all non-repeated words in the word sets of the two texts; The similarity of the semantic information is expressed as, in, and are the word vector representations of text word set A and text word set B respectively, represents the vector dot product, A i Represents the i-th word in the A text word set, B i represents the i-th word in the B text word set, ||A|| and ||B|| represent the modulus of the A text word vector and the modulus of the B text word vector respectively; The formula of entity similarity is expressed as: Overall Similarity=0.4·Jaccard Similarity+0.6·Cosine Similarity Among them, Overall Similarity represents overall similarity, Jaccard Similarity represents text similarity, and Cosine Similarity represents semantic similarity.

4. The method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to claim 3, characterized in that: The feature learning method based on the autoencoder extracts the implicit structural features of the data, including constructing an autoencoder structure based on the multi-source fused charging station operation and maintenance data, extracting the potential feature representation of the data through the encoder, and performing high-dimensional reconstruction through the decoder; The formula for extracting the potential feature z of the data is expressed as, z=f(x)=σ(W e ·x+b e ) Among them, W e represents the weight matrix of the encoder, b e represents the encoder bias, σ represents the activation function; The formula for high-dimensional reconstruction is expressed as, in, Represents the data after high-dimensional reconstruction, W d represents the weight matrix of the decoder, b d Indicates the decoder bias.

5. The method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to claim 4, characterized in that: The dynamic consistency detection includes calculating the reconstruction error and anomaly score based on the deviation between the input data and the reconstructed data to form an evaluation index for inconsistent point detection; The formula of the reconstruction error is expressed as, Among them, e i represents the reconstruction error, x i Represents input data, Represents the reconstructed data corresponding to the input data; The formula of the anomaly score is expressed as, Among them, s i represents the abnormal score of the data, and e represents the set of all reconstruction errors.

6. The method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to claim 5, characterized in that: The completion includes completing the data through the confidence of the data source, and building a confidence-weighted repair model based on the source reliability and historical consistency index of each data item in the multi-source data. The formula is expressed as follows: C(x i →v)=P(v|context(x i )) Among them, v represents the repair value, C(x i →v) means to convert x i Repair to the confidence of v, P(v|context(x i )) represents the conditional probability of v as the repair value under the current context conditions, Indicates the repaired data.

7. The method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to claim 6, characterized in that: The correction includes correcting abnormal data through information entropy fluctuation characteristics, building an information entropy minimization model based on the distribution characteristics of data attributes and their uncertainty levels, and performing probability optimization repair on inconsistent data. The formula is expressed as: Among them, H(x i ) represents the input data x i The information entropy of , V represents the set of possible repair values, and P(v) represents the probability distribution of v; The reconstruction includes building an abnormal data learning and reconstruction model based on a generative adversarial network structure. The reconstruction model consists of a generator G and a discriminator D, and uses game optimization to model the relationship between data and semantically reconstruct missing areas. Constructing the objective function The formula is expressed as, Among them, x z represents the real data, P data (x) represents the distribution of real data, z z represents noise, P Z (z) represents the input context features and the distribution of noise, G(z) represents the output of the generator, that is, the reconstructed value, and E represents the mathematical expectation; The discriminator maximizes the loss function L D The objective function is expressed as follows: The generator minimizes the loss function L G Expressed as, 8. A system for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion, applying the method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion according to any one of claims 1 to 7, characterized in that: It includes data acquisition module, fusion pre-processing module, detection module and repair module; The data acquisition module develops a hierarchical acquisition strategy based on the heterogeneous data sources in the operation and maintenance of charging stations, according to the data source, granularity characteristics and update frequency, and collects multi-source raw data including equipment status, operation and maintenance records and environmental information according to time windows; The pre-fusion processing module cleans and standardizes the collected data, eliminates field naming ambiguity through attribute alignment, and uses entity recognition to complete cross-source entity alignment to generate a consistent fusion data set; The detection module builds an inconsistency detection model based on the fused data. The inconsistency detection model extracts the implicit structural features of the data based on the feature learning method of the autoencoder, uses the reconstruction error to identify potential anomalies, and performs dynamic consistency detection and classification annotation on multi-source data; The repair module completes, corrects and reconstructs abnormal data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a method for detecting and repairing inconsistency in charging station operation and maintenance data based on multi-source data fusion as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a method for detecting and repairing inconsistencies in charging station operation and maintenance data based on multi-source data fusion as described in any one of claims 1 to 7.

Images